Computer VS Human Brain

[adwilk]

I have a simple idea how a computer works, from the cpu all the way to a printed masterpiece from mspaint... i have little idea how the human brain works but i was wondering if the two could somehow be compared. ie.

what is the FSB of the nervous system?
how many GB of info can the brain contain...?
at what ghz does the brain process info?
anything else that a comparison could be made. This isnt for homework or anything, i was just wondering.... can a comparison even be made? why or why not? sorry if something similar has been posted before... thanks

 

[CTho9305]

at what ghz does the brain process info?

At most 200Hz, or 0.0000002 GHz

 

[Soccerman06]

The only thing that could be directly compared would be the storage capacity of the brain. I dont know the specifics, but there are 100 billion neurons, each with their own set of memories and branches. If you think about it, the human memory has the capacity to remember everything it sees over its lifetime (but it doesnt remember everything because its still being made at a young life).

 

[CycloWizard]

Lots of anecdotes get thrown around, but I have never seen any real, hard data. I don't believe such quantification of the brain is necessarily possible, as we don't even know how it stores memories. However, I think it's safe to say its storage capacity is in excess of the equivalent of exobytes. Since the nervous system is analog rather than digital, it is again hard to quantify an FSB. You could say that it is infinite, since one analog signal is essentially infinitely many bits, but this doesn't really mean anything. The only frequencies that I'm aware of are in the 20-60 Hz range, but these are sophisticated mechanisms far removed from the brain, so I'm sure it really works much faster than that (again, probably in an analog fashion, so it might not be comparable to a frequency).

The best I can tell you really is that the brain operates in an inherently different manner from modern computers. It functions in an almost infinitely parallel fashion, while most computer parts function in series. This is why we can recognize complex images and patterns without a thought, while the same task is impossible, regardless of the CPU power, for a computer. The converse is also true: the simplest computer can perform limitless digital calculations infinitely faster than your brain. Different functional specialties lead to different levels of performance depending on the task at hand.

 

[bobsmith1492]

There's no comparison. Computers are for raw data; minds are for raw life.

 

[Gannon]

As always with everything, there are most likely better things neurons can do then transisters, and vice versa, we'll probably develop hybrid models based on biology + technology, i.e. a trans-neuron.

 

[iwantanewcomputer]

completely different. human brains have evolved to do some insanely complex calculations/programs nearly instantly, yet we can't do math(adding/multiplying/other operators) as well as the most simple transistor computers

 

[Xyo II]

Originally posted by: iwantanewcomputer
completely different. human brains have evolved to do some insanely complex calculations/programs nearly instantly, yet we can't do math(adding/multiplying/other operators) as well as the most simple transistor computers

Our subconscious handles most everything. Our conscious mind isn't aware of the subconscious's computing, only when the subconscious tells us "hello, that semi is going roughly 45 mph at less than 200 yards in a path that intersects your intended path of travel", and all we see is, "I'm not going to make it, I have to wait"-our mind is too busy to be thinking about little things like 125x45= ? Just think about how many calculations and judgements the mind has to make in just walking, this is why robots are slowly evolving. The world is more complex than a operating system.

Edit: oh, and the storage capacity of the brain greatly exceeds our life span, so the best that psychologists have come is by saying that the brain has an infinite storage capacity, since nobody can "fill" it up in their lifetime. For another reasoning, Look what Scirus brought you. Good boy.

 

[harrkev]

My take is that you cannot compare silicon and flesh. A computer can beat any human alive at chess. It can numerically solve integrals that would take a human years to do with a pen and paper. So computers are far better.

On the other hand, a simple dog can fetch your slippers, run around in almost any environment without getting stuck, find his own food (in some cases), and has a visual system more advanced than our best computers. So even a dog is smarter than a computer.

This is sort of like asking if a spoon is better than a fork. The answer depends on whether you are eating steak or soup.

For reference, the brain is rather slow compared to computers. However, the brain maikes up for it by being incredibly massively parallel. In a computer, there is one processor which has to do everything. In a brain, you have millions of neurons, each working together.

 

[spinaltap]

In addition to being massively parallel, because all neurons are connected, the brain also works in a relationary manner, rather than exact matches. This is why people are good (for the most part) at recognizing faces and language even though it may not be an exact match from they are used to. This is why people can also draw context from writing and search for things based on meaning rather than set algorithms.

