
Originally Posted by
epimetheus
While popular amongst neuroscientists, and the most likely, the idea that the mind is entirely computational is still debated. There are theories such as the Penrose-Hameroff theory that model parts of the brain as a superposition, and hence non-computational.
Current AI architectures are not nearly as complex as the brain. In terms of raw weights the biggest networks are close to the number of neuronal connections - but raw numbers don't mean much. Counter-intuitively too many connections is probably the problem, hence the recent uptake of sparse models. We see this in babies - far denser networks than adults but can't even control their bowels, only when those connections are pruned do they manifest more sophisticated behaviours.
The question should be what is the human brain doing. There is growing evidence to suggest that it is 'running equations'; it is the prevalent view of neuroscientists. If so, then Lee has a point about the substrate being irrelevant.
Bookmarks