Slashdot Mirror
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."
Does anyone *really* think that computers and the brain work in the same way ? Or even in a significantly similar fashion ?
Well, by 'processors', I assume you mean neurons. These are activated to perform a firing sequence on output connections dependent on their input connections and current state, heavily modified by chemistry, propogation time (it's an electrical flow through ion channels, not a copper wire), and (for lack of a better word) weights on the output connections. To compare the processing capacity of one of these to a CPU is ludicrous. On the other hand, the 'several' in the quote above is also ludicrous... "Several" does not generally correspond to circa 100 billion...
No-one has a clear idea of how the brain really processes and stored information. We have models (neural networks), and they're piss-poor ones at that...
The brain behaves less like a computer and more like a chaotic system of nodes the more you look at it, and yet there is enormous and significant order within the chaos. The book by Kauffman ("The origins of order", I've recommended it before, although it's very mathematical) posits evolution pushing any organism towards the boundary of order and chaos as the best place to be for survival, and the brain itself is the best example of these ideas that I can think of.
Brain : computer is akin to Warp Drive : Internal combustion engine in that they both perform fundamentally the same job, but one is light years ahead of the other.
Simon.
Physicists get Hadrons!
Analog computers still exist in some places, but you list discrete values. An analog computer works with an essentially continuous range of charges instead of discrete values; and it works continuously in time, instead of in discrete steps. They're very good at integrating, which is the application I used them in.
I believe that the mind is (simply?) a quantum computer, and the article seems to support that idea. The human brain utilizes a sort of general interconnectedness of things to process thoughts as dynamic probabilities of state, with conclusions only being properly arrived at after a certain ammount of calculation has occured, but with all probabilities esiting well before the completion of the thought.
Anyhow, I should probably stop rambling and go outside or something.
Never eat more than you can lift -- Miss Piggy
How is this different than a schema? Haven't we known this since the 70's?
The path planner goes slower and generates paths that are initially ambiguous when faced with multiple alternatives. That's no surprise. I'm working on the steering control program for our DARPA Grand Challenge vehicle, and it does that, too. Doesn't mean it's not "digital".
I'd have put this in the "water still wet" department. People have known for decades that the brain used continuous, or analog computing.
That's actually the first book I read on the subject too, it's really great I agree.
The logic problem you refer to is modus tollens, "mode that denies" and people do find it extremely difficult.
Here is a rule.. if there is an E on one side of a card, there is a 5 on the back. The fronts of cards have letters and the backs all have numbers. What cards must you turn over (minimum) to prove the rule?
Here are the cards as you see them on the table:
N 5 9 E
Modus tollens is also called proof by contrapositive which could be a hint for solving this problem.
It is also true that there is no instantaneous jump from 0v to firing voltage. A different types of neurons require more or less neurotransmitters to reach the threshold voltage.
Well, there is no instantaneous jump in digital comptuers either, however once theshold voltage is reached in a neuron it fires very quickly and very sharply.
In any event, the voltages inside a neuron are quantized, always an even multiple of the charge on an electron, which obviously can be stored in a computer program, as could the finite number of molicules of neurotransmiter around them.
autopr0n is like, down and stuff.
Now, the output cannot be any more stable than the input, so if you have a fluctuating input, you will have a fluctuating output. However, I'll assume that the input is stable to some high level of precision. (This requires a screened input and a screened device, but those are doable.)
So we now focus on the device itself. Resistance varies with the exact composition of the material, the exact temperature of the material and the exact thickness of the material.
Problem #1 - it is very hard to make a resistor that is of absolutely 100% perfect even consistancy. So if you move the dial N% of the total length, you expect to get a resistance of (total resistance)*100/N. In practice, this will merely be the average value, there will be some variance. That variance dictates the absolute upper limit of how finely you can tune the dial, because at some point the level of uncertainty will become comparable to the level of adjustment.
The second problem is heat. All resistors generate heat, but heat increases resistance. Thus, all resistors will fluctuate in value. Remeber, though, that the composition is not 100% even, so the temperature cannot be 100% even either. This means that the fluctuation in resistance will be dependent on where you are on the dial, increasing the uncertainty.
