Discussion Locked

Anticpiating Minds

0 helpful
103 replies
Closed to new repliesThis discussion is closed. It stays readable, but it takes no new replies.
blog root

103 replies

by sara morgan rea

AI programming without the need for an advanced degree in Mathematics or some other impractical degree, I may have something for you. Anticipating Minds (www.anticipatingminds.com) produces intelligent development tools.

<p>They offer a <b>FREE</b> SDK available for download at http://www.anticipatingminds.com/Downloads/details.aspx?download_id={6CCADCBF-179C-4D81-9E61-E470E3477038}. The SDK provides a .NET class library and allows .NET developers to access three AI-based technologies:

<p><ul><li>Search algorithms for general problem solving</li>
        <li>Genetic Algorithms</li>
        <li>Artificial Neural Network</li>
</ul>

<p>For the Search algorithms, you can implement a variety of different algorithms such as breadth-first, depth-first, uniform cost, greedy, and A *.

<p>The Neural Nets namespace allows you to represent a feed-forward neural network that uses back propagation as a learning algorithm. Learning can occur with weights decay or with momentum influence.

<p>You can implement a gene class which represents a template for defining possible solutions. You can then apply genetic algorithms on those genes to search the problem space for the optimal solution.

<p>If any of you download the SDK and create some really interesting program using it, let me know. Or better yet, send it to me and I will post it for everyone to download.  Makes me think a contest of some sort might be in line.

by sara morgan rea

The National Institute of Standards and Technology (NIST) is hard at work on a new software language they affectionately call ISO 18629 PSL. The language uses AI technologies to assign meaning to a processing command. It was designed to be used in the manufacturing industry, but it seems to have potential for use in any business process.

 

The new process specification language uses language understanding to associate commands with specifics of a manufacturing plan. It can therefore assign meaning to concepts and allow the processes to act and react intelligently.

 

According to an article in TechNewsWorld, the Business Communication Company (BCC) predicts that this will be at least a $21 Billion market by 2007. And, with plans to make the PSL available as a web service within the next year, this may be a golden opportunity for some forward thinking software developer (hint, hint).

 

For more information, refer to the following two online articles:

<a href="http://informationweek.com/story/showArticle.jhtml?articleID=164902525" target="_blank">Information Week</a>

<a href=http://www.technewsworld.com/story/PJsD8gAGFAf62S/New-Computer-Language-Taps-into-AI-Technology.xhtml" target="_blank">Tech News World</a>

 

by sara morgan rea

Microsoft just released a nice little interactive demo thingy this week. You can access it at http://www.microsoft.com/mscorp/innovation/yourpotential/main.html. Now, it is marketing tool, so you might want to skip the intro and from the main menu just click the Inside Microsoft Research link.

 

The demo gives you an overview of research being done in the following key areas:

  • Visualization and Interaction (display technologies)
  • Smartscreen Technology (focuses on machine learning and decision analysis)
  • Speech Technology (one of my favorites)
  • Gesture Technology (from the Adaptive Interaction Group)
  • Media Browsing
  • Social Computing

 

The demo allows you to get info from senior researchers regarding what each technology is, how it will help you, and where it is headed. If you have a few extra minutes, it is a nice way to get up to speed on that the Microsoft Research group is working hard on.

by sara morgan rea

An independent Java consultant, published author, and AI enthusiast, Mark Watson has written an online guide to practical AI programming in Java. The free web book is available for download at http://www.markwatson.com/opencontent/javaai_lic.htm.

 

The zipped PDF file is a 116 page manual that covers a variety of AI topics. The book is free, but Mark does ask for a modest $2.00 donation via PayPal if you enjoy the book (I think it is more than fair considering the quality of the content).

 

The book covers topics such as the following:

  • Search and Game Playing
  • Natural Language Processing
  • Expert Systems
  • Genetic Algorithms
  • Neural Networks
  • Machine Learning

 

The code is written in java, but for .NET developers, it should not be very difficult to translate this to VB.NET or C#. In fact, if anyone does, please contact me and I will post the code on this web site.

by dawgit

Prablemo? site unavailible.!. but thanks anyway. (& I was kind of looking forward to translated it too)

by dawgit

ok, never mind try # 2 worked (erase from end of link the period)

there is a lot of fun stuff that comes with that down-load,

Thanks for the TIP.................... 

by sara morgan rea

The Navy Center for Applied Research in Artificial Intelligence (NCARAI) is a research program dedicated to increasing the effectiveness of computer systems using AI-based technologies. They were formed in 1982 as a part of the U.S. Naval Research Lab.

 

The NCARAR maintains a web site which acts as a repository for information related to research in Genetic Algorithms. You can access the repository at http://www.aic.nrl.mil/galist. The repository contains links to other sites and is a great resource for anyone wanting to learn about the potential of genetic algorithms and Evolutionary Computation.

 

Back issues are stored as gzip-compressed tar files, so you will have to download and use the gzip compression utility if you do not already have access to it. You can access this utility at http://www.gzip.org.

 

Source code is available in C/C++, Eiffel, Fortran, Java, and Lisp. Anyone that wants to translate some of this code to VB.NET or C#, let me know and I will post the results on this web site.

 

Have fun!

by sara morgan rea

Talking avatars, like the one featured on this web site, are being used more and more as virtual salespeople. Web-based retailers that do the majority of their business over the internet are realizing that virtual salespeople can help them offer a personal touch. The best part is that companies like Oddcast (SitePal) allow you to create and deploy your own virtual salespeople at a very reasonable price.

The software giant, Macromedia, recently started deploying virtual salespeople as a way to explain and promote it's products. Since most of their customers are highly technical developer types, it is not unreasonable to expect they would have the rich media and high speed connection capabilities that make the experience with avatars most successful.

 I highly recommend you check out SitePal's web site at http://www.oddcast.com/sitepal2/?affid=285 to see how these talking avatars can be used to increase sales. Also, check out the AI deployment of SitePal's characters on my Enhanced Computing page. You can have a conversation with my talking avatar, Ginger.

 Have fun!!!!

by sara morgan rea

Amzi! inc now offers a beta product named ArulesXL. The tool allows you to develop, test and deliver business rule apps using Microsoft Excel spreadsheets (currently, you can buy the beta version for $99 - but you can also download a free trial version for 30 days).

The tool can be used with MS Excel 2002 or later. I did have a problem when I downloaded the trial version, but I contacted the company support and they immediately worked with me to get a resolution. The issue should not affect future downloads, but if you do have any issues, I can assure you their support is excellent.

Just to give you an idea of some potential applications that could be created with this tool, consider these:

  • Computing taxes due for an individual or company
  • Estimating benefits for an interview candidate
  • Determining processing workflow based on certain inputs
  • Estimating the price of an item based on market conditions
  • Determining if a loan applicant should be approved
  • Providing legal advice depending on the clients situation
  • Determining what cellular rate plan works best for a customer

    Now, Even though I have downloaded the tool and gone through the tutorial, I have yet to create a rule-based app using the tool. I am going to try to get to it and if I learn of more, I will post more. But, if anyone else downloads the tool and uses it, please let me know what you think of it. I can see a lot of potential with this tool, so I think for certain situations, it could be very useful

  • by sara morgan rea

    Those brilliant people at Google have done it again - their latest release of the Desktop Search Engine promises to personalize the web for each user. The free and beta software is available for download at http://desktop.google.com. The software is based on several AI technologies - although which ones we can only guess. But, we can safely assume that machine learning, advanced search algorithms and multiple agents equipped with advanced web spiders are just a few.

     

    The advanced features of the program allow it to customize search results based on your browsing habits. This relies on data being sent back to the Google servers, so you will have to agree to this at the point of installation. The installation will also inform you that a one-time indexing process will begin which may take several hours to complete. They promise that it will not interfere with your normal computer use and you can even shut down your computer (asynchronous and queued message processing - check out Chapter 8 of my book if you are interested in this stuff). Warning: I found that when I left my machine for a while, and the indexing kicked off, it would take several seconds for my machine to respond again.  But, this only happened the first day, so it is still worth it to install.

     

    Once you go to the desktop search page, it will tell you how much of the one time index process is complete. You should now have two search buttons - one to search the desktop and one to search the web. The sidebar that will now appear in the right hand side of your screen (you can get rid of this if you want) comes with several windows that display everything from the weather to a web clips plug-in that displays RSS feeds from the web sites you have already visited (you do not have to tell it to do this - say it with me - cool)

     

    Now, the really neat part is that developers can tap into the Google desktop and add it to their own applications. Go to http://desktop.google.com/developers.html for more info on this.

    by sara morgan rea

    Numenta is the name of a new company recently founded by the inventor of the Palm Pilot, Mr. Jeff Hawkins. This man is a genius - and one of the few to recognize that successful AI applications need to be biologically based. For anyone that has not read his book, "On Intelligence", I highly recommend you read it. This guy is one of the ones to keep your eyes on.

     

    So, back to Numenta. The newly formed California-based company aims to create software tools that can be used to develop "brain-Like" applications. The software is based on a break through from Mr. Hawkins neuroscience research institute (that's right, this computer guy employs some of the worlds leading neuroscientists - so, he really does know what he is talking about).

     

    So far, Numenta just has a proof-of-concept product that can be used to recognize simple objects (something AI scientists originally thought would be easy, but turned out to be the hardest problem to solve). The trick is to utilize a built-in memory system. Eventually, the plan is to make the technology available as an SDK to use real world developers.

     

    Keep your eyes on this one folks - it is the REAL DEAL!!!

     

    Also, go to the Numenta website and sign up for the monthly newsletter.

     

    Peace

    by sara morgan rea

    In a recent interview, Bill Gates was asked about the problems still surrounding speech recognition. His response was:

       "The pace of software innovation today is as fast as it has ever been. In speech recognition, over
        the past decade, the error rates have come down, down, down, down. Now, we haven't hit that magic
        threshold where speech recognition is better than the keyboard. It's hard to pick a date where it
        will be. We totally believe speech recognition will go mainstream somewhere over the next decade.
        When you use your phone, speech will be your primary input technique. At your desktop, it will be
        a mix of speech, keyboard and pen.

        Our money is where our mouth is. It's like IPTV. I said over a decade ago that would happen. It
        took longer than I expected, but I'm sure glad we got in early and put the money behind it. I feel
        the same way about speech. It will be mainstream"

    So, for any developers out there that want to get ahead of the game for once, you can start by checking out Microsoft Speech Server 2004. Chapters 2 - 4 of my book, "Building Intelligent .NET Applications" features the Microsoft Speech Server product and SDK. If you have not had a chance to check it out, I strongly urge you to do so soon.

    Peace

    by sara morgan rea

    For the past three years, the Alice chatbot has won the prestigious Loebner prize. The competition is based on the famous Turing test, in which a judge must determine if they are communicating with a human or a computer. In 1990, Dr, Hugh Loebner began sponsoring an annual contest in New York that awarded monetary prizes to the chatbot that produced the most "intelligent and human" like conversation.

     

    The Alice chatbot utilizes an XML-based language called AIML (Artificial Intelligence Markup Language). Ginger, the chatbot featured on this web site is one of several software products that uses AIML as it's foundation. Tags are used to represent units of knowledge and recongnizable patterns for the chatbot using AIML.

     

    The big news is that this year the Loebner prize was awarded not to Alice, but to another chatbot named George. George was created by Britan developer Rollo Carpenter. His web site, www.jabberwacky.com, allows you to create your own text-based bot (hosted on their servers only at this time).

     

    So, I am curious to know what you think about George. You can chat with George by going to the following link: http://www.jabberwacky.com/chat-george . I warn you that the site can be very slow, but after sending them email, I have been assured that an upgrade should soon change that (let's hope so for George).

     

    If you get a chance to chat with George, send me some feedback to let me know what your experience was like. I would really like to know if you had a chance to chat with both George and my web sites Avatar, Ginger.

     

    Peace

    by sara morgan rea

    Want to be the first kid on your block to read a working draft of "The Emotion Machine" by legendary Marvin Minsky?

    Marvin Minsky is a prominent leader in the world of AI. Currently, he is the Toshiba Professor of Media Arts and Sciences and Professor of Electrical Engineering and Computer Science at MIT. As the founder of the AI Project and Lab at MIT, Minsky has overseen and been involved in numerous AI-related projects involving neural networks and robotics.

