View Single Post
  #24  
Old 12-08-2012, 11:43 AM
Dragar's Avatar
Dragar Dragar is offline
Now in six dimensions!
 
Join Date: Jan 2005
Location: The only hill in Oxford
Gender: Male
Posts: VV
Default Re: Ensign Steve waxes philosophical on the Singularity, a thrad by Ensign Steve

So my opinion is a bit above my station, but I work with a number machine cognition people (one of whom transferred over from a neuroscience academic career path) and seems to align roughly the same with them.

We're in no danger of a singularity any time soon - our computers and our way of programming computers appears to be fundamentlally different to how biological computers work. And I am fairly convinced that our notions of intelligence hinge on that sort of functionality.

And to make things worse, to draw on Ensign Steve's point that we don't understand how our current computers work - we really don't understand how biological computers work.


Let me unpack these some more.

Our computers are built out of very specialised components, with minimal interaction between components. There is typically one component that does actual computations, which only just now has started expanding to contain multiple cores to allow for some measure of parallelism. The hardware is fixed and unadaptive.

This is almost the opposite of biological functionality. Biological brains are massively distributed, where specialism is implied but by no means fixed. Memory is distributed, not local, and processing is massively parrellel. And worse - the interconnections are huge in number, and are able to rewire themselves. What a difference! And what a mess!

The prevailing, very high-level view I have is that brains function as adaptive Bayesian reasoning networks; where learning is essentially a process of cementing (but never quite fixing) priors into that network. This makes them hugely adaptive and - while they can't have the raw focused attack power of an electronic computer - extremely powerful in a variety of tasks where a raw computation would otherwise take too long.

That's not to say we can't reproduce biological computers with electronics - we can! But programming parallel code is not easy for humans. I don't even know how we'd begin to start handling the adaptive nature of biological computers. And the massively parellel level of human brains is far, far beyond anything we're used to programming. In some ways it's a good job we are now limited in how small we can make our processors - this is going to force chip developers to start looking at these complicated problems that have so far been ignored.

But as far as understanding biological computers goes, we've been on completely the wrong path with electronic computers ever since their invention. We're essentially starting right from the drawing board again. It's going to be a long, long time before we start understanding and reproducing what brains are doing. And let me stress that understanding is key - people have made complete electronic reproductions of small biological brains - the human cortex even, I think. But when your model is as complicated as the system you are trying to understand, it's not clear what you can learn from this model - if anything!

So we are a long, long way from figuring out biological computers. I think we're only just starting to figure out why we don't understand them at all.
__________________
The miracle of the appropriateness of the language of mathematics for the formulation of the laws of physics is a wonderful gift which we neither understand nor deserve. -Eugene Wigner

Last edited by Dragar; 12-08-2012 at 03:30 PM.
Reply With Quote
Thanks, from:
But (12-08-2012), chunksmediocrites (12-08-2012), Crumb (12-09-2012), Ensign Steve (12-08-2012), lisarea (12-08-2012), Shake (11-10-2015)
 
Page generated in 0.06630 seconds with 11 queries