A Thousand Brains: A New Theory of Intelligence - Jeff Hawkins

[nafregnum]

A little book report, for anybody who likes to geek out about contributions of neuroscience to the development of AI.

 

Author is Jeff Hawkins, one of the founders of Palm Computing.  In 1986 he was very interested in studying neuroscience and AI for speech recognition, but back then his ambitions were way too big for the PhD board in the neuroscience department of [Cornell or Berkeley, I don't remember which, maybe both]

 

https://en.wikipedia.org/wiki/Jeff_Hawkins

 

He tells a story of how he gave a presentation to Intel leadership that predicted everyone will one day be using handheld computers and getting a skeptical reception.

 

After Palm Computing, he founded a neuroscience lab called Numenta.  One day, holding a coffee cup, he asked himself a few thought experiment type questions about what his fingers would expect if he adjusted his grip on the cup and what would be going on in his brain as he did so.  He had a eureka moment about how it all fits together, and called it the Thousand Brains theory of intelligence.  It has to do with "cortical columns" which are like small 1mm x 1mm squares of lasagna layers spread out all over the neocortex.  One cortical column will be tied to a little patch of nerves in a fingertip, another tied to a small patch of input from the eye, etc.  In each cortical column, the top layer is made of neurons that receive the direct sensory inputs and they fire almost nonstop with the "blooming buzzing confusion" of constant inputs from the outside world.  Traveling down the lasagna layers in each column, the neurons are firing less and less often because each layer is "refining" its prediction about what inputs will come next from above based on which "reference frame" the brain has chosen as the best prediction for what the current context looks like.  To borrow a metaphor he doesn't cite in the book, think of that poem about the bunch of blind men who encountered an elephant, except imagine that they each share what they feel with each other and then vote on what it is that they're all encountering.  A voting process among the deeper layers of the cortical columns help us determine where we are at the present moment and activate a reference frame kind of like a learned map of the territory.

 

He has a pretty decent chapter which patiently explains why "uploading your brain to a computer" is a fun sci-fi concept but is pretty much problematic even if it one day became possible to do atom-by-atom scanning of a person's brain.  One example: You would have phantom limb syndrome for every part of your body all at once unless you virtualized the entirety of your nervous system.  

 

He explains how all the current work in AI hasn't delivered any actual "intelligence" yet, just machine learning.  He believes we need to be using everything we're learning about our neocortex brain structures to make better advances toward a real kind of intelligence, perhaps by aiming to produce a machine as smart as a kindergartner at first.  He explains how we have an "old brain" from the 100 million+ years of evolution when the name of the game was just staying alive and reproducing, and that we do not need or want to reproduce the old brain in machine intelligence, so the future intelligent machines wouldn't develop fear of dying when being turned off.  He believes we humans with our old brains are much more likely to cause our own extinction than to create a Terminator type future (overpopulation, nuclear annihilation, making earth uninhabitable among the threats).  Intelligent machines would be more capable than humans at establishing a hypothetical 2nd home for humans on Mars because they could deal with the radiation and lack of atmosphere while they do their jobs to build things or dig tunnels or whatever the job requires.

 

The book steers into some fun sci-fi musings in the last 3rd.  One chapter near the end is titled "Estate Planning for Humanity" ... he suggests that in the billions of years of time in our universe, intelligence like ours may spring up from time to time and only last a brief moment before the candle flame goes out again.  So as we look for signs of life out there, the odds may be against us that a similar kind of intelligent life would be "awake" at the same time as us, so he suggests we should develop some kind of long lasting beacon in our solar system which would be capable of signaling to distant observers that intelligent life was once here.  Think: some kind of window blinds orbiting the sun which block/diminish its light in a regular intelligence-signal pattern -- something which could function for millions of years without maintenance.  Fun sci-fi thoughts if nothing else.

 

If you've read it before, or do so in the future, I'd love to know what you found interesting!

[pricingpower]

I enjoyed this book, pretty inspiring to see someone push science forward like this for the rest of humanity and to be reminded that bubble valuation times can create some nice positive externalities for society.