Search This Blog

Sunday, April 26, 2015

All you wanted to know, but your brain didn't let you


In our age intelligence has occupied a position of God. It exists (or we believe it does), we can't see it, but there are signs of it, everyone has opinion about it, but we can't prove it and changing the point of view – changes its definition. Jeff Hawkins in his book “On Intelligence” wittingly notices that the position of scientists (some of them) about complex and incomprehensible human brain is similar to old centuries' position about the futility of exploring the sky (where God lives) since we will never reach it. (Have you notice how the semantic of “never” changes depending on whether past or future is concerned? Compare “It has never happened before” and “It will never happen!”).
Yet, as seeking the glory of God, we are contemplating to create it: Artificial Intelligence! What a courageous pursuit to create something that we cannot yet define. J.Hawkins explores this fallacy in detail: we create robots, that can do complex calculations in nano-seconds, but cannot distinguish the picture of a dog from the picture of a cat (well, update, apparently now they often can http://xenon.stanford.edu/~pgolle/papers/dogcat.pdf). Long quantitative repetitive predictive tasks are easy for machines, yet our intelligence get tired even to remember more than 7 numbers. In the same time 6-year old will easily name a range of animals from the pictures, whereas computer will get stuck, especially if the picture is not of a full animal. Artificial Intelligence may be intelligence (depends on how you define it), but it is indeed very different from ours.
Interestingly, our brain is rather slow compared to transistors. And the information we have to deal with every second of our existence is enormous – sounds, light, people speaking, memories, books, televisions, driving – and it all changes all the time! How do we cope and even completely outperform super-fast computers with endless memories? We have an interesting brain. Surprisingly it works mostly on memories. Yes, when we need to make a new step or catch a ball – we do not calculate the trajectory and our relative positions. We just remember. Remember so well that it happens automatically. And the ultimate paradox is that any step, any ball-catch is different. We can only remember what we've learned before and yet – before we've never experienced what is happening now. And still we don't calculate as a robot would do.
No, I don't have an answer. The working hypothesis is that we only remember invariant patterns and adjust each time (I assume by remembering the adjustment pattern). It's fascinating. It's even more fascinating that you can actually try to program a brain yourself, totally free. Yes, you are welcome:

http://numenta.org/

No comments:

Post a Comment