Jeff Hawkins has a new theory of the brain. It is interesting and persuasive. But to me, the most interesting observation in the book is not about brains – it’s about how an intelligent species could make its presence known across the vast distances of space. It seems that was also the most interesting part of the book for Richard Dawkins, who contributed a foreword. More about that later: first the brain stuff.
The old brain and the new
Hawkins has been wondering how brains work for a long time, ever since he read a 1979 essay by Francis Crick (the DNA pioneer) about how mysterious the brain remains. Frustrated by the field’s slow progress, he took a detour away from neuroscience, and founded Palm Computing, which developed some very successful early smartphones, including the PalmPilot and the Treo. He left Palm in 2005 to establish Numenta Labs, a brain research institution in California.
Our brains have two parts. The old, reptilian brain, or hind brain, is heterogenous, composed of several differentiated parts, like the medulla, the pons, the cerebellum, and so on. The neocortex, which exists in all mammals and only mammals, is where our advanced intelligence arises, but it is structurally homogeneous: any section of it looks very much the same as any other section. Hawkins was introduced to this fact by Vernon Mountcastle, an American neuroscientist, and it has inspired much of his work.
The neocortex is dense. A hundred thousand neurons are crammed into the space of a grain of rice, connected to each other at half a billion synapses, via several kilometres of cabling – axons and dendrites. There are many different activities going on inside different parts of the neocortex, and what differentiates them is not their structure, but their connections. When we learn something, these connections are strengthened. When we forget something, they are weakened.
Cortical columns
The basic units of the neocortex are cortical columns. Each column contains upwards of a hundred neurons, and there are between one and two million of them in the neocortex. The key insight in Hawkins’ theory is that what cortical columns do is to attach reference frames to objects in the world – and also to abstract concepts. There are “what” columns that attach frames to external objects, and “where” columns that attach frames to your body. This is what enables your brain to understand where it is in the world, and to navigate it.
Collectively, the neocortex “learns a model” of the world, and continuously updates it. Hawkins says that there is no central control room in our brains. Instead, our perception is a consensus which the columns reach by voting. Within the columns – indeed, within the neurons – predictions are made, and depending how successful their predictions are, the neurons will vote for their version of events. The model which emerges is the result of their aggregate strength of those predictions.
Machine intelligence
I’m not qualified to say how much of this theory is original, and how much it owes to, say, Marvin Minsky’s 1986 book, “Society of Mind”. It is certainly interesting and persuasive. But Hawkins is far from finished. Slightly more than half of his book is not actually about how the brain works, but about whether human-level artificial intelligence will be developed, and whether it is a threat, and then about how humanity should address posterity.
He has a surprisingly anthropomorphic view of intelligence. Several of his remarks suggest that he thinks the only way an organism could be intelligent is by doing what our brains do – in the way they do it. “I don’t believe any kind of deep learning network will achieve the goal of AGI if the network doesn’t model the world the way a brain does.”
He is also surprisingly complacent about the threat from advanced AI. He thinks that artificial general intelligences (AGIs – artificial beings with human-level intelligence) will arrive in the next two or three decades, and that they might well be conscious. But he thinks they will be affectless – uninterested in their fate. “I would have no concerns about unplugging a conscious machine.
First, consider that we humans turn off every night when we go to sleep … Unless we go out of our way to give machines equivalent fears and emotions, they will not care at all if they are shut down, disassembled, or scrapped.” He acknowledges that many smart people disagree, and think that machines with goals could well develop goals which are inconsistent with ours.
His claim that “intelligent machines will not have human-like emotions and drives unless we purposely put them there” is not wholly reassuring. As Professor Stuart Russell and others have pointed out, an intelligent machine with no goals would be rather pointless, and a machine with any goals at all will derive other subordinate goals, notably the intention of continuing to exist in order to accomplish the main goal – and to acquire resources that will help it to achieve the main goal.
Posterity
The last third of the book is a digression further away from research into brains and intelligence, but it is a digression which will probably engage most people who are interested in how brains work. It is how we should respond to the knowledge that we are likely to go extinct at some point in the future.
Hawkins has no truck with hopes for extending our lives with uploading or similar tricks, and he worries that we will disappear, leaving no collective footprints in the universal sand. He takes the extreme view that “being lost forever is the same as never existing.” So if we had not come along to dig up and admire their bones, the dinosaurs’ 160 million years of planetary dominance would not only have been meaningless – it would never have happened.
This belief makes Hawkins a fan of METI, which stands for “messaging extra-terrestrial intelligence”. Many people (me included) think that sending a message which alerts aliens to our existence and location is a thoroughly bad idea, because we cannot predict what they might come and do to us. Hawkins shows the same complacency about the potential threat from aliens as he does about the potential threat from AGI.
No alien signals means no aliens?
Hawkins thinks we will probably go extinct, and that if aliens never learn about us, then our existence will have been meaningless. He therefore suggests that we build an archive of our civilisation in satellites orbiting the sun, and then he proposes an intriguing idea about how to advertise it. I don’t know if it’s original, but I haven’t heard it before, and neither had Richard Dawkins, it seems.
We could, he suggests, place enough other material into orbit around the sun to occlude its light in specific patterns – patterns which could not have occurred naturally. This would be far more visible, and far longer lasting, than sending out Voyager probes, radio signals, or artificial beams of light. Hawkins is confident that the universe contains a lot of intelligent life, so our message will be received. He does not appear to ask himself why we have seen no evidence of any extra-terrestrial civilisation doing the same thing. It might be that we haven’t had powerful enough telescopes yet, and perhaps the new ones coming online this decade will jolt us with some starling discoveries.
Or perhaps no ETs have hung up any such sign. Because, perhaps, they are not out there.
