AI is already all around us: but when is your PC going to be smarter than you?
AI, I've heard of this. It's when machines try to take over the world and eliminate all life on earth isn't it?
Sounds like you've been watching too many Hollywood blockbusters.
The Matrix, Terminator, Blade Runner - awesome!
While those are indeed awesome films, they are more sci-fi than AI - at least the kind of artificial intelligence that's around today.
What kind of artificial intelligence is around today?
Let's go back to the beginning. Back in the 1950s when the research field was being established one of its early pioneers, the scientist Marvin Minsky, defined it as the science of making machines do things that would require intelligence if they were done by humans.
Nowadays, however, scientists sometimes refer to this as the 'old definition' or the 'classical definition' because the focus of research has shifted since those formative years. Instead of its early emphasis on human intelligence, the AI umbrella today covers all sorts of processing mechanisms that have been made by humans.
What sort of mechanisms?
Both technological and biological.
In the biological realm, some scientists are already conducting experiments with hybrid AI, connecting biological neurons taken from rat brains to electronic robot bodies via a multi-electrode array and using the biological component of the entity as the controlling mechanism for the body, for example.
Other, more straightforward technology mechanisms can also be considered AI: software brains driving and manoeuvring a robot body, for instance, or a program running on the internet.
But how do you actually go about determining if those mechanisms are intelligent or not?
A good question. Intelligence as a concept is subjective and hard to definitively pin down.
What about the Turing Test? Doesn't that prove when machines are artificially intelligent?
The Turing Test was the brainchild of the English mathematician Alan Turing who, all the way back in the 1940s, was interested in the notion of intelligent machines and whether machines can think.
In a paper published in 1950, Turing set out an idea, not of testing whether a machine is actually thinking - he decided it was simply too subjective and difficult a concept to pin down - but whether a machine can at least appear to be thinking, when judged by a human.
The test he proposed was for a human judge to converse separately with one machine and one human via a text terminal and then have to decide which was which.
If the judge did not consistently identify the machine as a machine, it would have succeeded in appearing human and thereby passed the test.
Since 1991 the Loebner Prize for artificial intelligence, sponsored by US inventor Hugh Loebner, has run versions of the Turing Test with AI chat programs pitting their digital wits against human interrogators.
While machines regularly fool judges into thinking they're human, these days the Turing Test is considered by some in the AI field as an interesting experiment or a tick in a clever box but not much more a sign of intelligence than when Big Blue beat Garry Kasparov.
What's this about Big Blue?
Back in 1957 another early AI pioneer, Simon Herbert, predicted a machine would beat a human chess champion within a decade.
His prediction came true - albeit 30 years later - when IBM's Deep Blue supercomputer beat world chess champion Garry Kasparov in 1997.
IBM's Deep Blue
(Photo credit: Pedro Villavicencio via Flickr.com under the following Creative Commons licence)
So did the match herald the era of intelligent machines has arrived? Not exactly. Let's just say the prevailing sentiment became more that just being able to play chess was not a sign of genuine artificial intelligence and was instead more a matter of brute processing force and good engineering.
In essence, Deep Blue is an example of narrow AI: a system designed to carry out a specific application rather than to exhibit the kind of general intelligence we humans have.
Narrow AIs are where a lot of real world examples of AI can be found.
What sort of examples?
AI applications have found their way into many systems underpinning Western society.
If you take a broad view of what AI is, it's possible to count hundreds of apps in use throughout our infrastructure - from intelligent algorithms routing comms data, to computer-assisted design that produces sophisticated gadgetry, to autopilot programs that fly and land aeroplanes - that developments from more than half a century of AI research have contributed to.
Many very modern applications such as web services are essentially made possible by aspects of AI research too.
Google apps for example - its search, translation, voice recognition, advertising and spam filtering to name but a few - are all dependent on AI techniques.
Google's spam filter for instance uses machine learning - a scientific discipline that feeds into AI where systems typically improve over time as they handle more data. In the case of Google's spam filter, it asks real users to label what's spam and what's not and then improves its own filtering by studying their results.