I’ve been listening to a fascinating book (recommended by TWiT and purchased from Audible.com) written by Jeff Hawkins called “On Intelligence
The gist of the book is Hawkins’s theory that the function of the neo-cortex is to form predictive models of the world based on past experience, communicated to it as inputs through the senses. He points out that each sense is processed similarly to one another even though they seem to be radically different to the untrained eye. An example Hawkins uses that caught my attention was of the transition between typing, speaking, and hand-writing. Hawkins points out that to communicate the Gettysburg Address one need not consciously learn it for a new medium. Our brain learns the famous speech by hearing it or reading it and our brain can rehash it by totally different means. The example is rather simple but amazing nonetheless.
The neo-cortex is called the neo-cortex because it is the “new” brain when compared to a lesser species that did not evolve it. Our “lower” brain is also called the reptilian brain, since that’s the sort of brain we have in common with Gojira and his friends (not Mothra of course, that’s another blog post). What I found fascinating about the descriptions of the neo-cortex was its uncanny ability to discern and absorb information. Watching a baby in a new room gives one a sense of this. As the baby girl looks around you can visualize all of the identification and cataloging that is going on inside her head. The cortex functions by translating any possible input into what it already knows. If the cortex does not have a pre-existing model to translate to, it files the new input away as a potential model.
Areas of the cortex work on bits of input to determine if they are suitable towards the purpose of that particular identifying bit. An oversimplified example would be a tiny part of the brain that looked as some input (say, something that was translated from the optic nerve and further broken down into vague shapes) and determine if the input contained an eye. If the input did contain an eye, further processing could then take place and our brain would communicate back and forth to tell the optic nerve to expect a nose in this general direction.
According to Hawkins this is how we work. Given this, Hawkins expects electronic brains to be feasible in ways that traditional artificial intelligence will never be. Some day Google may actually fully understand what the following question means: “is a dandelion more yellow than a canary?” Based on a lot of input (and correctly labeling such input) software should be able to learn about qualities in the way humans do. Whether or not this is a good thing and will actually be truly useful is to be seen (hopefully in my lifetime).
I think that one could easily create a sociopath computer just by taking advantage of how it learns. I joked with my wife one time about purposefully teaching the incorrect words for various colors to the children of some friends of ours. “What color is the grass? That’s right, it’s pink.” Giving a learning machine similar false positives could really screw up its view of the world. I wonder if such a thing could cause a machine to feel hate, or would emotion truly be more complex than a brain mechanism (i.e. necessarily involving hormones).
Based on my listen I also wonder about whether or not organized society is simply an evolutionary step. If our neo-cortex is all about taking in input, cataloging it, and making predictions based upon that stored knowledge, perhaps society rose about as a way to distribute such knowledge? What would the implication be if we determined that the entire function of families, cities, and nations was simply to propagate knowledge? If this would be the truth, in what direction are we evolving?
One final thought I wanted to share from this book was one on DNA and species-memory. Hawkins describes a scenario that explains the course of evolution by way of DNA being the most primitive form of brain: certain aspects necessary for survival are encoded into our DNA and this is how further generations know certain things. I don’t really know if I buy into this (I’d love to see more evidence and experiments for it) but the notion is intriguing. If this is true, the notion of clones becomes interesting to me, if nothing else but for the purpose of seeing how different they become over time. Not in a Multiplicity way of course but over generations of “self-duplication.” Would clones evolve over generations? Perhaps a series of clones would become impotent based on the new way of propagating genes? Okay, now I’m beginning to sound like some cheap sci-fi.
All in all, this work has been wonderful to listen to. Though it’s taken me weeks to get through, “On Intelligence