AI has a history of overpromising and under delivering

Starmind founder Pascal Kaufmann explains the difference between machine learning, natural language processing, automation, and the other components of artificial intelligence.

AI has a history of overpromising and under delivering Starmind founder Pascal Kaufmann explains the difference between machine learning, natural language processing, automation, and the other components of artificial intelligence.

Starmind founder Pascal Kaufmann sat down with TechRepublic's Dan Patterson to discuss his skepticism about what's considered AI today, and the great difference between small machine learning (the human brain) and big machine learning. AI, he asserts, is just what the original programmer input. Kaufmann, applying his biology background, looks for more creativity. The following is an edited transcript of the interview.

Dan Patterson: Pascal, let's start with the practical implementations of artificial intelligence. In fact, let's get rid of the word artificial intelligence. There are so many preconceptions about what AI is. Help us understand the difference between machine learning, natural-language processing and the other components of AI that make practical sense for business.

Pascal Kaufmann: There is a hype in AI, and there's a lot of confusion about terminology. We have to distinguish between automation, digitalization and artificial intelligence.

Once we can tell these terms apart, all the others fall into port. Automation and digitalization, they are a synonym these days. We automate things in several thousand years. It's nothing new. I do not think, however, that you have artificial intelligence these days.

Dan Patterson: Automation, the magic word there, which might be better than machine learning or AI. We hear so much about automation changing the workforce, changing the economy, changing jobs. What's the reality?

Pascal Kaufmann: Automation, for example, a hammer is also automation. I mean, you could use your human hand by planting a nail. That's very cumbersome, so we take a hammer, and you automate, you amplify this process. Automation is something that is really not new.

When you do it with computers, you automate, for example, calculation. You automate image recognition. Digitalization and automation are very closely related.

Dan Patterson: A moment ago you said you don't think we have AI yet. I know we are being sold AI as B2B packages, and the consumers are certainly being sold AI as something that fits in their smart phone and their smart homes, smart offices. Why don't we have AI and when will we have it?

Pascal Kaufmann: AI has a history of overpromising and under delivering. We don't even know what intelligence is. To sum it up, everything that we label AI these days is nothing else, as the intelligence of a human programmer put in the source codes. Like an engineer has given a lot of thought: what happens if this is this, and then we call it AI? I find the limits cheating. I'm interested in the real thing, a machine that is really creative and doesn't just replicate what has been put into it.

Dan Patterson: So, often an algorithm is used and referred to as AI, or packaged as a business tool we call AI or machine learning tool. I assume you have your own definition of AI, but when your colleagues meet, and we know that you have spent a lot of time trying to crack the brain code, what do professionals in the space consider artificial intelligence?

SEE: Digital transformation: A guide for CXOs (Tech Pro Research)

Pascal Kaufmann: It's interesting... the difference of people. A neuroscientist or brain surgeon, for example, they very rarely talk about artificial intelligence. They're humbled by the human brain. They noted we are actually not even close to cracking the brain codes.

On the other hand, there are people that are constantly talking about artificial intelligence. Often these are vendors of software products, and it can be much better sold if you call it AI inside. I really think it depends on your background and your expertise. People who know a little bit about the brain are very skeptical about today's progress in artificial intelligence.

Dan Patterson: Your skepticism is also coupled with an optimism. You're one of the people I know that I can really get real AI answers from. What is your background in brain science and how has that informed your work in terms of artificial general intelligence?

Pascal Kaufmann: Almost 20 years ago at the Chicago Medical School, I dissected brains out of animals, put them in dishes with artificial blood and kept those brains alive for a few more days. It looked like Frankenstein's office. My task was to enter the electrodes into the nerves of these living brains, connect these living brains with machines. As soon as you have to talk with the brain, you know exactly what you know and what you don't know.

SEE: Artificial intelligence: Trends, obstacles, and potential wins (Tech Pro Research)

Twenty years ago, I figured out that we don't know a lot about the human brain. Because of this biological background, I'm very, very skeptical these days. To put it a little bit bluntly, if you need 300 million pictures of cats in order to be able to say it's a cat, a horse, or a cow, I do not deem that very intelligent. So, whenever I read deep learning or artificial intelligence, I get a little bit skeptical. To me, it would be much more about the small data. The brain is definitely a small-data machine and not a big-data machine.

Dan Patterson: I'm so glad you said big data because it sounds as though that runs counter to the notion that well, with more data we can create better AI and we can have better systems. You're saying that's incorrect?

Pascal Kaufmann: Yeah, it's actually even the other way around. If you really need so much data to figure out that it's a cat or a horse or a cow, I mean it can be so smart, the algorithm behind it. A child, for example, looks at one cat, cuddles it a little bit, and knows once and for all what a cat is. So, the brain is definitely a small-data machine. If you go for the big-data processing, I think you go away from understanding the brain. It's just like very powerful statistics. I never was good at statistics. Maybe that's the reason why I don't like statistics. But to me, statistics and the intelligence are really not related at all.

Also see