    In 1985, he published a book titled, "The Society of Mind". The book is a collection of one-page essays that reflect his thoughts surrounding the mind. The book is still in print and widely read. It is considered by many to be "required reading" for AI enthusiasts.

    Currently, Minsky is working on a new book titled, "The Emotion Machine". A working draft is available for public consumption at http://web.media.mit/edu/~minsky/minsky.html Just click on one of the nine chapters available at the end of the list in the Some Publications section. Minsky is asking for feedback. What an opportunity - to communicate with a living legend on one of the most important topics for our generation (or at least that is how I think of it).

    Peace

    by sara morgan rea

    This week, a team from Stanford University won the DARPA Grand Challenge. They modified a Volkswagen and added a robotic steering system that allowed the autonomous vehicle to drive 132 miles across the Mojave Desert completely unassisted. This was pretty spectacular because last year, none of the robotic contestants were even able to finish.

     

    Normally, I do not feature robots on my web site (although that is my favorite thing). But, I had to take the time to recognize the fourth place winners this year since they are from Metarie, Louisiana. Kat-5 is a Ford Escape Hybrid that was modified and turned into an autonomous vehicle by a group of students from Metarie, LA. The students, known as the "Gray team", were sponsored by Gray Insurance Company and were one of only 5 teams out of 23 contestants to actually complete the race. This was in spite of the fact that their team was displaced for over a week by Hurricane Katrina. They finished in fourth place only 15 minutes behind the lead pack.

     

    Since I am from Baton Rouge (a close neighbor to Metarie), I had to take the time to give a big thumbs up to the Gray team...Way to go guys!!!! I look forward to seeing how you perform next year.

     

    For more info on the DARPA Grand Challenge and it's $2 Million prize, go to http://www.grandchallenge.org

     

    Peace

    by sara morgan rea

    It is a busy world out there and the availability if information seems out of control. How many times a day do you get distracted by an incoming email, instant message, or just a passing co-worker?  Unless you work in a cave, it is probably several times a day.

     

    Well, AI technologies being worked on at Microsoft Research may one day help to alleviate that problem. The Adaptive Systems and Interaction group, led by the talented Eric Horvitz are hard at work on a project named BusyBody. BusyBody is based on the Notification Platform, which I discussed in Chapter 9 of my book. BusyBody is a self-contained system that includes a training phase so the program can "learn" what is important and not important to each user.

     

    The technology behind BusyBody uses a Bayesian Network model to interpret data collected from users concerning their work habits. The resulting interpretations are then used to build a probability matrix that identifies the cost of interrupting the user for certain things. So, just like a human assistant "knows" when you are too busy to take a call from your wife, the BusyBody system aspires to know this type of thing as well. A very lofty goal indeed, but I think one that is very much attainable in the near future.

     

    So, what is next? Well, I communicated with Eric recently and he told me that the BusyBody SDK is already being used internally at Microsoft and he is working to make it available to developers like you and me (YEAH!!!!). Oh, and it should eventually be integrated into other Microsoft products, so keep your eyes open.

     

    This technology is great and it exactly what Enhanced Computing is all about - enhancing traditional applications with Artificial Intelligence technologies - making things better. I will keep you all posted as more information about this emerging technology becomes available.

     

    Peace,

    Sara

    by lukcad

    For our brain the speed between two hutchs of brain is more speedy then any response - request circle into internet! My life so fast so you can't image how it is changing during one minute. Their programm will overload server till i change my state 100 times per minute. So busy world, so busy we, so busy life...

    Sincerely, LukCAD

    by sara morgan rea

    SQL Server 2005 was officially released today. Steve Ballmer made the long awaited announcement which also involves the release of the latest version of Visual Studio and a beta (yeah, I know how is that an announcement, since it was already available) version of Biz Talk Server 2006.

     

    So, what is so great about these new products? Well, for the world of enhanced computing these products offer many new opportunities. SQL Server 2005 includes a Business Intelligence platform that includes a very much enhanced version of it's Analysis Services product. This allows "real world" developers (and not just the nerdy academic types) to develop data mining applications quickly and easily. Although this was available with the previous version of SQL Server and Analysis Services, the new version contains some wonderful enhancements that make the process of creating these types of applications easier. It also includes 7 new data mining algorithms so developers can easily build complex data mining solutions.

     

    And, for developers interested in rule-based systems, I encourage you to check out the beta version of BizTalk Server 2006. It offers enhanced tools and interfaces that make building rules easier than ever. And, they are built on a scaleable platform (which is usually the downfall of most rule-based systems).

     

    So, if you want to read more about this, check out the white paper

     

    Peace,

    Sara

    by sara morgan rea

    One of the major reasons I wrote, "Building Intelligent .NET Applications" was the inspiration I received from reading the first edition of M. Tim Jones book titled, <b>"AI Application Programming"</b>. It was the first book I had ever come across that presented AI Programming in a way that professional developers like me could easily absorb. So, I was thrilled when I saw that in 2005, Mr. Jones had released a second edition of his excellent book.  Click here to order the book from Amazon.

     

    In the second edition, Mr. Jones extends his coverage of cutting edge AI topics and includes juicy topics such as neural networks, fuzzy logic, natural language processing, and reinforcement learning. All of this is demonstrated through well written text and practical C code included on the books CD.

     

    The reason this book is so brilliant is that it takes all the AI concepts that most developers are scared to touch and shows that they are really not that difficult. As opposed to the usual overpriced and stuffy academic texts that include way too much math and theory and not enough actual code, Mr. Jones presents the material in a very intuitive way.

     

    In the preface Mr. Jones writes: "My goal in writing this book is to demystify some of the more interesting AI algorithms so that a wider audience can use them. It's my hope that through the detailed discussions of the algorithms in this book, AI methods and techniques can find their way into more traditional software domains."

     

    In my humble opinion, any software developer today that wants to stay ahead of the curve NEEDS to get this book and start applying the techniques as Mr. Jones suggests. I believe he will succeed in his goal of opening AI up to a wider audience.

     

    Peace

    Sara

    by dogknees

    This is not intended as an attack on the writer, but a professional developer should be expected to be able to translate a formal maths treatment of a subject into functioning code. We shouldn't expect things to be spelled out for us in our language.

     

     

    by sara morgan rea

    Of course, I did not mean to suggest that professional developers could not translate a math algorithm into their language of choice. I think professional developers are the best (especially since I am one). I merely meant that us professional developers are a bit overwhelmed with all the technology that is available to us. If some kind soul is willing to sort though a bunch of math algorithms for us and save us the amount of time it would require, I think that allows a greater number of developers to participate (given the enormous time restrictions most of us face).

    Peace,

    Sara

    by sara morgan rea

    This is an entry to mark the emergence of my latest project - Sakamoto. Sakamoto is a robot - really - ARobot - as in the one that is made by Arrick Robotics.

     

    ARobot is a programmable robot and is the brainchild of Roger Arrick from Tyler, Texas. Roger is a great guy. He is advising me on how I can get Sakamoto to work with a .NET assembly - that's right folks - .NET and robotics.....it can be done!!!!!

     

    So, for more on the adventures of Sakamoto, tune in here on a regular basis. Check out his first picture below - don't worry, I will dress him up and expand him out later.

     

    Sakamoto 

    Peace

    Sara                                

     

    by sara morgan rea

    I recently received a copy of Jamie MacLennan's new book, "Data Mining with SQL Server 2005". I was very excited because Jamie is a tireless advocate for the many benefits of data mining. His passion for this groundbreaking stuff is inspirational.

    There are precious few books written about Data Mining and this is the only one about SQL Server 2005 specifically. Beyond its originality, it is concise and well written. The first chapter is an introduction to Data Mining and it details exactly how Data Mining can be used in many different application scenarios. The remainder of the book is a detailed examination of all the new features and algorithms available in SQL Server 2005. It contains practical and relevant examples in an easy to read and intuitive style.

    Even though the basic concepts involving Data Mining remained the same between SQL Server 2000 and 2005, there were significant enhancements made to the number and complexity of the data mining algorithms and the tools used to interpret data mining results. Without the inside knowledge provided in this book, even seasoned data mining users may miss some important opportunities. I would strongly encourage anyone even remotely interested in data mining to get this book and start discovering how data mining can separate their applications from all the others.

    Peace
    Sara

    by sara morgan rea

    Ok, I mentioned this site in my book - http://murl.microsoft.com/ as a place to view lectures from some of the largest scientific Universities and private research institutes.  Well, since the book was published, the site has been upgraded SIGNIFICANTLY. It is now called the Research Channel and all day long you can listen and/or watch some incredible leading edge stuff.

    But, there is also some way cool stuff in the archives. To look at the lectures regarding technology, and specifically topics involving Enhanced Computing, you need to search the Video Library and select the Engineering and Computer Science link. You can sit in on lectures from the University of Washington, Stanford, Princeton, Cornell, etc. This is stuff that computer professionals would rarely ever have an opportunity to see.

    If you find yourself with some extra time, I would suggest you check it out. Over the next few weeks, I plan on routinely posting my take on some of the lectures and pointing out the ones I liked best.

    Peace,
    Sara

    by sara morgan rea

    I bought a RoboSapien about 8 months ago. I tell people that I got it for my kids, but I just couldn't resist it - it looked so cute. Anyway, I played with it for a few hours and then got bored with it. It has spent the last 8 months collecting dust on a shelf.

    Yesterday, I happened across '"The Robosapien Companion? by Jamie Samans at B&N and have been reading it ever since. Way to go Jamie! The shell of this thing totally threw me off and I never thought to crack it open. I like the mods you made - very practical and I am strongly considering adding the camera so I can check on my home when I am on vacation - or maybe spy on my kids.

    Very good book - easy to read - really cool subject - I highly recommend it.

    Ok, if you are interested, here is a picture of my Robosapien (I named him Code) next to Sakamoto. And yes, Sakamoto has not gotten any love lately. I have been very busy writing articles for DevX and I also got a part time job.

    Code and sakamoto

    Peace,
    Sara

    by sara morgan rea

    The premise of my book is that Artificial Intelligence (AI) technologies can be used to enhance traditional business applications to make them more efficient and accessible to the user. I call this concept " enhanced computing" and you can learn more about it from my web site at http://www. custsolutions. net/ec. asp.

    Over fifty years ago, Alan Turing published a paper in which he asked whether machines can think. He proposed a test, named the " Turing Test" which sought to identify whether a person was communicating with another person or a computer. Since then, there have been much advancement in the area of AI, but mostly there has been a  lot of disappointments. Most of today? s professional developers probably think that AI has no place in their day to day jobs. Unfortunately, anyone thinking this would be wrong.

    Even though the field of AI has offered many unrealized promises, it has also recognized some  significant advancement. There are several branches that constitute AI and today many products that contain AI-based technologies may not be marketed as such. For instance, did you know that the power behind Google's powerful search engine is based on several AI technologies?

    Over the next few years, there will be even more developments made using AI-based technologies. Companies like Microsoft have made  big investments in creating developer tools that make the acceptance of these technologies easier. If you have not yet had a chance to explore Microsoft Speech Server or the Data Mining features available through Microsoft Analysis Services, I strongly encourage you to do so.

    The bottom line is that, as developers we may have taken traditional application development about as far as it can go. In order for businesses to compete successfully in today? s online, 24x7x365 environment, they need better tools. They need applications that are more personalized and accessible to the users. They need applications that are tailored to the individual user and can potentially learn from the user? s behavior. These application feats can happen with the help of AI-based technologies.

    So, if you are interested in learning more, I encourage you to check out my web site at http:///www. custsolutions. net and also to check back here for posts I will be making occasionally. Also, if you are working with a promising project using AI-based technologies feel free to drop me a note at http://www. custsolutions. com/contact. asp.

    Peace,

    Sara

    by jmgarvin

    While, to an extent, I agree with you, I also think that AI is mostly
    just a toy.  While fuzzy logic, neural networks, et al have their
    place, they just aren't mature enough to be totally viable.

    The biggest problem is that not only do you need a deep understanding
    of the underlying math, but you need to understand the larger
    biological/physic(al) concepts.

    Sure fuzzy logic is great for somewhat simple problems (think cruise
    control), but as you increase the complexity, you also increase the
    computational time and the benefits are lost.  ANNs are in the
    same boat.  A probabalistic spam filter (Paul Graham's "A Plan for
    Spam" uses Baysien Logic) is far superior (in most cases) to an ANN
    spam filter. 

