On Intelligence - Jeff Hawkins

Can we create intelligent machines based on an understanding of how the brain works? 

 I liked this book a lot, since it stimulated my thinking off in new ways. However, whilst it has some great stuff in, there's also a fair amount of sloppy thinking. At least one of his central premises I question.

Hawkins gives us a useful intro to the field of AI and some of his own history. There's no doubt of his passion for the subject. In a nutshell, AI has concentrated almost entirely on the view that intelligence means intelligent behaviour. Hawkins questions this view and says there's a lot more to intelligence than behaviour. I agree. But then he swings to the opposite extreme and claims that a person sitting quietly in a chair thinking is intelligent. Here's a sample of his writing (p33): We demonstrate human intelligence through our speech, writing and actions, right ? Yes, but only to a point. Intelligence is something that is happening in your head. Behaviour is an optional ingredient.

Its his 'yes, but only to a point' response that I simply don't buy. Its sloppy thinking - we don't demonstrate intelligence in any other way than our behaviour. So, far from behaviour being optional, its essential. That's not to say that intelligence is merely behaviour though. That's not true either, so Hawkins does have a point, and its his focus on what's going on which isn't directly reflected in behaviour that makes this book a worthwhile read.

Here are some more examples of muddled thinking - (p40): Your neocortex is reading this book. He's obsessed with the neocortex, having studied it so long - the neocortex is an essential component of the whole system, for sure, but it of itself doesn't read a book - I read the book, and there's a lot more to me than my neocortex. This might appear to be hair splitting ( that's what philosophers do, after all, isn't it ? ) but the error runs like a thread through the book. (p41): An intelligent machine need not have sexual urges, hunger, a pulse, muscles, emotions or a humanlike body. Again, I do disagree though I can't find arguments right now - how can a machine be intelligent without being curious about its environment, curiosity being an emotion ? (p43, speaking of the neocortex ) Those thirty billions (sic) cells are you. Nope mate, they're not - they're certainly an essential part of me, but they're not me ! Still, on p43, we find Hawkins agreeing with Crick - That the cells in our brains create the mind is a fact, not a hypothesis. I still don't buy it simply for just Hawkins stating it.

Mountcastle's theory is elegant and Hawkins explains it well, I buy this because it explains a lot of things. There's basically one cortical algorithm running wherever the sensory input comes from - ears, eyes, nose etc..  Each sensory input simply produces a pulse train. The brain then gets to work on that pulse train to create our perceived world. There are patterns from the ears, patterns from the eyes and patterns from the other senses. So intelligence is separate from the need for having one particular sensory organ. Where Hawkins is at his weakest is when he starts to talk about philosophy - consider this excerpt :

Finally the idea that patterns are the fundamental currency of intelligence leads to some interesting philosophical questions. When I sit in a room with my friends, how do I know they are there or even if they are real ? My brain receives a set of patterns that are consistent with patterns I have experienced in the past. The patterns correspond to people I know, their faces, their voices, how they usually behave, and all kinds of facts about them. I have learned to expect these patterns to occur together in predictable ways. But when you come down to it, its all just a model. All our knowledge of the world is a model based on patterns. Are we certain the world is real ? .... Existence may be objective but the spatial-temporal patterns flowing into the axon bundles in our brains are all we have to go on.

This kind of writing raises all kinds of questions of consistency from a philosophical standpoint. What's 'real' - this is a question Neo asks Morpheus in 'The Matrix'. Morpheus points out what Hawkins has already noted - what's 'real' is simply electrical impulses in your brain. Even that view is not consistent since Morpheus failed to factor out that the brain is part of the 'reality' we have begun to question. Electrical impulses are also being called into question - since those are observed in our 'reality' but who knows what's really going on behind the scenes ? Thus we need Donald Hoffman's style of thinking to go further - that 'reality' is a user interface and tells us nothing of the underlying hardware. The brain itself is part of that user interface - its daft to exclude it and assume that when we talk about electrical impulses in neurons, that's not also part of the illusion of reality. Having said that, then Hawkins is going beyond his remit to say 'Existence may be objective but...' - nothing is objective ! Here's where for my money, Hawkins displays a complete lack of humility (p64):

Can we trust that the world is as it seems ? Yes. The world really does exist in an absolute form very close to how we perceive it. However, our brains can't know about the absolute world directly.

Complete poppycock Mr H - you have a brain just as I do, so what's your point of reference for these claims ? Or are we to assume that you have a hotline to God ? Hawkins hasn't consistently embraced constructionism, he's straddling two worldviews.

