Last week I discussed the progress that Wolfram|Alpha, described by its publisher, Wolfram Research, as a “computational knowledge engine” has made in the 12 months since I first discussed it. As I concluded last week, “[the] progress of Wolfram|Alpha is spectacular in such a short period]”.
To briefly recap, Wolfram|Alpha is built as a collection of customized knowledge domain and sub-domain specific modules on top of Wolfram Research’s Mathematica engine. What’s intriguing is that this architecture has an interesting parallel with the architecture of the human brain in so far as human brains have many areas dedicated to recognizing and responding to very specific stimuli.
For example, there are high level areas in the brain that handle functions such as vision, hearing, and balance while at a very low level it has been found there are individual neurons that become dedicated to responding to very specific stimuli, such as the amusingly named Marilyn Monroe and Halle Berry neurons.
So, my thought was that Wolfram|Alpha, with its vast store of knowledge and many tens of thousands of modules that recognize symbols (stimuli) that they specifically and idiosyncratically respond to, could eventually have enough embedded knowledge to provide a useful form of artificial intelligence, capable of, for example, answering complex questions with real “insight” and understanding.
Of course, the sheer quantity of specialized modules required would probably have to number in the millions to rival the complexity of response of the human brain. Now, before you argue that millions of modules wouldn’t be anywhere near the scale of neuronal connections in the human brain (estimated at around 100 billion) remember that much of the underlying mechanics of the biological processing system is paralleled by the coding that implements the modules. Thus what matters is the richness of knowledge, not the richness of connections.
So, how would the AI version of Wolfram|Alpha (Wolfram|Omega?) present itself? Obviously, this is a matter of pure speculation but one thing it would most likely not appear to be would be a human-like presence. In our discussion for last week’s newsletter, Stephen Wolfram, the father of Wolfram|Alpha, explained that already the system would fail the Turing Test simply because it knows too much.
Imagine Wolfram|Alpha with image handling capabilities and a speech interface backed up by robust linguistics. Perhaps integration with Numenta’s Hierarchical Temporal Memory (HTM) technology would provide the link to the real world (I reviewed a product, Vitamin D, which is based on HTM in my Network World Gearhead column).
So, let me ask you: what do you think an artificially intelligent system would be like? What would it, ideally, be capable of? Would you, or could you, ever trust such a system to do things in the real world? Let me know your thoughts …
