Monday, September 29, 2014

Artificial Intelligence Take Over?

Artificial Intelligence is looming, they say. I mean, real AI, the kind of AI that would potentially measure up if not flatly beat our best brains ever. The argument goes with examples such as how powerful at chess the computer has become, or those software which are able to tell stories by themselves, and so much more. The total computing power is exponentially exploding and soon the number of chips that can be put on a grid to build the Internet of CPUs will bear such a colossal neuronal capacity that our brain in comparison will be dwarfed away like if in prehistoric ages. Well, I may be totally wrong, but I seriously doubt this thesis.
Because this analysis forgets what counts most: software. Billions of CPUs are dumb if they do not run the right software. Computing power is indeed exploding and now makes possible applications that were unrealistic, even undreamt of a few years ago. More sophisticated algorithms and intense data crunching are now processed in much less time and make it possible for complex decision making to happen in human-scale real time, this is absolutely correct. And these algorithms give the impression of a growing computer intelligence, laptops are getting smarter and smarter. Well, so they seem.
But all the given examples, all the magical abilities of computers rely on some existing algorithm, precisely. Chess playing can be impressive, it is in fact not very clever. It is not difficult to write a simple chess playing program – ok, a really good one is harder, but still pretty primitive in terms of cleverness. There is no true intelligence in those codes, they are pretty boring in many respects. If you find a way of playing that the coders had not anticipated, they cannot adapt and fail miserably. All of them.
And this is the point. Computers can only be so clever as their coders enable them to be. They simply cannot invent a behavior that their designers had not foreseen, directly or indirectly. Thus, foreseeing AI to soon come true either is totally wrong or requires or foresees that soon we will be able to code software that makes possible behaviors or decisions that were not anticipated by the designers. I am probably mentally limited, but this sounds hard if not totally unlikely to me.
On a TED talk, Larry Page from Google took the example of a start-up they recently took over that has developed a software that can ‘view’ videos and detect patterns out of nothing – example was that of a cat – and then matches such a pattern to its related concept – here the word ‘cat’. Fantastic indeed. But again, this is not intelligence, but data crunching. All the intelligence if any was given by the ability of the programmers to express an algorithm that would make it possible to match similar patterns on videos. The computer is not intelligent, the software is neither, only the programmers were, if anyone.
Yet, there is a way, say the visionaries. The exploding computing power will make it possible to mimic the brain in its design, resulting in conscience and intelligence to emerge from the Grid. Maybe, one day. But before, to be able to do that, we need to understand the brain itself to code it into the Grid. Or we need to devise a neuronal mechanism that, when injected into every single CPU-turned-into-a-neuron will make it possible for the neuronal activity to produce some kind of “thinking” at Grid level.
It might, it will probably eventually happen. Not soon I believe, only when we understand our own brain, our own neuronal software, so that we can then code it. And even then, I am still wondering: Will Alan Turing’s theorem not forbid such a possibility to ever happen?

1 comment:

Mako Program said...
This comment has been removed by a blog administrator.