Our Brains Don't Work Like Computers

Roland Piquepaille writes "We're using computers for so long now that I guess that many of you think that our brains are working like clusters of computers. Like them, we can do several things 'simultaneously' with our 'processors.' But each of these processors, in our brain or in a cluster of computers, is supposed to act sequentially. Not so fast! According to a new study from Cornell University, this is not true, and our mental processing is continuous. By tracking mouse movements of students working with their computers, the researchers found that our learning process was similar to other biological organisms: we're not learning through a series of 0's and 1's. Instead, our brain is cascading through shades of grey."

Comments

The Network is the Computer

[Doc Ruby]

Each neuron is like a tiny, slow analog DSP, feeding back FM around a base frequency (eg. about 40Hz in the brain's neural tract). The neurons have feedback among themselves locally, and send out some larger feedback in fiber bundles, signalling other clusters along the way. It's like a teeming kazoo symphony, without a conductor.

Re:The Network is the Computer

[SilentChris]

Well, actually, from the article it sort of sounds like a multibranch computing article I read a while back. I'm not sure if Intel actually went through with this, but the idea was to have a CPU process multiple "paths" ahead of time.

So, for example, for a simple if statement waiting on user input, part of the CPU would process the "true" result of the statement and part would process the "false" one. When the user made a decision, one would be used and one would be thrown out. In theory, computing these branches ahead of time was supposed to be faster than doing things linearly.

Again, though, I'm not sure Intel went through with this. They were the subject of the article.

Newsflash

[tupshin]

Headline: Brains More Like Neural Nets Than Traditional Programs

Who woulda thunk it.

ftp://ftp.sas.com/pub/neural/FAQ.html%23A2

'Most NNs have some sort of "training" rule whereby the weights of connections are adjusted on the basis of data.'

Insert joke about the 1980's (or 60's/50's/40's) calling). Somehow I don't think Norbert Weiner would be the slightest bit surprised.

-Tupshin

also worthy of note

[twiggy]

The book "On Intelligence" by Jeff Hawkins (of Palm fame) and Sandra Blakeslee is all about how the brain works, and why people's approach to AI is not going to come anywhere near emulating the brain...

Figured it was worth mentioning given the subject matter of the thread... I liked it.. good read, if a bit dry at times...

Let's see the numbers

[DynaSoar]

"In this model, perception and cognition are mathematically described as a continuous trajectory through a high-dimensional mental space; the neural activation patterns flow back and forth to produce nonlinear, self-organized, emergent properties -- like a biological organism."

Fine, let's see the math. Let's see the trajectory calculations. How about those calculating the space? Calculating the number of dimensions the space has, and how fast that number changes over time?

40 years ago brain scientists realized that computer architecture made a good metaphor for how the brain works. (They did NOT assume there was no feedback, contrary to the article). It made a handy and productive way to look at things so they could figure out more about what was really going on.

10 years ago brain scientists realized that they could use the way cool chaos stuff the describe the way the brain works. Believe me, I know; I've been to the Santa Fe Institute twice. It worked particularly well for me because I'm essentially a signal analyst -- I HAVE to define a set of variables, estimate how well they work, and decide how many of my arbitrary variables to keep or throw out.

It's still only a metaphor. And unlike the specific specific processes described by cognitive science, the dynamic system stuff remains nebulous. It claims a mathematical legitimacy which it can really claim only in concept because the actual math of the acutal operations are is beyond the abilities of anyone making the claims. The fact that it *can* be described this way is no less trivial than the fact that processes can ge grouped according to the traditional cognitive science concepts.

Trajectories on phase space are soooooooo sexy. But if it's any good, it'll result in something more concrete than more people picking up this flag and waving it while shouting the new slogans and buzzwords. Until that happens I peg this with the study that "calculated" the "fractal dimension" of the cortex just because it has fold and folds in the folds.... so fsking what.

Re:Fascinating

[nmoog]

Thats why everyone needs to install this super dooper greasemonkey script: De-Piquepaille Slashdot

It blocks stories submitted by Roland. Of course, you'd have to have installed greasemonkey. Which I forgot to do on re-install and hence saw this fucking stupid article.

predictive branching

[rebelcool]

Modern processors do in fact, do this. They maintain statistics on the branches and go forward on the branch deemed most likely to be taken. Its based on a simple principal - if you've taken the same branch a few times before, you're likely to keep taking it from now on. Think of how loops work.

Granted, if the processor is wrong, it has to clear the pipeline and start anew (which is costly), but the benefits outweigh the negatives.

Re:comparisons

[MarkusQ]

But a computer cannot demonstrate this truth. I don't claim to understand why not, but it clearly says in the wikipedia article that it can't.

Short answer: you're incorrect.

Long answer: The reason you seem to think that you are correct is also, I believe, incorrect. Godel's proof basically involves forming the statement "this statement is false" in a specialized language that allows you to do so without reference to pronouns--instead, he assigned each symbol a unique integer, and worked out ways of manipulating them both with and without regard to their "meaning". That part would be easy to do with a computer (e.g. asci/text editor/compiler).

Next, he posited a string of symbols where the meaning was related to the process for the manipulation of the meaningless symbols (this is also easy on a computer--sort of like using an editor to edit its own source code).

Using these, he constructed a relatively normal argument about the meaning level that coresponded to an argument at the symbol level--an argument that said "the argument represented by this long string of digits is unprovable"--but the kicker was the long string of digits was the coded representation of the argument itself. If false, the system could obviously not prove it (since we are assuming here that it only proves things that are true). Therefore it must be true, but that means it can't be proven within the system. Tricky, but there was nothing magical about the logic--no quantum mechanical must-derive-this-step-from-the-sprit-world voodoo that would make it impossible for a computer to follow.

--MarkusQ

P.S. A computer might not be able to understand the proof, but that shouldn't be held against it--after all, most of the people who discuss it don't understand it either.