height=”192″ width=”256″ alt=”Blue Scenic :image” />
I wrote the other day in the Beginning of Wisdom piece about a neural network system that was trained to discriminate US from Soviet Tanks in the cold war. Only it didn’t - it made its discrimination on the basis of lighting conditions - the US tanks had been photographed in the morning, the Soviet at night!
We found the same kind of thing with my son James, aged 2 1/2, the other day. We’ve just got a new car, a Renault Scenic. This has caused great excitement, with my daughter (Emma, aged 7 1/2) showing it off and counting all Renaults she sees. This has sparked interest in James who is going all out to spot all Scenics he sees too, but his discrimination is slightly wide of the mark like the neural network is - he claims all cars of the same shade of blue as our Scenic are Scenics, and all others, including non-blue Scenics, are not. Ergo, all cars are red - including non-blue Scenics - and Scenics are blue - including any blue cars!
This discrimination thing, It’s the way that we learn, example, example, counter-example etc. Its not about neural networks specifically. I remember a book called ‘Induction’ by John Holland in the 80s about all this stuff and how it relates to how we problem solve. We don’t get told all the rules, we get told some, and we infer the others. And often, in a blink, we don’t know why we think something, but we do. Sounds dumb, but it’s actually what make us smart.
RSS Feed