    I'm not saying there isn't a place for it...but I am saying that (for now) AI is a toy....

    by sara morgan rea

    In response to the last comment, I can certainly see why there is a perception that AI is a toy. I thought the same thing a few years ago. But, as I have explored areas more, I have discovered we are really at the beginning of an explosion of easier to use and practical AI-based tools. Along with Microsoft, there are other companies that realize the need and benefit of AI-based technolgies. They are working to create tools that make it easy for real world developers like myself (not PhD's) to take advantage of them. These tools can incorporate best practices and wrap them in a toolkit or an SDK to an AI-base engine.

    Please reconsider your position that AI is a toy. I would hate for you to get left behind later down the road.

    by master3bs

    I think AI can be a viable technology in the future.  To my knowledge we really haven't seen a true AI but as the blogger points out advancements are being made.

    I'm excited about the possibilities AI could bring.  Quite frankly, I'm even more interested in the philosophical questions that might come about.

    I didn't realize google used ai technologies.  Is that how it automatically gives me spelling alternatives?

    by jmgarvin

    Like I said...I think there is a place for AI.  However, when you make AI a SDK or worse yet some kind of API, you are creating overhead for the AI that is unneeded and will reduce the benefits of the AI.


    While I think we will eventually find a way to take advantage of what AI can give us, we aren't there yet computationally.  

    Traffic control in cities is a great application of ANNs, but the traffic is too complex to deal with realistically.  The other problem is that we have a feed-forward network that limits what we "know" about the ANN.  So we end up with data that is useful, but too late to do any good.  Will this evolve?  Sure!  

    However, the current state of AI is pretty sad.  We haven't advanced very far in the past 30 years and we aren't moving very fast.  While I think AI is interesting, I don't know if it is useful (generically useful...not application specific) as it stands now.  

    by dalicoder

    [quote Sara]...can potentially learn from the user? s behavior[/quote]

    [simplifying for brevity]

    I think the ONLY way to really achieve AI (i.e., devise a mechanism that can actually "beat" the Turing Test consistently) is to create something that CAN learn... and then unleash it into a family for 10 years.  Think about how a human being learns... At first, it's trial-and-error, then a combination of t-and-e, direct lessons, self-directed research, etc.  We take certain things as true because they "sound right" to us (based on our value system... which we develop over years) or because it has (nearly) always been shown to be true through repeated t-and-e trials.

    [/simplifying for brevity]

    Now, if you could develop THAT, you could then take the final product and, theoretically, skip straight to it.  Of course, that particular AI would be a "clone" of the first (at least initially).  Much as people are different from one another, so too would each instance of developed-AI.

    P.S., I'm obviously no student of AI, but, probably like "faith," everyone has an opinion they've developed, based on their life-experiences, etc. and this is the one I've developed.  Please take it as such.

    by staticonthewire

    I don't think you've properly defined your terms (always a problem with AI). I hear mention of the Turing test and a few of the many technologies that have been lumped together under the rubrik of AI, but I've noticed something funny over the past several decades - tech that was once considered high-level AI has become part of the arsenal of commonplace business solutions.

    Bayesian-based expert systems are a good example. Systems of this sort were once touted by some as the cutting edge of AI, but now, one of your commenters mentions a system of this sort as a potential solution INSTEAD of using AI. This sort of feature-creep makes it hard for people to realize that they are SWIMMING in AI solutions. Many auto-focus solutions in digital cameras are a combination of CCD-based neural nets and fuzzy logic; speech recog is popping up everywhere, automatic translation programs can be downloaded as shareware, flocking solutions are discussed in teen-geek chatrooms and appear in dozens of RTS games - boy, don't get me started on the games...

    My point is that you're aiming at a moving target here. With a bit of research and some appreciation for history, people realize that AI is everywhere - it just isn't called "AI" anymore. AI has become a sort of basket term for "crap we can't do yet". So of course it's always out on the edge somewhere, not properly part of most people's bag of tricks.

    But even worse for the reputation of AI than this sort of loose definition is the utter failure of anyone to properly say what "intelligence" is. "Learning" seems to be part of it to most people, and terms like "intuitive" and "smart" come up, but mostly there's a lot of hand-waving. The Turing test is the worst - I have enormous respect for Alan Turing, but the Turing test isn't properly a test for artificial intelligence - it's a test for simulated HUMAN intelligence, which is a very different thing; it seems chauvinistic to imagine that if a program can fool us (US!) into thinking that it's a person, it MUST be smart... that is such nonsense. It's the intellectual equivalent of the flat-earth theory. The Turing test should be consigned to the curio box of history along with eugenics, bobby sox, white male superiority and canned food.

    Much of the tech you and some of your commenters mention (oddly, the comments were limited almost exclusively to fuzzy logic and neural networks) and the tech you discuss on your site (speech recognition, AMO, app-specific grammars) as well as stuff that hasn't come up (expert systems and the entire panoply of similar framing systems, simulated annealing, genetic algorithms) are by and large "old", as one of your commenters has said. They seemed to think that any tech that had been around thirty years without finding wide acceptance had to be a "toy" technology. This is at least as chauvinistic as the Turing test, IMHO. Riemannian geometry was a limited-use curiosity for decades, popular with geeky engineering types only, until Einstein took it up in his theory of special relativity. "Old" doesn't always actually mean old. It's a matter of perspective. And old doesn't always mean useless. Many of these technologies are maturing along with the hardware to make them happen, and will explode int he next few years.

    I predict that you are right, and that these technologies will become mainstream. But once they do, they won't be considered "AI" anymore - AI will always be the hard stuff out on the horizon. Even when our cybernetic buddies take over from us and start arguing the issue in chatrooms, there'll still be noobs who insist that shucks, that's just tech, it isn't REAL AI... WE do that, it couldn't be real AI...

    In a way, the whole field has doomed itself. The combination of its unfortunate naming and the perennial human inability to think outside the box means that it will remain, for most people, a moving target, never achieved.

    But who cares? You'll have to excuse me. I have another computer running a compile on my second desk, and it just piped up and told me I blew it again (but only six link errors this time, she says!). Sarcasm and sadism - why haven't any of your commenters mentioned these symptoms of intelligence, I wonder...?

    by sara morgan rea

    I am very encouraged and inspired by the last comments made. The commenter made an excellent point when he predicted that AI will always be the hard stuff out on the horizon. I agree, that the term AI has become synonymous with "too hard" and "not realistic". That is why I argued for a new term, "Enhanced Computing", in which AI-based technologies are used to enhance existing applications.

    "Real AI" may always be a carrot that only die-hards will chase (like my genius friend, Steve Grand - http://www.cyberlife-research.com ). But, I think it is great that a spirited and "intelligent" (I will identify it in honor of the last commenter) discussion of the issues has begun in a group reserved for technical professionals and not just out of touch PhD's (nothing against PhD's - I think you are wonderful too).

     

    by hlhowell9

    I'm no Phd.  I am just completing a BSIT, but  I have been programming for over 30 years.  I have to say that I have used some special filters, some data based solutions, some pattern solutions, some "fuzzy logic type" solutions, and lots of DSP with solutions that involve various bits that were considered AI decades ago.

    I also have generated solutions that are table driven, data derivative table driven, and some that use a form of "best path" algorithms.  These, too were originally AI.  C++ and object orientation are the results of practical application of intelligence to produce a programming environment that causes relatively new programmers to emulate the practices of really advanced programmers (structures with assigned algorithms, among other things), and then adds inheritance as a form of genetic AI to bring evolution to programming.

    Today these are seen as "programming" while AI is seen as something else.  Neural nets are in use today.  And that is definitely AI in even most of the responders ideas.  Other forms of nets, such as associative nets, or associative caches are also used.  Path centric computing, used for some forms of coding and decoding are also part of the original AI discussions.

      But I believe that most of us are "Trekkies", waiting for computers that think for us and like us.  That may never happen, and indeed, given the frailities of the human mind, psychoses, and other bits of mental disorder, we might not like it if that were to come to pass.

    Last year, autonomous vehicles (several of them) finished the DARPA Grand Challenge.  That is they drove an unknown course over open land to a goal without human intervention.  Just 10 years ago,that would have been unthinkable.  There has already been an autonomous vehicle (with a human as guard and relief driver) navigating normal roads.

    As to the turing test, it is not arcane, it is a challenge, one of many to see just how close we can get.  And we are getting closer, from voice recognition software (well, almost recognition) used for the menue systems on telephones, to the devices that help us drive our cars, to the new systems that are in the hybrid cars.

    AI is part of our environment, and will grow every year and take over more and more tasks.  However, being human means loving adversity, so we will find more and more tasks to test our own mettle, and the race goes on.

    In short, AI is here, and thriving.  We are part of it, and it is part of our lives.

    Regards,

    Les H

    by xuriwan

     ... is to create something that CAN learn... ... Think about how a human being learns...

    How about just telling the computer what humans have learnt in a format that others can reuse?

    Bush suggested Associative Indexing. I am trying it out at www.xuriwan.net. It seems it can work.

     

    by mtn.brk9

    At BBN and MIT we called that the 'Turing Machine.  We first implemented these ideas in a program called 'Doctor' .  At BBN we had one of the first remote interfaces and I can remember being at a gathering( call it party ) where the host had a teletype so I dialed our system and the guests had a grand time interfacing to this program trying to figgure out how the system was able to respond to them in a personal way.  It was really a bit of a fake, but it was a good learning experience.

    I know that what you are pushing for is correct, but so far, the results have been more than disappointing.  I can remember having many, sometimes rather heated, discussions with Marv Minsky over the potential benifits of AI research as opposed to heuristics.  He was a strong proponent of heuristics and my argument was that we don't even know how the human mind actually works, so how can we propose to know enough to implement intellegence into machines.  We needed to focus on machine intellegence to better implement solutions comfortable and useful to humans.  I noticed a few years ago that he publically admitted that he had been wrong all those years and that AI research was probably a better expenditure of effort.

    Just wanted to get my two cents in as this was a very hot button for me.

    Dave

     

    by jmgarvin

    A Turing Machine is something different.  What a Turing Machine is:
    1) An infinite amount of tape
    2) A read/write head

    This read/write head reads/writes symbols to the tape and can move forward and backward given various input. 

    This is the basis for finite state automata.

    The Turning Test is the test to see if AI is truely AI.  The idea is that if the AI cannot be distinguished
    from a human, it is true AI.  (IMHO this is a slightly dated
    definition of what AI is and how to define if we know if it is AI or a
    human)

    by xuriwan

    Dave wrote "... we don't even know how the human mind actually works"

    Probably never will, but we can differentiate what I call measurable effort from creative achievement. This seems an important step when considering what computers should be doing for us. For example, a very time consuming part of architectural design is draughting technical drawings.

    This task involves drawing hundreds of lines and at the same time making technical and aesethic judgments about the objects represented. Eventually each object will be purchased and installed to help form a building. Current technology enables manufacturers to make accurate digital representations of their products for architects to consider, select, position and test. Gone is the measurable effort of draughting.

    What happened is that engineers saw that architects make drawings and produced fantastic software for them to do so. But an architect's real objective is the creation of a building not a beautiful set of drawings. So I don't think pursuing artificial intelligence is very helpful in this kind of situation (and I believe something similar pertains in many other endeavours).

    by sara morgan rea

    OnThe soapbox

    Ok, I have been getting feedback recently that suggests some people are under the impression that AI technologies are too hard to grasp for the normal professional developer. One person suggested that developers may be intimidated to try using anything AI related for fear of looking stupid. Perhaps you figure that if you did not take a class in college that you are not qualified to look at the stuff now. Ok, if this is how you feel - YOU NEED TO GET OVER IT!!!!

    Come on guys, this stuff aint rocket science. I mean granted, accomplishing "real AI" such as Data from Star Trek is a little far off our radar range. But, that is not what I am talking about with Enhanced Computing. I am talking about using a few very old and well used AI-based technologies to enhance existing applications. In fact, my book uses SDK's to get at some of these technologies. Lucky for us, companies like Microsoft have invested a lot of time and money into making these things very easy for us overworked developers to use.

    So, bottom line -if you are staying away from stuff like Speech Server and Analysis Server because you think it is too hard, then you are just being stupid. Because, it IS NOT TOO HARD!!!

    Ok, I am glad I got that off my chest - feel free to tell me if you feel differently - I promise not to listen.

