11 April 2008
Jeff Hawkins’ HTM: Hype And Hot Air
It has been just over a year since Jeff Hawkins’ Numenta released the platform implementing the Hierarchical Temporal Model (HTM) that Hawkins was so sure would allow computers to perform human-like thought. Well one-year-old humans can walk, distinguish between their parents and strangers, demonstrate fear, and a whole ton of other stuff that HTM doesn’t appear to have achieved yet. Perhaps this has something to do with the fact that despite Hawkins’ attempts to convince everyone that HTM was some kind of magical new AI, it is really only an implementation of the adaptive resonance models which have been around for over a decade (and which were introduced with appropriate humility and without unnecessary fanfare).
The whole situation is altogether unsurprising since, according to Hawkins, he came up with the idea when reading Vernon Mountcastle’s 1978 work on organizing principles of human cortex, a work which he decided (based on his extensive knowledge of what people read) that pretty much everyone else had entirely ignored. Right, because no one pays much attention to stuff written by members of the National Academy of Sciences, and certainly not to the stuff written by people who win the National Medal of Science.
Well, there you have it. And I didn’t even say Foleo, once!
2 Comments
Phillip Shoemaker
15 April 2008
What does the foleo even have to do with the neocortex? Are you implying that because Jeff made a single mistake (although it is yet to be seen if the Foleo is actually a mistake), that all of his ideas are incorrect? You of all people should know better than that.
With regards to the timeline (a year since Numenta released their HTM implementation), I should point out that the first version has yet to be released. The one that released back in March 2007 was an alpha version. And despite being the early version it is, there are numerous problems currently being solved with it.
Suggesting that HTM is another model of adaptive resonance models is laughable; according to many AR folks, AR has solved most of the world problems more than once, and is nearly the perfect model of the brain.
And, regarding mountcastle. Jeff’s argument is not that he can read minds, but that if nobody talks about a book after they’ve read it, they tend not to believe it. Hardly ever was mountcastle’s theories quoted in other works.
Open your mind.
TerminalDigit
15 April 2008
Relax Phillip. The Foleo bit was a joke. Unclench your intestines.
And while you’re manipulating body parts, try opening your own mind. You really think you’re looking at things objectively? You work for Numenta! How could you continue to perform your job in good faith if you believed Jeff’s ideas to be fallible? Your paycheck depends on blind belief, but please keep your PR campaign on your own site.
As far as adaptive resonance, you can laugh all you want, but I notice that you never offer that HTM is not an implementation of adaptive resonance. Whenever people have been able to corner Jeff with this directly, neither has he.
And my response to your comment about Mountcastle can be found here, beneath the other damage-control post you made here.