TEDxWarwick - Professor Steve Furber - 2/28/09

[gespilk]

I have heard this talk from many people and many times in every decade.

This is not the way to model a human brain. It may give us some better chips but it will never give us working Artificial Intelligence.

One promising way is to use CYC-like technology where the "facts" are sensory inputs. However current CYC is limited because its higher levels still have to done manually.

Another way is "the machine with dreams" meaning abilities for confabulation, association, analogy, + all reasoning.

[CommandLineCowboy]

@gespilk The question is not will this create AI but is the Spinnaker system useful. Neuroscientists today use neural net systems that can understand speech. They then 'break' the neural net in ways that simulate known pathologies of brain disease and in this way come to a greater understanding of the causes. Spinnaker will be a much larger neural net, be able to recognise more speech be able to simulate subtler pathologies. I thinks that's useful.