On Jeff Hawkins

As there are so many similarities between Jeff Hawkins’s book ‘On Intelligence’ (2005) and my recent ‘Intelligence and the Brain’ talk, it might be natural to assume that I had copied him without crediting him. But I have only just discovered Jeff Hawkins, afterwards.

blah
On Intelligence

I think it is interesting that both approaches can be so similar when, as well as being written independently, there are no common references.  I cite ideas from a whole host of thinkers e.g. Rodney Brooks, Daniel Dennett, Karl Popper, Richard Gregory and Karl Friston. Hawkins only really cited Vernon Mountcastle.

In this posting, I just want to highlight the similarities by quoting Hawkins and pointing to recent talks and postings which have covered almost identical ground. Other postings on his book will hopefully follow later. For a review of his book, I recommend Sunny Bains’s.

An overarching theory of the brain

In ‘Intelligence and the Brain’, I presented Friston’s ‘variational free energy’ theory which has been hailed as a universal theory of the brain. Hawkins presents a similar vision, albeit somewhat vaguer.

 “One of the favourite technologies of the 1990s, aka the Decade of the Brain, was functional imaging… The result is that we have lots of data on where in the brain certain tasks occur, but little data on how realistic, time-varying inputs flow through the brain.” (page 31)

“…the field of neuroscience itself was awash with details. It still is… There’s still no overall theory…  History shows us that the best solutions to scientific problems are simple and elegant…. The failing isn’t one of not having enough data” (pages 33-34)

“The situation is analogous to the work of biologists in the 1800s. They spent their time discovering the minute differences between species. For many years Darwin followed the same course … but Darwin eventually had the big insight to ask how all these species could be so similar” (page 50)

“What has been lacking is putting these disparate bits and pieces into a coherent theoretical framework … and it is the goal of this book.” (page 90)

Prediction, hierarchical representation and time

Prediction, hierarchical representation and time are three key concepts for Hawkins:

“I have seen how the ideas of prediction, hierarchical representation and time have crept into the language of neuroscience.” (page 233)

Prediction and hierarchical representation are key concepts for Friston and time was a key concept in ‘Free Will / Free Won’t’.

The regular architecture of the cortex

Hawkins picks up on Vernon Mountcastle’s work on the uniformity of the neocortex. (And incidentally, likens the cerebrum to a cauliflower.)

 “…the outer surface of the brain seems highly uniform. A pinkish grey, it resembles a smooth cauliflower” (page 40)

“Vernon Mountcastle… published a paper titled ‘An Organizing Principle for Cerebral Function.’… The regions of cortex that handle auditory input look like the regions that handle touch, which look like the regions that control muscles, which look like Broca’s language area, which look like practically every other region of the cortex. Mountcastle suggests that since these regions all look the same, perhaps they are actually performing the same basic operation!” (page 50

“The thing that makes the visual area visual and the motor area motoric is how the regions of the cortex are connected to each other and to other parts of the central nervous system.”

“When I first read Mountcastle’s paper I nearly fell out of my chair… In my opinion it was, is, and will likely remain the most important discovery in neuroscience.”

Note the uniformity of behaviour is a basic assumption for Friston.

Prediction

For Hawkins, like Friston, prediction is key:

“I argue that an important function of the neocortex is to use its memory to make predictions.” (page 82)

“I had an ‘aha’ insight… what would happen if a new object… appeared in the room… When I look around the room, my brain is making predictions about what it expects to experience before I experience it. The vast majority of predictions occur outside of awareness.” (page 86)

“Prediction is so pervasive that what we ‘perceive’… 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.” (page 87)

“Prediction is not just one of the things your brain does. It is the primary function of the neocortex, and the foundation of intelligence. The cortex is an organ of prediction.” (page 89)

“IQ tests are based on making predictions… Science is itself an exercise in prediction.” (page 96)

“Prediction requires a comparison between what is happening and what you expect to happen. What is actually happening flows up, and what you expect to happen flows down.” (page 113)

“All predictions are learned by experience… You were not born with this knowledge; you learned it.” (page 119)

Hierarchy  with feedback

Hierarchy, with communication between levels in both directions, was emphasized in ‘Intelligence and the Brain’ and also earlier in ‘Free Will / Free Won’t’.

“The notion of a hierarchy is critical.” (page 44)

“I want to draw another picture of the cortex that highlights its hierarchical connectivity… sensory input enters at the bottom, the lowest region, and flows upwards from region to region. Notice information flows both ways.” (page 109)

“…there are as many, if not more feedback connections in visual cortex as there are feedforward connections.” (page 113)

“…information goes up and down your cortical hierarchy … may remind you of the hierarchy of military command… The general does not get all the details. There is an exception to this rule. If something goes wrong that cannot be handled by subroutines down the chain of command… What was the unanticipated problem to subordinates is just the expected next task on his list.” (pages 132-133)

 

The role of time

The idea that lower levels of the hierarchy are faster was covered in ‘Free Will / Free Won’t’.