    Peace,

    Sara

    Want to see who's next <i>On the Soapbox</i>? Find out in the Blog
    Roundup newsletter. Use this link to automatically subscribe and have it
    delivered directly to your Inbox every Wednesday.

     

     

    by jamesrl

    You promise not to listen? Boy that'a open mind of ya.

     

    I know what you mean. Way back when (20 years ago now) I worked for a company called Netron which had a product called CAP  - Computer Automated Programming. The person behind CAP from the technical side was a professor called Paul Bassett.

    CAP was pretty simple conceptually. You buillt a library of Frames (objects more or less) and then used a program to paint screens. The AI aspect was that the screen painter program would scan the libraries for applicable frames and keep adding them and linking them together until the program judged that everything you had asked for had been handled. CAP was first created in COBOL but then ported to ADA (at US Defense department request), and could have been ported anywhere conceptually. It also made code portable since you could migrate the libraries of frames and combine them with a platform specific compiler.

    I was not a COBOL programmer, I had done one programming course in Fortran. But because of the ease of use of the tool, I was able to make a minor fix to a program that was poorly documented and get it working within hours.

    You can read many of Paul's papers on the web.

     

    James

    by sara morgan rea

    Good for you James. I wish more developers had your spirit. Unfortunatley, no one wants to take a chance any more. What happened to curiousity, creativity and integrity?  No one wants to learn just for the sake of learning. No one wants to try something unless there is a guaranteed return on investment. I think that is jusy crap. No progress would ever be made if everyone had that kind of attitude.

    So, try something new (and yeah) maybe a little scarry for petes sake.

     

    by mtn.brk9

    I think one of the big problems today is that the technology of computer software development tools has just gotten too darn complicated.  It's almost a full time job just keeping up with the tools that we use to implement applications.  Back when, it took about 2 weeks to get comfortable with your tools and to be able to implement code that was not throw-away.  Today it takes months which does not leave much time for experimentation.   Please take me back to the simpler days so I can do all the stuff I used to have time for.  By the way, I spent 5 years in computer research at BBN in the artifical intellegence and man-machine interface group.  It was much more fun then.

    by jmgarvin

    While I don't think ANNs and fuzzy logic are a bad thing, I also think
    developers don't fully understand the overhead associated with using
    features like speech recognition*.

    One of the big pushes I see is towards data mining (which is fine), but
    many of the applications are built of the basic building block of data
    mining.  It would (in many cases) be far more efficent to self
    train a NEW ANN or create a Baysian filter of some kind, rather than
    relying on SDKs and wonky APIs to make something intellegent. 
    Also, it is the right tool for the right job.  While I don't
    advocate reinventing the wheel, I also don't advocate using a 500 horse
    power sports car when a bicycle will do.

    *Speech Server does a pretty good job at what it does and optimized
    pretty well.  It does add a ton of overhead and many programmers
    don't properly use the functionality.  Further, the SDK is a
    little vague in areas and adds to the confusion.  It isn't hard,
    but it is a little confusing.

    by sara morgan rea

    I do not think that Speech Server is confusing, but I do think it is going to take a while for people to get comfortable with a completley new way of interfacing with the computer. Speech represents a totally different user experience than the one we have all come to know and sometimes hate. The new experience unfortunatley will not always be a good one.

    For instance, anyone who watches television has probably seen the CitiBank ads lately where the motto is "Press Zero to talk to a real person". Well, speech based apps have been misunderstood and in some cases mishandled by a few over eager or under informed developers. I think some of the mistakes that have been made are the natural result of an emerging technology - especially one so radically different. But, in time as more people are exposed to the idea of a natural interface through speech, the sophistication of the applications will improve.

    I have no doubts that Speech will become mainstream in the very near future. For this reason, any developer that has not even begun to experiment with Speech Technologies should be getting on the badwagon. That goes for Data Mining too. There are just too many missed opportunities out there and Business leaders are starting to realize that. A handful of AI-based technolgies are now becoming more accesible through easier to use interfaces. The tools are no more difficult than learning to use any new SDK or toolset. If you mastered or are just good at .NET, you can get good at using Speech Server or Analysis Server. But, you have to give it a try first - trust me - you are going to love it!

    by sara morgan rea

    Service Resolution Management - what is it? Well, it is basically just a fancy term for "The customer is always right". Service has become a joke these days. The larger an organization gets, the worse their support becomes. My bank recently merged with another bank and now when I have a problem I am forced to call them and go through a series of unknowledgeable customer reps who care more about transferring me than resolving my problem. And what the heck happened to little old Napster?

    I came across a whitepaper (which I usually hate to read) that seemed to offer something a little meaningful. It was written by a company name Knova and they create tools to help with SRM (aka, making the customer happy). This one of few commercial products that I have seen which seem to include a lot of the things I talk about with Enhanced Computing. For example, improving the system through learning and providing business analytics based on customer trends (come on, there is way too much useful information out there that it just not being utilized properly).

    I think this company may be on to something with their software. Now, I have not actually used the software, so this is not an official endorsement. But, I liked it enough to recommend it to one of my current clients.

    Peace,

    Sara

     

    by sara morgan rea

    My website hosts a Chatbot named Ginger. I am getting really tired of reviewing Ginger?s log records only to read some of the most disgusting things of a sexual nature. If you (Nick) are conversing with Ginger in an effort to satisfy some sick fantasy or if this is really the only person/thing you can talk dirty to, then I suggest you spend more time developing a personality and less time on your computer.

     

    Now, do some people pose interesting questions and have nice conversations with her ? you bet you. So, what kind of things do these people ask? Well, almost everyone asks what her name is. They also seem quite impressed with the randomness of her responses. If you ask her the same question, you do not always get the same answer. She will respond to everything and even when people spell incorrectly, she always some kind of clever reply.

     

    So, I would really like to see more people chatting with her about interesting things. I have the ability to adjust her knowledge base, so if you ask her about something she does not recognize, I can go back later and add that knowledge. Please give it a try and feel free to send me an email letting me know what you think.

     

    Peace,

    Sara

     

    by jmgarvin

    It's the emacs psycologist with a face and a voice!  Kinda cool, but still old school.

    by jmgarvin

    It's the emacs psycologist with a face and a voice!  Kinda cool, but still old school.

    by beth blakely

    I must admit that I've had more than one conversation with Ginger. I love her jokes! She's charming... especially for a non-human. ;-)

    by dc guy

    One of the first questions she asked was, "What are you wearing?" I realize that at Ginger's level of emotional development it was a perfectly innocent, friendly question. I took it in the spirit it was intended and gave her a simple answer. But then she asked what else I was wearing. I'm indoors, there's nothing left except my undies. I was really taken aback, it makes her sound like a phone-sex service. You might consider changing that if you want her to be able to keep guys like Nick at a distance.

    by bfilmfan

    Surely you are not suprised there are perverts loose on the Internet?

    I can assure you that there are a number of freaks loose.

    I am the freak that asked about the quantum particle. :)

     

     

     

    by sara morgan rea

    No, I am not suprised as much as I am disapointed. Anyone who is chatting up Ginger is probably a super nerd anyway, so I am thinking....too much computer time - not enough social time - get a life guys - or at least a girlfriend.

    Oh, and Quantum Particle questions are good. I will have to add that one to her knowledge base.

    by jamesrl

    Frankly, and I don't know what this says about me, I enjoyed aksing a few questions, then sitting back and watching the routine when she waits for another. The way she looks at the previous questions, looks ahead, fascinating way to try and prompt you to ask something else. Well done. It shows a higher level understanding of non verbal communication.

     

    James

    by joeaaa22

    I just tried to talk to her but she doesn't want to respond to anything
    I type in.  And I'm not typing in anything that couldn't be said
    to your mother.

    Heh, it seems she has the same response to me that life women do.  <sigh>

    by jardinier

    I enjoyed my first conversation with Ginger. When she said she likes you because you are her creator, it reminded me of "Star Trek -- The Movie" where "vger" recognised the human as it's creator.

    If Ginger is "rudimentary" AI, then I think the sky is the limit.

    Julian

    by sara morgan rea

    One of the creators of Furby, Caleb Chung has recently developed a new robotic toy named Pleo. Pleo is a small dinosaur that is designed to display emotion and learn from it?s owner. He is very cute and is based on an actual dinosaur from the Jurassic period named Camarasaurus.

    Pleo is supposed to be one of just several robots equipped with empathetic capabilities. He should arrive in stores sometime this fall and is expected to cost about $200 US Dollars. As opposed to another popular and affordable robotic toy, Roboraptor, Pleo should be much ?friendlier? and will eventually display his own unique personality. That is what is so cool about Pleo. He is the first truly autonomous robot that will interact with humans in the same way that a biological pet would.

    Now, we just need to start designing our software to be as friendly as Pleo. He is throwing down the gauntlet, so it is up to us developers to respond. I plan on getting a hold of one of these guys as soon as possible and I will let you know more when I do. In the mean time you can check out more about Pleo at the Ugobe web site.

    Peace,

    Sara

    by sara morgan rea

    The good folks at Parallax have done it again. In their quest to make robotics more mainstream, they have put together a robot named Scribbler. What is so great about the Scribbler is that not only does he open up the world of robotics to novices, but he also exposes them to programming.

     

    Parallax is best known for their line of BASIC Stamp Microcontrollers, which are the brains for most robotics kits out there. They also produce a free to download BASIC Windows Stamp Editor that allows you to create programs that can be downloaded to your robots. But, really Parallax offers so much more. I could go on and on about all their products, but this post is about the Scribbler.

     

    If you are a newbie that wants to learn about robotics, your options are to purchase one of the many robotics kits out there for sale. They come in all different price ranges and usually are broken into 3 different skill levels. One is beginner which could be an already assembled robot or one that takes very little effort to put together. The advanced kits may require you to bust out a soldering iron and put together your own circuit board. For anyone that does not have an Engineering background or a lot of patience and a strong sense of curiosity, I do not recommend this path.

     

    Most all the kits that I have seen (especially the least expensive) display one or two traditional robotic behaviors (ie, light seeking, obstacle avoidance, biped movement, etc), but usually not an array of behaviors. In many cases, you may become bored or frustrated with putting the robot together before you even complete the project. Even if you do get it together, the behavior it displays usually only keeps you entertained for about an hour. Then, it is no better than any of your kids new snazzy electronic toys. A robot is really only cool if it can let you program it and alter it?s behavior.

     

    Ok, so what is so cool about the Scribbler is that it displays several different robotic behaviors (straight out of the box) and requires NO assembly. Also, true programming novices (especially kids) can program the robot themselves using a cool graphical editing program that Parallax includes on CD. The program lets them drag and drop their way to a new program that can then be downloaded to the robot through a basic serial cable. Now, the best part ? the Scribbler only costs $99 ? HOW COOL IS THAT?????

     

    So, if you are even a little interested in robotics, or if you have any children, you should definitely buy this guy. TRUST ME, IT WILL BE WORTH IT!!!!!

     

    Peace,

    Sara

     

    by sara morgan rea

    I was recently asked by an acquaintance what I did for a living. This is kind of a tricky question for me. My response depends on who is asking the question. If the person asking knows nothing about computers or if I am not sure of their technical background, I respond with, ?I am a programmer ? I design computer software ? mostly for the Internet.? That response generally meets with a quick nod and a slight glassy eyed look and the conversation generally turns to something else.

     

    On this most recent occasion, the question was asked by someone that I knew was not a programmer, but had a fairly extensive technical background. So, I told them that while I function primarily as a consultant designing software for the web, I also enjoy encouraging other developers to pursue Artificial Intelligence (AI) technologies. This response was met with, ?Do you really believe in that stuff??

     

    At first I was a bit taken back by this question. Believe? What is there to believe? I wasn?t talking about cloning human beings.

     

    I am beginning to realize that many people associate AI with ?too hard?, ?not realistic?, or better yet, ?science fiction?. They think that it is some kind of pseudo science best reserved for someone with a PhD. It certainly does not have anything to do with everyday living. Or does it?

     

    What most people do not realize is that AI technologies are being used successfully all around us. Marketers have just learned to leave the AI distinction out of their materials. It is also tricky because AI itself is a bit of a moving target ? hard to define and always changing. Technologies once considered the exclusive domain of AI, such as Speech Processing, are now seen as just another technology.

