This is a review on Chapter 3, in Ray Kurzweil’s “How To Create A Mind”: The task of reverse engineering the brain might be one of the biggest tasks in human civilization, and is one that Kurzweil casually approaches in his new book “How to Create a Mind: The Secrets of Human Thought Revealed”. Kurzweil’s work has undoubtedly been instrumental to the field of AI, and as a pioneer in speech recognition technologies he is the right man to consult on the topic.
That the brain understands patterns through a hierarchal grid system is nothing new, and humans have been under this assumption since the late 1950’s. We have been able to create pattern recognizers able to perform intelligent tasks to a certain extent, IBM’s Watson clearly being the best performer (although it’s parameters where very closed and focused) and Google only last year tried to create an algorithm using the same logic, but only reached an accuracy rate of about 18%. Yann LeCun, the N.Y.U. researcher who was just appointed to run Facebook’s new A.I. lab mentioned the that: “AI [has] ‘died’ about four times in five decades because of hype: people made wild claims (often to impress potential investors or funding agencies) and could not deliver. Backlash ensued. It happened twice with neural nets already: once in the late 60’s and again in the mid-90’s.” The point is, that the brain learns by chunks of low level to high level patterns is not the interesting part – the interesting and still unknown part is how?
That we are able to replicate human thought with AI using this logic is yet to be proven, which leads to the main criticism I have – why did he not test his theory like any good scientist would do? He seems to make a lot of claims about how this theory would solve everything with very little evidence. Moreover, he claims to reveal the secret of the human brain, thought, and emotion without fully making the connections. He is clearly merely interested in simply replicating certain “thought processes” mathematically.
Much of his logic I would agree to, for example that a computer is much more able to process the 300 million pattern recognizers a human brain presumably has (how he came to this number is unclear). Also the idea that there are directed and undirected thought processes, which explain our state of dreaming or mediation vs. goal oriented thinking makes much sense and pertains to the variables we are currently working with in this class – attention/meditation with various BCI technologies are to date the most reliable. The way in which information flows down the conceptual hierarchy as well as up thus enabling us to predict the future is also very plausible, however how we are able to do this or remember new patterns is unclear.
Over all, the PRTM indeed gives the unfamiliar some thoughts on how we would begin computing a human brain. However nothing new is revealed or tested, as he even falls short in proving the differences between his and Hawkins theory. Actually, many of his unsupported ideas may in fact merely be … dreamt up.