AI

Or the poor state of AIs.

People are scared of AI. Commonly they fear when Google seems to be to smart to be benevolent, when retargeting hit them hard or when we are confronted to our own limits that computers seem to ignore. Hollywood doesn’t help. Even Stephen Hawking and Elon Musk seem to share in their ways this fear. Sometimes for good reasons, sometimes for not so good ones. But, AIs don’t exist, at least not yet.

When people are talking about AI, they often make references about Google Search. Google is not an AI. They wish! It’s, at best, a gigantic crawler that put links into an enormous database. That’s it. They can sprinkle some social features, videos, images, rich page content, add a calculator, add weather forecasts, add a zerg rush… it’s still just a bunch of scripts that loop, fork, iterate around the same data. We can’t even say that’s successfully mimicking basic intelligence.

My definition of what’s an AI is indeed strict. A.I. stands for Artificial Intelligence, not Virtual Intelligence. I am not considering faking intelligence as the goal that to be achieved with AIs. It will be actually sad if we can’t pass that. Even if you manage to put all the knowledge of the world in a well-designed machine and empower it with complex conditions, statistics and machine “learning”, and it appears to be able to converse and make decisions. It will still be confined into what has been coded for it. Not being able to “truly” learn. Not being able to look past statistics. Not being able to achieve consciousness. And above all, not being able to evolve, replicate or reproduce.

Following this definition, I won’t consider, obviously, Siri or IBM Watson as an AI. They aren’t even embryos of AIs. That doesn’t mean they are not achievements and can’t be useful. Considering even the best chat bots, they are laughable and won’t foul a monkey for more than 30s. A better contender would have been IBM Synapse where neurons are basically artificially recreated. When they have launched this project a few years ago, it seemed exciting with buzz accomplishments like real auto-learning or the completion of an artificial brain said to be already as smart as a living mouse. But, now, the project seems to have failed its ambitions and they are back to more traditional approaches where the chips are just a different kind of CPU coupled with a special kind of RAM where you need to use a special programming language to take advantage of this special architecture. In top of that, there is no commercial release dates at the horizon.

ttn-emacsA good example about how AI are nowhere near to be intelligent is to study the game of Go. (Or also just considering how much computation is needed to challenge a simpler chess game.) In Go, despite decades of works and brilliant engineers passionated by the task, strongest algorithms can’t beat professional Go players and any average player can actually beat any software currently on the market. Because these softwares all lack the natural intuition coupled with a global sense of the board that humans gain automatically by playing the game. A true AI should be able to achieve that. Even without being explicitly taught anything.

Recently, a lot of hype has been around quantum computers. The hype is real but maybe there is no reasons for it. Reading scientific studies, quantum computers are just theories. For now, it’s not has been proven that any of the prototype are working nor that anything exploitable has been extracted from any of these prototypes. Companies currently operating in this industry (I am looking at you D-Wave!) seem to be nothing that elaborate scams. I presume the future will tell us if they are going to deliver anything of what they are promising.

Don’t be afraid of something that doesn’t exist yet and may not even exist in the humankind future. Don’t be afraid of something that we can’t understand yet. I do believe though if a true AI finally emerges from us or by accident, the odds that it will consider us as nuisance that needs to be dealt with, in a gentle paternalist way or a more aggressive one won’t be nil. So, actually, yes, you can be afraid a little if you want! But we are far from that.

PS: One of my deep personal hopes is to be able to see the emergence of a true AI before I die.

Advertisements