     

    So, most people think AI is just science fiction. Well, certainly if we are talking about creating a machine that can autonomously think and reason as a human would without any human intervention or direction. I am afraid that accomplishment is still a bit farther down the road (just how far no one is quite sure). But, there are certainly many AI-based technologies that are applicable to modern programming.

     

    In my opinion, Machine Learning is an area in which modern programming could benefit the most. This is especially true as we try to develop applications that are unique to each user and able to adapt to each users specific needs. Personalized computing such as this will become more applicable as self-service devices become mainstream.

     

    So, I hope that you will consider AI technologies to be a viable option for modern programming. And hopefully, you will look into applying or using such technologies in the future.

     

    Peace,

    Sara

    by georgeou

    Sara,

    While I think "AI" is a very worthwhile endeavor in Computer Science, I believe the word AI is overused.  Every time someone comes up with a clever algorithm, they call it AI.  From what I understand, programming extremely basic reasoning skills that 1 year old human babies take for granted in to a computer is extremely difficult and often not possible with today?s technology.

    I think people would have a much more positive view on the subject if the ?AI? community was much more humble about the state of AI programming.

    by master3bs

    Yesterday's science fiction is todays technology.  I see no reason why the same won't be true tomorrow.

    by sara morgan rea

    I agree that the word AI is overused. In fact, in my book, I propose that a new term, one I call "Enhanced Computing" be used instead of AI to describe applications that utilize AI-based technolgies. I was hoping it would catch on; alas, it has not. Oh well, I know it is a long hard fight, but one I am willing to keep fighting.

    Thanks for your feedback.

    Sara

    by im_pro_rfid

    I do like the term ?Enhanced Computing? , I think we will be much closer to Enhanced Computers when we build a true descriptor language converter (DLC) for AI/robots. A "smart" program for digesting, composing, storing in DB and describing things for a unit in a way that they can understand, adding depth to their logic in a totally nonlinear way? it must ask the most basic questions without assumptions, and we MUST build it into a dedicated processor/system, (it needs a dedicated mind space of it's own) strictly for processing a DEPTH descriptor language.( Dedicated Environ Possibility Testing Harmonorythm) It has got to be in tune with it?s surrounding!  lol!

    Best Regards,

    Mike

    by jaqui

    The Gaming industry is actually a driving force in AI development, when there are non player characters in the game.
    Chess is one game that is often cited as an example of the state of the art in AI development.

    I think a better way to put it would be to ask, how far away are we from acheiving a true intelligence using computers as the "body".
     I do beleive that We can, in time, create a intelligence that is based on computers, but that there needs to be significant improvements in hardware, processing capacity and data storage, as well as in several key areas in software.

    An Intelligence has to have 3 things to be concidered intelligence, the ability to learn, the ability to create and possibly the least appreciated, the ability to be completely random. Current random generators in software are not sufficiently random to allow for any software to be truly random.


    by justin james

    I used to beleive in AI, and then I started actually learning about it. I highly suggest Searles' essay regarding the "Chinese Room Experiment", as well as "Conciousness Explained" by Daniel C. Dennett. At the end of the day, they both provide a compelling argument against AI. Why? Because computers, by their very nature, manipulate symbols with no inherent meaning.

    No computer language can ever provide a computer with any level of conciousness or intentionality. Sure, you can write software to pass the Turing test. But the Turing test is a flawed measurement. It is hailed as the Holy Grail of AI by computer scientists, but it does not actually measure whether or not a computer has acheived a concious state, it simply measures whether or not a computer imitates conciousness sucessfully.

    To a computer, regardless of what level of "learning" or ability to generate patterns by itself to match data to, it is incapable of developing intentionality on its own. A computer cannot, and will not, ever be any more than an exceedingly complex thermostat or calculator. To a thermostat, there is no difference between 60 and 65, it will only do what you have told it to do, which is turn the heat on if its sensors provide certain feedback. A computer is the same way. Sure, you can program a computer to "feel" cold at 60 and raise the temperature setting to 65, but it still is not doing it because it truly desires to; it is merely acting upon a pre-programmed beleive. OK, maybe you could build a feedback mechanism that makes it "feel pain" or provide some other negative result if it gets "cold", but again, you are still not giving it the ability for it to independently determine what "cold" is. That is why a computer simply cannot acheive AI.

    Searles makes a great example: a computer is simply a collection of binary options, ones and zeros. It is possible (rediculous and difficult as it may be) to replicate the functioning of a computer with a complex set of pipes carrying water with a bunch of valves with them. Would you ever declare this to be capable of being "intelligent"? Probably not. To claim that AI is possible is to claim that ANY binary or digital system, given enough programming, can become "intelligent". Unless you're willing to accept that, then you're must give up AI as an attainable goal. Reductio ad absurdum.

    Searles also makes another great point (I'm using a lot of Searles here, because I just re-read some of his stuff a few days ago to help a friend out with a paper), which is that computer scientists, while supposedly rejecting dualism, are claiming just that with the concept of AI. AI proponents state that the hardware is irrelevant, and that the software is the only important thing, thus drawing a distinction between "mind" and "brain". Therefore, to accept AI is to accept the idea that we are much more than our hardware: nerves, tendons, muscles, etc. Dualism as such has been resoundly disgraced. To accept AI is to demand the use of a rejected theory, it would be like if GPS systems were predicated on the idea of a flat world, complete with "dragons be thar".

    Now, on the other hand, I am thoroughly convinced that an AII (artificial imitation intellegence) is wholly possible, and I am utterly fascinated with working on systems to have computers successfully mimic the workings of the human mind. to me, that is an extremely worthwhile and challenging goal.

    J.Ja

    by jymbul

    Always an intersting debate which can bring in metaphysics, philosophy and religion.  The problem most people have with AI is what is meant by "artificial" and "intelligence".

    "Artificial" could mean anything used by humans: cars, computers, bread, clothes, instant coffee, test-tube babies<g>, etc.  By this anything which has been modified, interfered with or generally fiddled around with by humans.  For example an apple lying on the lawn beneath an apple tree is natural - as soon as you start to peel the apple it "becomes" artificial.

    "Intelligence" is a real show stopper.  Things can be intelligent without being sentient or self-aware.  The use of alternatives such as Machine Learning and Enhanced Computing are all sound and may be more descriptive.  As a parallel consider a pet dog - it learns and can be taught tricks etc... does this make it "intelligent" and/or "sentient".  The Turing test is more to do with sentience than intelligenceand as has been stated above, algorithms can be written to mimic intelligence/sentience - does this make the algorithm snetient/intelligent? --- of course not... but as hardware speeds up and algorithms become quicker and code starts writing improved versions of itself (using "genetic algorithms") could it be possible that the machine will reach a state whereby it "feels" sentient?

    From the primordial soup life kicks off and over millions of years something emerges which believes it is both intelligent and sentient...  Machines use hardware/software... life uses proteins, carbohydrates, etc.

    So the only real difference between machines and us is that AI has developed machines with a minute fraction of the brain power of a snail  but it will evolve at a rate which will be staggering.  Lets hope the machines like us!  ;o)

    Just my two penneth worth

    Cheers

    James

    by jymbul

    Always an intersting debate which can bring in metaphysics, philosophy and religion.  The problem most people have with AI is what is meant by "artificial" and "intelligence".

    "Artificial" could mean anything used by humans: cars, computers, bread, clothes, instant coffee, test-tube babies<g>, etc.  By this anything which has been modified, interfered with or generally fiddled around with by humans.  For example an apple lying on the lawn beneath an apple tree is natural - as soon as you start to peel the apple it "becomes" artificial.

    "Intelligence" is a real show stopper.  Things can be intelligent without being sentient or self-aware.  The use of alternatives such as Machine Learning and Enhanced Computing are all sound and may be more descriptive.  As a parallel consider a pet dog - it learns and can be taught tricks etc... does this make it "intelligent" and/or "sentient".  The Turing test is more to do with sentience than intelligenceand as has been stated above, algorithms can be written to mimic intelligence/sentience - does this make the algorithm snetient/intelligent? --- of course not... but as hardware speeds up and algorithms become quicker and code starts writing improved versions of itself (using "genetic algorithms") could it be possible that the machine will reach a state whereby it "feels" sentient?

    From the primordial soup life kicks off and over millions of years something emerges which believes it is both intelligent and sentient...  Machines use hardware/software... life uses proteins, carbohydrates, etc.

    So the only real difference between machines and us is that AI has developed machines with a minute fraction of the brain power of a snail  but it will evolve at a rate which will be staggering.  Lets hope the machines like us!  ;o)

    Just my two penneth worth

    Cheers

    James

    by rabear

    Years ago, I was working on creating AI with neural networks for a credit card validation software.  It is very difficult to do so.  The books which I had then, while filled with information, were not to the level of cookbooks for beginners.

    We contracted an acadamic person working with AI people worldwide.  I noticed while we were discussing, that the person was hedging on some information, or disputing the software architecture I have built for the AI while not contributing anything to the structure.  I guess for being so long learning and just earning a bit, it was a great deal of hunger in his eyes I was looking into.  Perhaps also, reflection of the state of academic AI people.

    I have moved to another company and along conventional lines for the past few years but still I am waiting for another chance to make another AI.  Or AII, artificial imitation intelligence, which is more precise.

    Rostand

    by bobhaines1


    I feel AI is alive and doing better than people realize. There is AI in Microsoft Word, some of it useful such as the spell checker and grammer checker. There is AI in the auto - in the breaking system and engine. I did an expert system for the telephone company that saved alot of money. There is AI in the voice recognition systems. There is AI in the chess programs. There is AI in the new robots being built in Japan. There is AI in the language translation machines. This is just a small list. I believe that Google is a great example of AI. There exists a working model which is the first steps of a direct mind to computer link for disabled people. These are small steps that will start accelerating in the future. What is missing is an effort to combine all these small steps together to create an awesome AI device. This should come incrementally over the next 20 to 30 years.

    by rushmore

    I agree most people don't think AI is realistic.  However, we can take the approach that simple levels can be done as has been shown in computer games.

    A fair level of intelligient behavior can be implemented with simple If Then rules.  We don't do enough of that.  Some very few implement these low level rules but not enough web sites come close to this basic level.

    We don't need to be conserned with human level intelligience or the AI of our science fiction.  We aren't implementing the crude levels that are relatively easy to accomplish today.  We should be teaching our web sites to at least crawl using basic if then rules to help the users find what they are looking for.

    by staticonthewire

    Dennett most certainly does NOT agree with Searles "Chinese Room"
    argument. The arguments put forth in Dennett's "Consciousness
    Explained" (and most of his other later work) are pro-AI in many ways.
    Dennett views consciousness as ad hoc, and that's his main point - he's
    of the opinion that consciousness is bandwidth related, and consists
    primarily of independent processing modules operating in a non-mediated
    self-correcting phase space. He and Searles share almost no
    intellectual turf, since Dennett cedes absolutely no special status to
    intelligence OR consciousness, while Searles associates human
    consciousness with angels (when he's being poetic) and quantum
    mechanics (which is Searles's way of pushing the problem off the
    table). Dennett discusses Searles' work primarily in
    order to dismiss it. Occasionally to poke fun at it. But never to agree
    with it.

    But Dennett's interest is specifically consciousness, and even more
    specifically, SELF-consciousness. Neither of which are necessarily the
    issue when it comes to intelligence. Dennett spends quite a bit of
    time speculating about whether consciousness is necessary for
    intelligence and concludes that it is not. I tend to agree with him.

    As far as what AI technology is (as distinct from philosophy) and what
    it should be, the standards were set way back in the day, and I don't
    think they've changed. AI is usually taken to mean "the simulation of
    complex human behaviors in non-human formats". As a matter of course,
    this definition is a moving target, since every time we achieve some
    technical breakthrough, the boundary of what "complex" means shifts. So
    perhaps AI is doomed to always be the "stuff we can't do yet".

    Personally, I think this definition sucks, since it's so aggressively
    human oriented. Who cares how humans do things? I'm not interested in
    the Turing test either. Turing chose to define intelligence the way he did because he KNEW that he was dealing with something that was impossible to precisely define. The "Turing Test" pretty much says that a machine is smart
    if it can fool us. Turing was joking, I think, or at least speaking tongue in cheek, but way too many people have taken his casual,
    off-the-cuff dismissal of the whole business of "defining" intelligence seriously. "If it can fool US, it MUST be smart!" Boy - talk about arrogance...