Chapter 4 is entitled 'Memory'. Not much here to disagree with - the brain doesn't function like a computer. Hawkins claims that the brain doesn't compute, rather it recalls pre-prepared answers from memory. Memory storage is also different from a computer - its auto-associative. It is also sequential  - we recall things in sequence - like the alphabet, which is much harder to recite backwards. The notion of the 'invariant representation' is also introduced. Once we start considering 'representation' then we might need to pay special attention - since the world in the brain isn't a representation of the (objective) world, it IS the world. Given that Hawkins hasn't embraced this view (at least he hasn't consistently) we're quite likely to find some inconsistencies here too. However Hawkins limits these representations to memory, so we might be fine.

So, what of these invariant representations ? (p75) 'The brain doesn't remember exactly what it sees, hears or feels ... the brain remembers the important relationships in the world, independent of the details'. Then we get some examples of what he says is going on (p76) 'You are probably holding a book in your hands right now... the pattern of light on your retina changes completely....Yet you don't have any doubt that you are holding a book, indeed the same book. Your brain's internal pattern representing 'the book' doesn't change even though the stimuli informing you it's there are in constant flux. Hence we use the term 'invariant representation' to refer to the brain's internal representation'.  Hawkins doesn't make clear here whether the representation is a visual one or a cognitive one, but its one held in auto-associative memory. The next example is recognising a friend - again, it doesn't matter what the angle or lighting or distance is - you can recognise someone if you know them. In this case, the internal representation clearly must be a visual one, when there's a match then the feeling of recognition appears. I think it was in 'The Holographic Universe' that there was mention of a holographic theory of performing this kind of pattern matching - whilst its practically impossible to get a match using straightforward images, if they're converted to holograms then it becomes much easier.

Hawkins says that in the hierarchy of neurons, the lower levels don't show invariance but the higher levels do. Some neurons display stable firing patterns only when the recognised friend is within the field of vision. Hawkins extends the concept of invariant representations to the way we remember songs irrespective of the key they're played in. It means we must relative-ize the intervals rather than recalling the absolute notes. If we recall a face, we go for the relative dimensions of the face - the proportions of the features, not the absolute values.

Chapter 5 is all about how the brain makes dynamic predictions as we go along, and constantly verifies those predictions against what happens. We only get things flagged into awareness when the prediction mechanism shows some variance with reality - a sudden surprise. Here I think is where Hawkins gets his notion that behaviour is optional from : (p85) 'As I looked around my office that day.. my eyes saw...yet just seeing...didn't cause me to perform any action. No behaviour was invoked or required, yet somehow I "understood" the room and its contents". Hawkins had an 'aha' moment at this point, something dawned on him about the nature of understanding. Myself, I do wonder what this 'understanding of the room and its contents' means. My guess is that he was able to recognise the objects and attach labels if he so desired. Looking in more detail at his words, we do find behaviour where he noticed none. His eyes were actively looking, not just passively seeing. He was by no means a passive observer - as he's already told us, if the eyes don't move (saccade) then everything will fade to grey. So perception is activity, perception is behaviour too. 'Prediction is so pervasive that what we 'perceive' - that is, how the world appears to us - does not come solely from our senses. What we perceive is a combination of what we sense and of our brain's memory-derived predictions' (p87) Note once again, perception depends on behaviour here - with no past behaviour there can be no memory-derived prediction - hence behaviour can never be optional.

I think Hawkins is in love with the notion of prediction, since we get this on p89 : 'Prediction is not just one of the things the brain does. It is the primary function of the neocortex, and the foundation of intelligence...... Even behaviour is best understood as a by-product of prediction'. My, behaviour really does get a hammering ! Prediction leads to probabilities - the brain estimates the maximum likelihood it seems. This isn't surprising since the decoding of signals from noise using a Viterbi decoder runs along the same lines and doesn't fit very well into a algorithm. Much better done with a neural net...

Skipping along a bit, we find out where Hawkins believes prediction precedes behaviour (p102) 'Move your arm in front of your face....the cortex predicts seeing the arm, and this prediction is what causes the motor commands to make the prediction come true. You think first, which causes you to act to make your thoughts come true.' Of course, he can't mean we think consciously, with language, he must mean an unconscious thought or intention. However, this seems to be too granular a way at looking at movement, a kind of cause - effect, rather than a locus or trajectory kind of model.