On Intelligence By Jeff Hawkins
Few people out there are as smart as Jeff Hawkins. So it is quite fitting that he is interested in researching the human mind. The objectives of this research are unique though. His objective is to learn the workings of the brain and apply this knowledge into creating better computers. Being a computer engineer myself, I find such work fascinating.
So why do I think Jeff Hawkins is wicked smart? Well just take a look at his background. He is the father of handheld computing; founded Palm Computing, where he invented the PalmPilot and Handspring, where he invented the Treo. Ofcourse after making a few hundred million dollars, he can afford to research brains on a full-time basis.
The book "On Intelligence" by Jeff Hawkins is a good read. My knowledge of biology is pretty sketchy for most parts and of neurobiology was as good as non-existent. So just learning the basics of brain physiology and anatomy was educational for most parts. The book also talks about the failures of the field of Artificial Intelligence and why the approach is wrong. Neural Networks, he says kind of started in the right direction but suffer from drawbacks such as not giving enough importance to feedback connections in the networks.
The neo-cortex is the top layer of the brain that developed most recently on the evolutionary scale. It consists of 5 layers with the large-scale analysis being done in the topmost layer and fine-grain analysis in the lowermost. This is where the signals from all our senses, get sensed. The basic function of this region is basically to store patterns and then matching new patterns to the stored patterns to identify or extrapolate from them. Based on his research of the neo-cortex, Jeff has founded another company called Numenta, that works on pattern recognition software and hierarchical memories.
I've been trying to find out university research groups that do similar research when I landed upon some excellent work by Kwabena Boahen at Stanford. His work was even featured in the Scientific American last year. The objective of his group being to produce silicon versions of parts of the brain, albeit very rough approximations. The power efficiency and coding efficiency of the human mind is almost impossible to mimic in today's technology.
5 comments:
Reminds me of the term wetware. Used quite often in the science fiction that I read.
As for NNs failing, yeah, what happened in AI was overpromising. All these AI researchers promised talking robot dogs in a few years, but no one quite anticipated just how complex the brain was.
At work, I have to write programs which find patterns in images. It's frustrating when a customer will look at an image and say, "That's easy!", when you know just how hard it is to write a program to do it!
Cool! Must check it out. I think one of the biggest hurdles to AI is that we haven't yet unraveled the role that emotiona and gut instinct play in what we call intelligence. And that's one hurdle of course.
You may want to google for Hubert Dreyfus, the big bad wolf of the pro AI lobby.
@Hob - LOL at the "That's Easy!" comment. :) Yeah and to think we got soo many hit AI movies. Arnold Schwarznegger wouldn't be governor if AI didn't show promize early on.
@Ashutosh - Yeah Jeff Hawkins work is aimed at just the neocortex to keep things simple. Emotions come from the older parts of the brain and he says are much harder to analyze. True there could be some connections there which even he might be missing. Good point.
hey saket !
Great blog! Keep in touch.
shrinivas
Thanks Shrini.
Saket
Post a Comment