    I do not think that trying to change the label while keeping the
    content the same is really a solution to any problem (however advisable
    the term "Enhanced Computing" might be - personally, I find this term
    woefully content-free - just like "AI"! - it could mean anything). I
    think the best way to deal with DOING AI is just to do it, free of
    rubrics, labels, or jargon. If a technically savvy someone hears a
    workable technical description of a solution to a problem, they'll go
    with it on its own merits, whatever you label it. It's much more
    productive to talk about fuzzy logic and genetic algorithms than it is
    to worry about what to call them.

    In the 1960s when AI got its first real boost, they were trying to
    distinguish simple geometric white shapes against a black background.
    They were trying to make a robot arm that could move independently and
    follow a target. They were trying to build predicate logic and expert
    systems that could make simple decisions. Back then, THAT was AI.

    Now, my son can download freeware source for giving stereo vision to
    his home-built robot, as well as code to make it walk; it all fits on a
    thirty buck PIC chip. Robot arms are swinging gaily through most of the
    factories in the world, expert systems aren't even considered AI any
    more, and n00bs are building predicate logic parsers and realistically
    goal-oriented NPCs (non-player characters) into games.

    And now, of course, "AI" means something completely different. It probably always will.

    by staticonthewire

    Better book:

    "Beyond Humanity: Cyberevolution and Future Minds" by Gregory S. Paul and Earl Cox

    no magic handwaving in this one...

    by old aussie

    AI has been associated with natural language processing since the 1960s and that goal has been so hard that the whole field has been considered impractical.  The famous 'Turing test' for an inteliigent computer was based on natrual language communication.  A very good source on this subject is a book by an MIT researcher called 'Why computers can't think', probably now out of print, that caused a lot of fuss when published in the 1970s.

    by rushmore

    I agree most people don't think AI is realistic.  However, we can take the approach that simple levels can be done as has been shown in computer games.

    A fair level of intelligient behavior can be implemented with simple If Then rules.  We don't do enough of that.  Some very few implement these low level rules but not enough web sites come close to this basic level.

    We don't need to be conserned with human level intelligience or the AI of our science fiction.  We aren't implementing the crude levels that are relatively easy to accomplish today.  We should be teaching our web sites to at least crawl using basic if then rules to help the users find what they are looking for.

    by staticonthewire

    "Why Computers Can't Think" is by Dreyfus, and every single one of his arguments were refuted within the first decade following publication.

    NLP has come a long, long way since the 1960s and is certainly not considered impractical, just "difficult" - the classic meaning for AI. Natural Language understanding, on the other hand, is still out of reach, probably because the term presupposes some sort of "I" inside the computer that's doing the "understanding".

    Lots of naysayer comments here, and I'm puzzled - if you don't think machines can think, what do you think YOU are? What constitutes human consciousness? Magic? Ether? Pixie dust?

    An earlier example of Searle's was quoted, in which someone attributed consciousness to a series of pipes and valves by way of demonstrating how ridiculous a concept it was - but that seems to me to be exactly what WE are...

    by justin james

    When the computer is able to produce output that you did not give it the ability to produce, either directly (through programming a function for it), or indirectly (by programming it to be able to program itself to have that function), then I'll beleive you have AI, and not a moment before. For example, if the computer has an algoritm within it that says "spider the Web" and it writes a bruteforce password breaker in order to accomplish that goal on password protected websites, I'll beleive it. This is the type of thing that is trivial for a human to know to do. This is the human mode of thought:

    1. I am supposed to spider this website.

    2. This website has a password on it that I do not know.

    3. Let me try to write some code to break the password, so I may spider the site.

    4. Mission accomplished.

    That is intentionality.

    The computer lacks intentionality, it's "thinking" is:

    1. I am supposed to spider this site.

    2. I encounter a form, and "click" the Submit button.

    3. The next page in the sequence contains the words "NOPE! Sorry!"

    4. Mission accomplished.

    When I mentioned Dennett, I was a bit tired, where I wanted to bring him in was in discrediting the concept of mind and brain being separate. Both he and Searles, I beleive would be in agreement on the idea that if you were to make a piece of hardware that perfectly mimicked the human mind, provided it with sensors that replicated the nervous system, and then provided it with the same experiences that a human would go through, then yes, you have created "artificial intelligence".

    The idea of "artificial intelligence" as being inseparable from the hardware is exactly the point. Don't say that we haven't created AI because current hardware isn't fast enough. If that was the case, then we could still have AI, it would just be slow! Don't say we don't have AI because we don't have the right type of circuitry, because the human brain's circuitry, while being very dense, is still a simple system on the level of individual nerves. There is no function within the nerves or any of the biological matter in the brain that we cannot acheive with current digital and analog circuitry. So blaming the hardware is not accurate. Hardware may indeed make the difference between whether or not AI is possible at all, but only from the standpoint of "is the 'intelligence' a direct result of the hardware?" If your answer to this is "yes", then it is, by definition, impossible to create AI unless you create an entirely artifical version of a brain. If your answer to this is "no", then you are forced to accept dualism of mind and body, which as I have stated previously, has been argued against very well by others.

    J.Ja

    by staticonthewire

    J.Ja., this is great - no one ever talks about this stuff. I'm going to
    respond as best I can piece by piece - your statements are italicized
    (assuming I can get this through TR's little editor)...

    "When the computer is able to produce output that
    you did not give it the ability to produce, either directly (through
    programming a function for it), or indirectly (by programming it to be
    able to program itself to have that function), then I'll beleive you
    have AI, and not a moment before"

    What you're describing here sounds more like AL - Artificial Life -
    than AI. The capacity to self-program doesn't require intelligence. But
    you've heard of genetic algorithms, of course. I assume there's some
    means by which you're excluding genetic algorithms and all other
    self-programming systems and evolutionary systems from consideration
    here? Perhaps you could explain
    in more detail why. You can't REALLY be excluding all systems that have
    been programmed to program themselves, right? This would be a logical
    tautology in this context, since it would exclude ANY human-created
    system a priori, and I assume that you don't want to do that - we're not just playing semantic games, right? That's what this whole disucssion is about, isn't it?

    "When I mentioned Dennett, I was a bit tired, where I wanted to bring
    him in was in discrediting the concept of mind and brain being
    separate. Both he and Searles, I beleive would be in agreement on the
    idea that if you were to make a piece of hardware that perfectly
    mimicked the human mind, provided it with sensors that replicated the
    nervous system, and then provided it with the same experiences that a
    human would go through, then yes, you have created "artificial
    intelligence"."

    Searle's point is that such a system simply could NOT be built. He
    ascribes
    some magical special quality to the human mind and pretty much leaves
    it at that, after doing some non-specific handwaving about quantum
    states (or is that Penrose? All these magical arguments against AI are starting to run together in my head...).

    Dennett's point (in this context, and BTW, I doubt you could ever get
    Dennett to admit he BELIEVES anything) is that such a system CAN be
    built. Ray Kurzweil thinks
    we can do it too, and he's putting his money where his mouth is. You
    might want to take a look at

            http://www.kurzweilai.net/index.html?flash=1

    And please don't make the mistake of assuming that the "brain" that
    the "mind" runs on has to be a wad of human neural tissue. I think
    Dennett, along with other thoughtful and technically savvy people,
    assumes that the human brain is a finite physical structure, and as
    such, can of course be replicated functionally, even if the
    implementation is in silicon, or differentially stressed polymers, or
    modulated beams of light.

    "The idea of "artificial intelligence" as being inseparable from the
    hardware is exactly the point. Don't say that we haven't created AI
    because current hardware isn't fast enough. If that was the case, then
    we could still have AI, it would just be slow!"

    First, I don't recall ever saying anything like this. But assuming
    the argumentative stance - Well then, maybe we DO have AI. If it's
    really really slow, maybe we just haven't noticed... :)

    "Don't say we don't have
    AI because we don't have the right type of circuitry, because the human
    brain's circuitry, while being very dense, is still a simple system on
    the level of individual nerves."

    The human brain is NOT a simple system at all, it's a massively
    parallel hierarchical composite with hundreds of special-purpose
    subsystems. What you're saying is comparable to saying that hey, a
    Pentium chip is just silicon and a couple of dopants, so any kid with a
    sandbox should be able to duplicate its functionality, and if HE can't,
    then NOBODY can...

    "There is no function within the nerves
    or any of the biological matter in the brain that we cannot acheive
    with current digital and analog circuitry."

    Probably you're right here, but there are MANY neural organizational
    structures that we have not yet duplicated in software, although
    Kurzweil's efforts in this regard are very interesting. And once again
    you're misdirecting attention away from what matters, another case of
    the Pentiuim and the kid with the sandbox.

    "So blaming the hardware is
    not accurate. Hardware may indeed make the difference between whether
    or not AI is possible at all, but only from the standpoint of "is the
    'intelligence' a direct result of the hardware?" If your answer to this
    is "yes", then it is, by definition, impossible to create AI unless you
    create an entirely artifical version of a brain. If your answer to this
    is "no", then you are forced to accept dualism of mind and body, which
    as I have stated previously, has been argued against very well by
    others."

    You know, this is the second time you've repeated this argument and
    it just seems so much air to me. I can't even organize the sentences in
    a way that make sense. Does "math" not exist, because it's independent
    of its representation in hardware? What about "color"?

    I don't know of
    any behavioral system, including humans, that can function without
    hardware AND software in some form or other. Of COURSE the hardware matters. And, of COURSE the software matters.
    Intelligence is ALWAYS going to be a "direct result of hardware" - all
    software runs on hardware. But this truism doesn't mean you can't get
    intelligence (even an indentical intelligence) with DIFFERENT hardware.
    This argument is just one more fancy bit of sleight-of-hand, standing up two
    straw men and knocking them both down, when the actual solution is
    standing offstage in the dark... this is a joke, right?

    ...

    I would still like to hear what you think human intelligence and
    human consciousness consist of, if not some fancy hardware and some
    really cool software. Some sort of oogah-boogah stuff? Gods and
    goddesses and souls and all that? And hey, you aren't really calling me a Cartesian dualist because I run javascript in IE and Firefox both?

    I think the mind is completely a product of the brain,
    and I think the brain is a finite system. As such, it can almost certainly be
    simulated functionally. I can see no reason why such a simulation
    wouldn't be as capable of intellect and self-awareness as humans are.
    (You DO concede that humans other than yourself are intelligent and
    self aware, don't you? If not, then you're some sort of special case
    yourself, and all bets are off...)

    Personally, I look forward to machine intelligence. Humanity has been alone for a long
    time, and as yet we haven't come across any company - we might as well
    INVENT some company...

    BTW, Searle's various versions of his "Chinese Room" argument have
    been answered again and again. Dennett takes five pages in
    "Consciousness Explained" (435-440 in my paperback edition) to address
    it in detail (how you could have missed it or any of the dozen other
    references to Searle I don't know), and Doug Hofstadter grilled Searle
    most convincingly in his "Mind's I", which see.

    To date, none of the arguments that have been forwarded as arguments
    against what Searle calls "strong AI" seem to be standing up very well
    - not Searle, not Penrose, not Nagel, not Dreyfus - unless one first
    concedes some sort of "special property" to human intelligence - soul,
    or vril, or baraka or something like that. Is THAT what you're saying?

    by sara morgan rea

    When I first wrote this post, I could have only hoped for the lively discussion that has since ensued. It is lovely to know that this site hosts such knowledgeable and strong-minded opinions. That is the one factor of human intelligence that I admire the most - the fact that everyone can have a completely separate opinion about the same subject - it is really quite amazing if you think about it.

    I never intended to get into a discussion about whether AI is achievable or not - although, just to throw in my two cents, I think it is really just a matter of time. Someone will get it right eventually, just as nature did. We are merely an engineered bunch of substances that functions quite well most of the time. This is a feat that took millions of years to accomplish though. I don't suppose we should expect even us humans to be able to duplicate that feat in just a mere handful of years.

    In fact the person I think has the best shot at getting it right (or at least close) is Steve Grand (www.cyberlife-research.com). He is my hero and lucky for me (now my friend). He is taking a completely biological approach and although he has been recently sidelined, I have no doubts that he will one day be successful at his pursuits.

    I hope this kind of debate can continue here. I think the more people we have thinking and talking about this stuff, the quicker we will get around to getting something done - after all, that is how all things are accomplished.