The third problem is the thickness. Resistance increases as the diameter of a wire decreases. Variable resistors involve two conductors in contact with each other, thus scraping. Unless the dial is 100% circular, you MUST go over the midsection of the potentiometer more than one or both ends. This means that even if you DO somehow achieve a perfect variable resistor at the start, you won't have one after you start using it. You will vary the thickness across the length, and therefore vary the resistance of any segment.
Normally, these variations are too small to notice, which is why these components are useful in the first place. BUT, as you increase the precision, you increase the importance of these variations. Eventually, the variations will swamp the signal. At that point, tuning the dial with even greater precision will be worse than useless, as the value is utterly non-deterministic.
In reality, power fluctuations are of vital importance and are a big reason ADCs and DACs have not exceeded 26 bits of precision. Nobody has figured out how to get a power source stable enough, or a chip screened enough, to transform signals of one form to the other with greater precision than that.
If the cleanest signals we can get from an analog system are 26 bits wide, then producing a simulation of an analog system that is 64 bits wide will be vastly superior to any actual system we know how to build.
Now, it is entirely possible that the brain has developed a level of precision and signal clenliness that exceeds 26 bits. I'm not disputing that. I am disputing that any physical system you can build can exceed 64 bits and it probably can't get even close to that. So, a 64 bit simulation of analog signals should be as good as the real thing.
But what of waveforms? Can you reproduce waves, using discrete multi-state logic? Sure. It's called a transform. The three best-known transforms are Z transforms, Laplace Transforms and Fourier Transforms. Using these, you can do a surprising amount. Transforms work by turning a domain you can't use into a different domain that you CAN use. They're very useful devices.
Fourier Synthesis (the theory that any wave, of any complexity, can be reproduced with a sufficient number of overlapping sine waves) makes this clearer. We can represent a classic sine wave by denoting amplitude, start point and end point. We just need to be able to build a set of any number of these, and we
It's a small world and it smells funny; I'd buy another if it wasn't for the money; Take back what I paid (SoM)
NO! The brain is NOT a Turing Machine.
There is something called the 'Halting problem'. Basically, for any computation, a Turing Machine can:
-halt with success
-halt with failure
-get caught in a loop
The question is, if it hasn't halted yet, will it halt in the future or will it get caught in a loop? And you can prove that it is impossible to construct a Turing Machine that is able to answer that question. This is called the halting problem.
It can be generalised to prove that you cannot construct a single Turing Machine to decide whether a given statement is true or false and this is where it ties back into Gödel's theorem and it is this argument that some people use to relate Gödel's incompleteness theorem to the brain (which I find intriguing but I'm not sure whether I agree with it.) The important point here, however, is that you, as a human being, can solve the Halting Problem. It follows, that you are NOT a Turing Machine.
It can also be proven that Quantum Computers are not Turing Machines by the way, but even Quantum Computers are unable to solve the Halting Problem, so our brain is a step up even from Quantum Computers.
Let me repeat the point here again: The brain is NOT a Turing Machine and as such the limitations of a Turing Machine do not apply.
At the theoretical level, my argument is that analog (or nearly analog) computing provides a much better simulation of the mind than digital computing theory. At the pragmatic level, my argument hinges on decades of research on Artificial Intelligence being motivated by traditional tree-search algorithms, production systems, and other discrete serial processing systems. In speech recognition, however, hidden markov models have more recently been the popular method for automated word recognition, and they do indeed perform in a way that is describable as representing multiple potential words at once. Therefore, probabilistic algorithms and neural networks (programmed on digitial computers) are indeed useful and informative ways to build simulations of various human mental processes.
One of God's properties is that He or It or Whatever is omnipotent, no? The _supreme_ being? Why would a supreme being need to obey logic? Your riddle supposes that logic is the supreme entity or force in the universe. I would expect a omnipotent, supreme-being type God to be able to do non-sensical, as well as sensical things.
Computers are useless. They can only give you answers.
-- Pablo Picasso