Joined: Aug. 2009
|Quote (Henry J @ Aug. 11 2010,19:29)|
|To figure out if language can be learned incrementally, just build a computer with a neural net, some sensors and manipulators, and an output device, and see if the thing can be taught a language.|
Most of the common neural nets don't actually learn, they get trained or configured using various techniques (like evolution) there are some types of network that include rules to model learning mechanisms - like adjusting synaptic connections based on firing rate - but there is still a bit of a gap between what the nets do and what biology does. (although I'm a few years out of date in my neural net knowledge)
There is also the question of archetecture. the brain is not a uniform structure, it has a lot of variety from the start and some of this may affect how and what it is able to learn. Even if we have a neural net capable of learning it still might not learn some things unless its archetecture is right.
An then there is the problem of power, to get close to a rat brain you need millions of neurons - thats a lot of processing power!