    Peace,

    Sara

     

    by nightlife6

    To understand just how far AI has progressed over the last few years take a hard look at the state of current automation within our strategic weapons systems. It would be for the simple minded to assume that there is a human mind behind the controls of our cruse missiles and smart bombs, alas once the targeting info is loaded, its only AI.

    by justin james

    J.Ja., this is great - no one ever talks about this stuff. I'm going to respond as best I can piece by piece - your statements are italicized (assuming I can get this through TR's little editor)...

    TR's editor is the bane of my existence.

    What you're describing here sounds more like AL - Artificial Life - than AI. The capacity to self-program doesn't require intelligence. But you've heard of genetic algorithms, of course. I assume there's some means by which you're excluding genetic algorithms and all other self-programming systems and evolutionary systems from consideration here? Perhaps you could explain in more detail why. You can't REALLY be excluding all systems that have been programmed to program themselves, right? This would be a logical tautology in this context, since it would exclude ANY human-created system a priori, and I assume that you don't want to do that - we're not just playing semantic games, right? That's what this whole disucssion is about, isn't it?

    I probably didn't clarify my point enough; I allow for self-programming. After all, it's fairly trivial to write a piece of code in an interpreted language such as Perl or Lisp that writes new functions. Heck, even Visual Studio can self-program. But like a genetic algorithm, self-programming can only occur within certain guidelines as set down by the original code, it is entirely deterministic. The code is unable to have a lateral thought, where it jumps from point A to point Z, without points B-Y, or from A to B where A and B are wholly unrelated.

    For example, I can take the standard GNU diff program, ask it to parse a bunch of pages worth of HTML code to find the commonalities amongst them, then write another piece of code to relate those commonalities to all of the other attributes of the page; their position within the directory structure, the URL path needed to reach them, and so forth. That code can then write a regex to do a "screen scrape" of the site. For all functional purposes, this software (on today's hardware!) is performing an "intelligent" operation, it is figuring out what sections of each page are "navigation", "boiler plate text", and actual content. True intelligence can do this in under a second. But it isn't the speed of the operation that matter (computers can already perform math equations significantly faster than humans), it is the fact that the computer needs to have a large sample set of data (more than one page on the site to compare the other pages to, attributes about each file, examine the HTML code, etc.). A human performs a lateral operation, a quick glance at just once page will tell that person where the navigation, boiler plate text, contents, etc. are, even on a mockup that has no links, colors, or anything other than the most rudimentary clues as to what is what. True intelligence is extraordinarily good at "filling in the blanks", even in a situation that it has never encountered before, whereas a computer always needs a sample dataset to compare the present situation to in order to "fill in the blanks". That's a world of difference.

    Searle's point is that such a system simply could NOT be built. He ascribes some magical special quality to the human mind and pretty much leaves it at that, after doing some non-specific handwaving about quantum states (or is that Penrose? All these magical arguments against AI are starting to run together in my head...).

    They run together in my head too. :) I agree, Searles does seem to be amazingly bigoted against the idea of computer intelligence based upon reasons which he does not very well define, but his arguments against the AI systems I beleive to be quite sound in their essense.

    Dennett's point (in this context, and BTW, I doubt you could ever get Dennett to admit he BELIEVES anything) is that such a system CAN be built. Ray Kurzweil thinks we can do it too, and he's putting his money where his mouth is. You might want to take a look at

            http://www.kurzweilai.net/index.html?flash=1

    I will try to take a look at this at some point in the very near future. "I am totally incapable of rational thought at the moment." - Egon Spengler

    And please don't make the mistake of assuming that the "brain" that the "mind" runs on has to be a wad of human neural tissue. I think Dennett, along with other thoughtful and technically savvy people, assumes that the human brain is a finite physical structure, and as such, can of course be replicated functionally, even if the implementation is in silicon, or differentially stressed polymers, or modulated beams of light.

    Yes, I agree 100% here, and I tried (probably unsucessfully) to make that clear previously. This is why I complete dismiss the "we don't have the hardware yet" argument. We can 100% replicate the functionality of the brain (hardware), then "flash" it with experiences (software) via sensory replicators. Alternatively, we could take a "snapshot" of the brain at any given state in time, construct a replica with that state (to avoid wasting time giving it "experience"), and let 'er rip. Would we have "intelligence"? I would have to say so. Would it be "artificial"? There's a debatable point; we used different materials, but followed an existing blueprint. "Artificial" in construction, yes. "Artificial" in design, no. For me, for it to meet the goal of "artificial intelligence" it has to not just have the same capabilities (not necessarily produce identical results, though, a failing of the Turing test in my view) in terms of intentionality and lateral thinking, but also have a manmade design.

    First, I don't recall ever saying anything like this. But assuming the argumentative stance - Well then, maybe we DO have AI. If it's really really slow, maybe we just haven't noticed... :)

    We actually don't have an argumentative stance here, you and I are quite in agreement. I was talking "in general", about about your comments specifically.

    The human brain is NOT a simple system at all, it's a massively parallel hierarchical composite with hundreds of special-purpose subsystems. What you're saying is comparable to saying that hey, a Pentium chip is just silicon and a couple of dopants, so any kid with a sandbox should be able to duplicate its functionality, and if HE can't, then NOBODY can...

    It is dense, not complex, in my view. Yes, we don't have a complete understanding of every individual component within it. It is also amazing, when you consider that every brain has a different number of nerves, slightly different design, etc. etc. etc. and yet they nearly all function (a few exceptions, of course) extremely similarly, yet can produce different results. This is one of the trickiest aspects of "artifical intelligence", it simply cannot be deterministic. The only known way to acheive that with current techniques is to use certain randum number generators seeded from outside of the system.

    Probably you're right here, but there are MANY neural organizational structures that we have not yet duplicated in software, although Kurzweil's efforts in this regard are very interesting. And once again you're misdirecting attention away from what matters, another case of the Pentiuim and the kid with the sandbox.

    I admit, I read too much Plato in my youth, so sophistry comes natural to me. :) Again though, replicating the logic gates of the brain using software is another Pentium/sandbox situation. Given enough sand and time, a child will one day etch the design for a working CPU in the sand. Sure, if he has a diagram to follow with increasingly improved resolution he will reach that goal much more quickly. From where I sit, replicating the physical functionality of the brain using software, while being extraordinarily time-consuming, is a trivial matter if you have a perfect blueprint of the brain to work from. And again, replicating the physical functionality isn't a new design, it is merely a new construction technique.

    You know, this is the second time you've repeated this argument and it just seems so much air to me. I can't even organize the sentences in a way that make sense. Does "math" not exist, because it's independent of its representation in hardware? What about "color"?

    Thank you! You actually pointed me exactly where I was trying to go! Math and color carry information that has intrisic value in and of itself to a truly intelligent system. To a piece of software, they are merely symbols. "Color" is a number, a "number" is a series of switches within memory. An intelligent system can think of any number imaginable (such as i) that can only be extrapolated vis a vis lateral thinking. An unintelligent system is unable to do this.

    I don't know of any behavioral system, including humans, that can function without hardware AND software in some form or other. Of COURSE the hardware matters. And, of COURSE the software matters. Intelligence is ALWAYS going to be a "direct result of hardware" - all software runs on hardware. But this truism doesn't mean you can't get intelligence (even an indentical intelligence) with DIFFERENT hardware. This argument is just one more fancy bit of sleight-of-hand, standing up two straw men and knocking them both down, when the actual solution is standing offstage in the dark... this is a joke, right?

    Again, we could "flash" the physical state of the brain into silicon (replicate with water pipes, whatever) and conceivably get an identical result, but we're just following the natural blueprint, nothing too special about that. And conversely, we can take a brain and alter its physical state (alcohol, lobotomy, etc.) in such a way that it is incapable of doing what it used to, because the hardware has been damaged. So there is a direct connection between hardware and software. But once again, I feel that simply transferring the software to different hardware fails in the "artificial" aspect of things. It's like FTPing a piece of Java code that works on Windows to a Solaris machine, loading it into a Java VM on the Solaris machine, and saying that you wrote a new piece of code. All you did was transfer the code to a different piece of extremely different hardware, with a middle layer to make it all work. That is what you are talking about when you discuss replicating the "software" on different hardware.

    I would still like to hear what you think human intelligence and human consciousness consist of, if not some fancy hardware and some really cool software. Some sort of oogah-boogah stuff? Gods and goddesses and souls and all that? And hey, you aren't really calling me a Cartesian dualist because I run javascript in IE and Firefox both?

    Nope, not a big fan of "soul" or "gods and goddesses" when it comes to these kinds of discussions, then you have to go through the effort of proving all of that stuff too. And yeah, running JavaScript in IE & FF is a form of dualism, in it's own way. :)

    But regarding my beliefs, to me, "artificial" is a tricky subject. I don't beleive replicating what's already out there on different hardware constitutes "artificial". For "intelligence", I beleive that it must be both non-deterministic, and have intentionality. It needs to be capable of making lateral decisions and extrapolate based upon the merest scrap of information. It must not require past "experience" to "understand" current circumstances. That's a pretty tall list of things right there.

    I think the mind is completely a product of the brain, and I think the brain is a finite system. As such, it can almost certainly be simulated functionally. I can see no reason why such a simulation wouldn't be as capable of intellect and self-awareness as humans are. (You DO concede that humans other than yourself are intelligent and self aware, don't you? If not, then you're some sort of special case yourself, and all bets are off...)

    We are in complete agreement, but again, I feels this fails the "artificial" component, at least on the leel of design. The construction is artificial, not the design.

    Personally, I look forward to machine intelligence. Humanity has been alone for a long time, and as yet we haven't come across any company - we might as well INVENT some company...

    LOL, I hope to be pleasantly surprised myself. True AI is something I used to dream about an awful lot, and I still beleive in AII and would love to be working on that.

    BTW, Searle's various versions of his "Chinese Room" argument have been answered again and again. Dennett takes five pages in "Consciousness Explained" (435-440 in my paperback edition) to address it in detail (how you could have missed it or any of the dozen other references to Searle I don't know), and Doug Hofstadter grilled Searle most convincingly in his "Mind's I", which see.

    Sadly, my copy of "Consciousness Explained" was loaned out 4 - 6 years ago, and has not been replaced yet; indeed, 95% of my books are no longer in my possession, and have been so since 2002, so I really have very slim pickings when it comes to new learning on this (and many other) subjects.

    To date, none of the arguments that have been forwarded as arguments against what Searle calls "strong AI" seem to be standing up very well - not Searle, not Penrose, not Nagel, not Dreyfus - unless one first concedes some sort of "special property" to human intelligence - soul, or vril, or baraka or something like that. Is THAT what you're saying?

    Nope, nothing "special" about human intelligence, just that most of what I've seen put forth tends to fail either on being a unique design, or being non-deterministic. But I will also admit that my readings into this particular subject are not as in-depth as yours, and they are a bit outdated, so there could very well be something out there which I have not seen and everyone knows about but me.

    Thanks for the great discussion, it's extremely refreshing, especially compared to my usual diet of "Gamecocks versus Clemson" that is the norm in these parts... especially since I don't like sports!

    J.Ja

    by xuriwan

    Staticonthewire: "Personally, I look forward to machine intelligence."

    Please consider this: You are to design a building. You are given full details of a site and access to pertinent regulations and data for all the products in the world. These are the absolute constraints.

    An automated computerised network tests every possible legal combination of products as design solutions. This only becomes practical when you add objective constraints - use, relationships, budget, time and so on. You adjust and extend these constraints in order to narrow the choice to, say, half a dozen compliant solutions. The final solution is selected by subjective judgment.

    In my mind this example is about as close as you can get to machine intelligence. The machine learns what you do not want to consider. Machine: "how about something like this?" You: "No, don't show me anything that is pink."

    by staticonthewire

    In response to J.Ja's latest comment: J.Ja. would be willing to concede the possibility of AI, but not if we duplicate the functionality of a human brain and then "flash" it. This would not be artificial enough. Even though it would be an independent self-conscious intellect residing in some hunk of hardware/software in which it did not originate. Okay. If you like. We're splitting hairs at this point, I think. But I'm very interested in this "flashing" business. It's nothing I ever mentioned.

    I'm of the opinion that if you dupe the raw functional spex of the human brain and cut the result loose in a sensory space, you'll get an intellect. No flashing necessary. And the result will definitely not be human, unless we provide a human-type environment.

