Tuesday, January 13, 2015

Q&A - 13/1


In May 2014, Baidu, the Chinese search giant, has hired Andrew Ng, a leading Machine Learning and Deep Learning expert (and co-founder of Coursera) to head their new AI Lab in Silicon Valley, setting up an AI & Deep Learning race with Google (which hired Geoff Hinton) and Facebook (which hired Yann LeCun to head Facebook AI Lab [who is a student of Hinton]).

The race is on

Ernest Davis (following a link posted by Yann LeCun)

[Criticizing the book Superintelligence by Bostrom] assumption [is] that intelligence is a potentially infinite quantity with a well-defined, one-dimensional value. Bostrom writes differential equations for intelligence, and characterizes their solutions. Certainly, if you asked Bostrom about this, he would say that this is a simplifying assumption made for the sake of making the analysis concrete. The problem is, that if you look at the argument carefully, it depends rather strongly on this idealization, and if you loosen the idealization, important parts of the argument become significantly weaker, such as Bostrom’s expectation that the progress from human intelligence to superhuman intelligence will occur quickly.

Of course, there are quantities associated with intelligence that do correspond to this description: The speed of processing, the size of the brain, the size of memory of various kinds. But we do not know the relation of these to intelligence in a qualitative sense. We do not know the relation in brain size to intelligence across animals, because we have no useful measure or even definition of intelligence across animals. And these quantities certainly do not seem to be particularly related to differences in intelligence between people. Bostrom, quoting Eliezer Yudkowsky, points out that the difference between Einstein and the village idiot is tiny as compared to the difference between man and mouse; which is true and important. But that in itself does not justify his conclusion that in the development of AI’s it will take much longer to get from mouse to man than from average man to Einstein. For one thing, we know less about those cognitive processes that made Einstein exceptional, than about the cognitive processes that are common to all people, because they are much rarer. Bostrom claims that once you have a machine with the intelligence of a man, you can get a superintelligence just by making the thing faster and bigger. However, all that running faster does is to save you time. If you have two machines A and B and B runs ten times as fast as A, then A can do anything that B can do if you’re willing to wait ten times as long.

Thumbs up