A book with many fascinating insights into how the mind works, sadly flawed with some brash assumptions and glossing over important issues.
So his aim is to correctly understand how the mind works – he says most other people have got it wrong, and no one has a grand theory of how the mind works like he does. Ahem, yes, bit grandiose and martyr-like. The book is that, i.e. a grand theory of how the mind works. It’s the first part of his work. The second will be to figure out how to truly put this algorithm to work inside a machine. He says that while the AI industry has come up with some great applications, they missed the point as they started implementing AI before they fully understood the intelligent brain they were aiming to base it on.
I laud his aims – he made his money building personal gadgets (he designed the Palm Pilot) and is now spending his gains on his intelligence institute.
In summary, he is a long way from his grand claim of a complete understanding of the way “intelligence” works, but what he has produced is a good step in the right direction towards increasing our understanding.
Let’s get the negatives out of the way first:
On p41 he makes the big, big assumption that only neocortex houses intelligence. This is pretty brash, given how we’re still learning about how the mind works and so much other stuff I’ve read shows how all the parts of the mind influence the others. That said, he hadn’t defined intelligence at that point.
And to say on p43 that the mind is produced only by the brain period is also bold given all the research about body-mind, and the nervous system around the body which many think has a lot more to play in the makeup of the mind than we intuitively assume. It has been argued that the mind wouldn’t function without the body, c.f. the feedback stuff he mentions. And this seems to be an unnecessary assumption.
And to his excellent paradigm for understanding how the mind works:
He posits that the mind can take any input – we have sight, hearing, etc. and learn to process it. It works the same for each: it takes in data over a period of time. Sight is not a snapshot – the eye has three “saccades” every second; it takes in a little part of the field of vision each time and builds up a picture over time. Similarly and more intuitively with hearing – we process a series of sounds over time. It wouldn’t make any sense if we simply had a snapshot of sound at a point in time. And so with touch – if you wake up touching something you can’t figure out what it is until you’ve moved along it, i.e. a sequence of touch input over time.
Then the cortex is made of 6 layers of neurons, each layer holding data that are an abstraction of the data in the lower layer. So for example when you hear music the lowest layer will register the notes, the next layer will put those notes into riffs, and so on. Or when you’re reading you’ll get letters at the lowest level, morphemes at the next, then words at the third, then phrases, and so on until you have more abstract understanding at the top.
When we’re learning something new, say reading, the simple part, i.e. the letters will go right to the top layer and we’ll be aware of that. As we learn, the letter bit goes down to a layer of which we’re not aware, and we can think more of the words, and as we get a bit more adept the meanings are all we need to consider at the top layer, and so on. So as we practise something, it gets so that we need to consider less of the details, which means we’ll only be conscious of the highest, i.e. most abstract layer.
Unless of course there’s an “error”. Only errors filters up the chain, say if you are walking into your house, the way you always do, but you suddenly notice that a floorboard is loose – that will shoot up the layers until the higher layers become aware of it. Otherwise all the actions are pretty autonomous.
So the abstraction process is what the brain excels at, and something I’ve always intuitively thought the brain did too, so good to see someone else confirming the theory.
Now given his system for how the mind works, the final chapter on applying this algorithm to all walks of life is most inspiring. The idea of this algorithm’s ability to learn given any input is powerful indeed. So just plug it into a camera and a car, for example, and it would use that same algorithm to figure out driving. And once we spend the time training one system we can simply copy it and refine it. It would truly revolutionise our world.