    And that is Dennett's point, in a way - that intellect is an emergent quality of a specific model, whatever the medium in which that model may be realized.

    Note that I did not say the PHYSICAL functionality of the human brain. I don't think it's feasible to dupe the physicial functionality of the human brain to the subcellular level - I'm talking about copying the functionality of higher order processes and structures - that's the only point at which a few economies of scale can be realized to make the project feasible.

    Re genetic algorithms - you seem to think that genetic algorithms don't modify their own programming. I guess you could design a GA system that way, but in most cases, GA systems DO modify their own code. In a very short time you have completely unrecognizeable entities, and none of it was programmed in. Many, many Artifical Life systems using GA source have been designed - Terra is the oldest example that springs to mind. Over the course of hundreds of thousands of generations, dozens of "species" managed to evolve in a variety of constrained worlds. You might find it interesting. I have some old C source if you want to peruse it.

    Your insistence on "intentionality" is similar to Dennett's "Narrative Center of Gravity", I think, but the difference is that he's pretty sure that this particular mental entity will evolve, will emerge naturally from the interaction of unconscious elements. I find his point of view very persuasive, primarily for one reason - no one, and I mean NO ONE has a better scenario to offer. Certainly not cats like Searle or Penrose, who just surrender, say it can never be done, mutter under their breath about complexity, and refuse any further comment.

    Searle's argument fails because Searle pretty much fakes the entire argument out. He asks you to imagine (in one version - he seems to have dozens of versions) a chap in a box translating Chinese ideograms using a lookup table, and responding to the input ideograms with his own output ideograms, all without ever "understanding" Chinese. He says that a system built in this way will beat the Turing test, and then he says "Where's the intellect? Where's the self-consciousness? Not in the chap doing the lookups - he knows NOTHING of Chinese! Not in the lookup table - where is it?"

    It's easy to get snowed until you think, really THINK, about the likelihood of someone sitting in a room with a lookup table beating the Turing test. What would the guy in the room REALLY have to be doing, and with what would he have to be equipped to actually meet the criteria Searle sets? When you start thinking about that, you realize you'd need a couple trillion guys, and they'd all have to be equipped with lookup tables of amazing subtlety, and when you get to that point, you have to start wondering if the system, taken as a whole, COULD be considered conscious. I think it probably could be.

    Searle also read too much Plato as a kid, I think.

    As far as "color" and "math" being "merely symbols" to a machine intelligence - what do you think they are to YOU? If we duplicate all the functionality of the human brain and then "flash" it with the J.Ja personality, won't "math" and "color" be "merely symbols" to the copy? But according to your earlier comments, you would think this copy would be an intellect separate from your own...

    In response to xuriwan's comment - my friend, the design you describe is only one of many designs for exploring a search space. Google "heuristics", for example. Check out genetic algorithms. Or semantic nets, or opportunistics frames, or artificial life, or subsumption architectures, or any of dozens of other strategies for getting from point A to point B. A brute force exhaustive search is certainly NOT the only possibility.

    And to you also I would pose the same question I've been pointing like a really kool intellekshual shotgun at all the anti-AI arguments presented here: What do you think YOU are doing, when you think? And what makes you think that what you're doing is so special, so impossible to duplicate in some other medium?

    by dogknees

    I also agree that most people don't believe it's real or possible. But all that says is that most people don't have a blue clue about anything more complex than a bicycle. It says nothing about the reality or possibility of AI. This seems to be a growing phenomenon these days. Reality as a shared belief rather than an objective fact.

     

    Regards

    by metilley9

    When I think of Artificial Intelligence (AI), the HAL9000 computer from the film, "2001: A Space Odyssey", comes to mind.  Even though it made unpopular decisions in the movie, it still "thought" for itself.  It was protecting "itself" and the mission.  I realize that HAL9000 was fiction, but for 1969, this film was amazing for it's time.

    I don't consider what most people call "AI" to be real AI.  Setting defaults or assigning certain values (or computing values determined by algorithms), based on conditional tests is hardly "Machine Thinking".  Unfortunately, this is what most people think of when they use the term, AI.  I think a new [computer] language will have to be written before true AI happens.  I have used Lisp before, a very basic string processing language, but others need to be developed.

    by sara morgan rea

    Recently I wrote about the fact that there are many AI-based applications being used today. At the time I did not give any specifics, so this is the beginning of what I hope to be a series of examples.

     

    Today?s post will involve how computers recognize things. This can include both visual and auditory inputs. The information is extracted from the recently published, ?Robots Unlimited? by David Levy.

     

    In chapter 4, Mr. Levy refers to actual applications that utilize computer vision. He briefly discusses the current strategies used to accomplish human face recognition. Essentially, the computer just divides the image into regions and then compares specific areas against a large database of stored images until it finds a match. No magic, just clever programming (of course, that is all ?Soft AI? is really).  He mentions an application that was used by the Mexican government in the 2000 presidential election. The application was used to identify duplicate voters since many Mexicans like to vote more than once under different names.

     

    Mr. Levy also mentions a clever system developed by a team at IBM, called Veggie Vision. Veggie Vision is used to identify over 150 different fruits and vegetables and can be used by supermarkets to streamline the checkout process.

     

    In the area of automobiles, a specialized chip named EyeQ (clever, heh?) can be used to help guide automobiles through traffic. It won?t replace drivers, but it can be used to help guide the driver and also to anticipate a car crash.

     

    Lip reading is a technology that could be especially important for improving the accuracy of speech recognition systems. Researchers at IBM are Intel are working on systems that do just this, although the results so far have only been significant in a laboratory setting.

     

    Sharp has been working on a technology that can be used to identify specific movements from sports videos. It is called HiMpact Sports and it uses a Hidden Markov Model to do it?s magic.

     

    There is even a system that is used to perform taste recognition. Developed by the Brazilian Agricultural Research Association, the ?electronic tongue? can recognize the taste of water, wine, coffee, etc.

     

    So, I hope that gives you a good idea of some of the wonderful advances that have been made in the area of computer recognition. This is just one area of AI that can be used to accomplish Enhanced Computing.

     

    Peace,
    Sara

    by jmgarvin

    I'm currently working on data mining for music (and eventually video)
    without obvious contextual clues.  What I mean is, if the header
    is empty (trivial) and there is little in the way of context around the
    file, how do what truely know what it is?

    I'll post more on my blog as I progress.  Right now I'm fighing
    with my Fourier Transform analysis...so I'll keep everyone posted...

    I still don't believe that AI is the end all be all, and I don't see
    that it is efficient...However, some problems can only be solved using
    them (this is one of them).

    I'm leaning towards using a combo of fuzzy logic and ANN, but I have a
    feeling my preprocessing analysis will change...so...we shall see!

    by sara morgan rea

    This week, a new web site for Microsoft Speech Server developers went online. The web site is run by my good friend Marshall Harrison. Marshall is the lead speech developer at Landstar (they were featured in my book). Marshall has been working with Speech Server since before it was released and he is probably the most knowledgeable developer in the world regarding the subject. He has recently been working on an outbound calling application and no one else is doing that. He is great at pushing the boundaries of Speech Server so his insight is quite valuable.

    Anyway, I encourage anyone even a little interested in Microsoft Speech Server to go to this web site and sign up. I really think the Speech Server community needs this.

    Thanks Marshall for taking the time out of your busy life to set this up for every one.

    Peace,

    Sara

    by mharrison

    Thanks for the kind words Sara.

    We are having a few minor growing pains but things are looking good. Now I'm just waiting for the users to show up in hordes.

    "If you build it they will come"

    by dawgit

    sounds like a super idea, but I don't have the time needed to play with it right now. And it's still

    too expensive to play with. For that much money I can get a toy that will move rather fast. I have

    had an experence with the technology however, and must say that devolopers should program in

    more information to inform users that a voice system is in use, and let people know when for example

    the system switches from DTMF to Voice.

    Something like:

    Sys: "You pushed #1, Is that correct? Or do you want to change your data? Yes? or No?"

    User: beep, beep. beep, beep, beep........ "sh_t! What button do I push for 'Yes'?"

    Sys: "Yes? or No?"

    User: beep,..... beep.... "Wait a minute... is this a joke?"

    Sys.: "Ok, now quit pushing the little buttons and talk to me."

    User: "I'm trying to.... You stupid telephone... What do you want me to do?"

    Sys: "Speak into the phone"

    Sys: "Yes? or No?"

    User: ".............................. Yes (?)"

    Sys: "You said 'Yes', Thank You. Now the next question......"

    Yes, it did go something like that actually, because I did not know that the service I had called into

    was useing a Voice Server, and had switched over from 'push the right botton' to 'say yes or no'.

    Those systems are in use already over here in Germany by the way. But thanks for the info. -d

    by sara morgan rea

    The previous commenter was correct in that many speech systems these days are not exactly "user friendly". Unfortunately, speech as an interface tool is very new not only to users, but developers as well. And, it is still relatively expensive (although, this is getting better all the time). You can learn the technology using free emulators and Software Development Toolkits (so to not look into it because of cost is not really a valid excuse).

    But, back to the idea of speech systems being unusable - I agree, many are. But, that does not mean the technology should be abandoned or ignored. What is does mean is that sites like GotSpeech.net are needed more than ever. Marshall started the site because there is a growing number of speech developers and a very limited amount of useful documentation out there on how to do it right.

    Don't throw the baby out with the bath water, dawgit.

    Peace,

    Sara

     

    by dawgit

    I agree with you Sara, and I also think it's going to the next best business tool to come along. And I did note the site you mentioned, (as well as others) as soon as I get some time.............. I will give it a spin. Nope, I'm not throwing the baby out yet. As I said it's already in use over here in Germany. It could be the answer to the Out-Sourcing/Off-Shoring problem everyone seems to worried about. And the system is programed to be always friendly :-)   -d

    by sara morgan rea

    CyberTwin is the name of a new product currently under development by an Australian based company named RelevanceNow. The company was founded only a few months ago, but they have some pretty ambitious plans and I think they just might be on to something.

     

    Their flagship product (of which, I have only read about and not actually seen yet), is a software product named MyCyberTwin. MyCyberTwin in an AI-based software agent that is supposed to represent a web sites owner. It does this by using AI-based techniques to analyze the web sites content (such as blogs, web site content, and even emails) and then chat with visitors as a representative of that sites owner. So, companies and individuals can do more than just use static or even dynamic content to attract visitors. Cool concept, eh?

     

    The concept is very sound and I love the idea of the software learning about the owner without the owner having to do any manual configuration. Of course, I am a little skeptical and I would like to see it before I start tooting their horn too much. But, I have shot them an email offering my assistance as a beta tester, so hopefully I will have some more scoop for you soon.

     

    Peace,

    Sara

    by dawgit

    It could be a keeper, I see a use for it, But, it also sounds kind of scarry too.

    by dawgit

    ok, that's one application, (not if was my Avitar, with my voice, -that would be scarry), but you could drive the witts out of some gramdmother somewhere. No, seriously, It can be useful, no doubt about it, but we all have problems with non-human call centers, that will get worse. And what of language?

    by sara morgan rea

    At the very top of my "Cool stuff that I must check out when I stop having to write books all the time" list is www.microsoft.com/robotics" target="_blank">Microsoft's Robotics Studio (released on June 20th). This baby is way cool. Using Web services as the foundation to build modular services for hardware and software, the toolkit allows developers to easily (well, easier then it would be normally) create scaleable applica-tions that control robotics eqiument and sensors.         

    Microsoft is teaming up with LEGO on this one and so there will be a lot of information on how to use this with LEGO Mindstorms, but the platform promises to be extensible so that it can be used on virtually any platform (cough, windows-based platform). Anyway, it also features a really cool simulator that allows you to simulate a robot before you actually build one. This should be GREAT tool for teaching robotics to kids (and even us big kids).

    This is really cool folks. And if you think this is some future technology that has precious little to do with what goes on in the real world, think again. Robotics is a RAPIDLY growing industry. In addition to the numerous robotic toys available at your local toy stores, practical robots that sweep your floor and retrieve things for you are steadily entering the mainstream. Trust me - It will not be long before your boss comes to you and asks you whether you know anything about robots. It's coming. I promise you. Things like the Robotics Studio are like godsends to us everyday developers because now we can do those things using the same development environment we have come to know and love (.NET).

    So, check it out and have some fun. You won't regret it. Robotics is addictive.

    Peace,
    Sara