 “If the robot waits to start moving until it knows exactly where the ball will arrive, it will be too late to catch it… Your brain has a stored memory of the muscle commands required to catch a ball (along with many other learned behaviours)” (pages 68-69)

Graduation from brain to environment

The inability the make a clear separation between whatever we might count as ‘self’ and the outside world was first considered in ‘The Extension of Mind’. For Hawkins:

“Your brain can’t directly know where your body ends and the world begins.” (page 60)

“To the cortex, our bodies are just part of the external world… From the brain’s perspective, it doesn’t know about your body any differently than it knows about the rest of the world. (page 199)

Artificial Intelligence is imitation; prediction is real intelligence

In ‘Redefining Turing’, I claim that the Turing Test is about proving that non-human things can be intelligent, not a definition of what is intelligence. Hawkins also rejects the idea that the test is a definition of intelligence, but in a different way:

“As inspired by Turing, intelligence equals behaviour… Behaviour is a manifestation of intelligence, but not the central characteristic or primary definition of being intelligent.” (page 29)

“The human mind is created not only by the neocortex but also by the emotional system of the old brain and by the complexity of the human body. To converse like a human on all matters (to pass the Turing test) would require an intelligent machine to have most of the experiences and emotions of a real human, and to live a humanlike life… it will not have a mind that is remotely humanlike unless we imbue it with humanlike emotional systems and humanlike  experiences. That would be extremely difficult and, it seems to me, quite pointless.” (page 208)

“Borrowing from nature, we should build machines along the same lines. Start with a set of senses… Next, attach to these senses a hierarchical memory system that works on the same principles as the cortex… our intelligent machines will build a model of its world seen through its senses. What makes it intelligent is that it can understand and interact with its world via a hierarchical memory model… Intelligence is measured by the predictive ability of a hierarchical memory, not by humanlike behaviour.” (pages 209-210)

“If Searle’s Chinese Room contained a similar memory system that could make predictions about what Chinese characters would appear next and what would happen next in the story, we could say with confidence that the room understood Chinese and understood the story. We can now see where Turing went wrong. Prediction, not behaviour, is the proof of intelligence.” (page 105)

“Behaviour and prediction are two sides of the same thing.” (page 101)

We need to build machines based on real, biological brains to really understand

Hawkins on AI computer scientists:

“Some took pride in ignoring neurobiology. This struck me as precisely the wrong way to tackle the problem. (page 12)

“The ultimate defensive argument of AI is that computers could, in theory, simulate the entire brain. … But AI researchers don’t simulate brains, and their programs are not intelligent. You can’t simulate a brain without first understanding what it does.” (page 21)

“AI researchers ask, ‘Why should engineers be bound by the solutions evolution happened to stumble upon? … How did we succeed in building flying machines? By imitating the flapping action of winged animals? No.” (pages 37-38)

Advertisements
This entry was posted in Uncategorized and tagged , , , . Bookmark the permalink.

3 Responses to On Jeff Hawkins

  1. Pingback: On Hawkins, Hierarchy, Homunculi and the Hippocampus | Headbirths

  2. Bell says:

    “If Searle’s Chinese Room contained a similar memory system that could make predictions about what Chinese characters would appear next and what would happen next in the story, we could say with confidence that the room understood Chinese and understood the story. We can now see where Turing went wrong.”

    Actually, this is where Hawkins went wrong — in fact, where he got it entirely backwards. The point is that if the Chinese Room could provide convincing answers to sufficiently complex questions, it must be utilizing a method that fulfills the requirements of understanding (which may be the requirements Hawkins describes here, if he’s right, or other requirements, if he’s wrong). Like Searle, Hawkins appears to have assumed a naive and simplistic implementation and then concluded “but that naive and simplistic implementation can’t be understanding anything!” Yes, of course not — but in that case it also couldn’t possibly have passed the Turing test. Turing had it exactly right: the ability to answer sufficiently complex questions is evidence that the system producing them is, in fact, “intelligent” by any reasonable definition.

    It was Hawkins’ failure to see the deep and glaringly obvious flaws in Searle’s Chinese Room argument that led me to me put down his book.

    • headbirths says:

      To say (as Hawkins does) that Turing was wrong is irritating. The ‘Turing Test’ was named after Turing and not by him.
      But my understanding is that Hawkins is not saying that behaviour is not a hallmark of intelligence. He would accept something that passed the Turing Test as being intelligent. But the Turing Test is setting the bar too high.
      In an upcoming post, I will be contrasting behaviourism with physicalism. I will (here in this comment) use the definition of what a metal is as an example. A ‘behaviourist’ definition would be one based on external appearances. For example, it might be something like ‘hard, opaque shiny metal that is a good conductor’. A ‘physicalist’ explanation proposes what is going on on the inside. For example, a metal’s atoms sit within a ‘sea’ of electrons that have been lost the outer shell of those atoms. A behaviourist explanation is often a useful stepping stone towards a better (physicalist) explanation (which then opens up the possibility of discovering new behaviour).
      I think Hawkins would agree with that relationship. As he says: “behaviour and prediction are two sides of the same thing” (page 101). A definition of intelligence in terms of what is being done (inside) to make predictions will admit some things that do not pass the Turing Test but are still intelligent.
      BTW I had a different reason for wanting to put Hawkin’s book down – the ‘I am so wonderful’ theme than runs through it. But I’m glad I persevered.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s