I think that as technology becomes more and more sophisticated, computers will start to adopt a lot of the principles which drive the brain. Search engines will become smarter in a sense where people can specify "kind of" what they are looking for and the best contextual matches will be retreived. Voice and photo recognition will be able to work more reliably based on advances here as well.

This is a main driver to why a lot of electrical/computer engineers are starting to study things like cognitive science. By learning how the brain transferres and processes information, principles in nature can most likely be reproduced in computers.

 

[concreteDonkey]

I think computers have got themselves mixed up, if they just copied nature they'd be really good!

Lets compare a computer with a bee hive:

A computer has a CPU which runs the show : The hive has a queen

A computer has peripherals which do it's bidding : The hive has workers, drones, bouncers etc

The CPU works really hard and the peripherals work relatively slow : The queen sits on her arse all day eating the best food while the bees work their arses off

Computers need complex protocols and lots of 'handshaking' to communicate : The bees dance to tell each other where the flowers are

Ok so I don't really know where I was going with that, but bees use space more efficiently than us humans(the honeycomb is made of hexagons which fill spaces most efficiently) and we design computers.

So should we let bees design computers? Or is the human brain like a hive? What was the question again?

 

[the splat in the hat]

Theres lots of documentries on the brain v computors , and so far the brain is way more powerfull, putting aside the accuracy of computors calculations , our brains can still do it if we are taught correctly. The brain can compute trillions of ''switches'' terrabytes , of info at a time , everything from physics's of a moving object ,to interpreting colours on the fly at 230 kmh , our brains are on one level far more advanced than any computor, but focused on 1 task for pure calculation the raw pc will win every time, the day the computor beat man at chess was a good exsample, the brain is just far more complex , consider that it can do all of this computing totally silent with no overheating problems and you soon realise how far our computors still have yet to come .
Over time the computor probably will surpass human brain for pure intelligence , it may end up running our world , it could even end up with a true and as real as yours or mine , concessness.
That debate is still out ........

 

[pinion9]

I have written AI programs before, so I have some real life knowledge of this. However, let get something straight about the computers beating man at chess.

Deep Blue beat the greatest chess player on earth. However, Deep Blue was a super computer with specialized hardware that was made specifically for playing chess. On top of that, Deep Blue had the storage capacity to actually store every possible permutation of the board at a certain level (say, 14 pieces left on the board or something like that.) Therefore, Deep Blue new very early on which states could lead to a win, and would try to force the board to a winning state, even if that state was 100 moves away from the end of the game. Deep Blue played a perfect game, analyzing every game board, and could look tens of turns into the future to see how a single move could be affected if the alternate player played perfectly.

In other words, before you even have a couple of pieces of the board, deep blue could have led you to a winning state and there is no human on earth that can make the sort of computations to tell that in two minutes. The algorithm simply searched for a winning board (marked as such by the programers) and the computer searched for ways to get to that winning board. Not that impressive.


Neural Networks, or genetic algorithms, are the current hype in AI. Essentially you create decision trees with many inputs and many layers. There are weighting functions between each layer of these nodes, and thresholds. Inputs are taken and passed from node to node, with essentially random weighting functions altering the values between nodes. Eventually you get 1 answer out of the entire network.

I built one for a game called mancala. I won't go into details of the game, but essentially there were about 12 different values that I used as inputs to my neural network. My weighting functions were random at first, and I made essentially 100 players with random weighting functions that would play against one another. The top 50 winning players would have their weighting functions altered very slightly, and then repeat. I did this for 50 generations of players over the course of 1 week on my home PC. The result? The first generations of players played randomly and were trivial to beat. The 50th generation of players were incredibly hard and I *never* beat the number one player at the game. My sister in law, who kicks my butt rotuinely, beat the number 1 player once out of 5 games. Amazing results.

The idea is that these inputs matter, but how much? How do they have an affect on one another? The neural network allows them to interact with each other, and the weighting functions allow them to interact at different levels. For example, if a node had a weighting function of .00000001, then its input obviously matters very little in the final outcome.

Edited for spelling crap

 

[bigal40]

Originally posted by: bobsmith1492
There's no comparison. Computers are for raw data; minds are for raw life.

:thumbsup: I think this sums it up

 

[yosuke188]

If you show me a computer giving offsprings, I will tell you the answer.

 

[TSS]

that would be robots reproducing their main proccessors

i belive that besides the fact we got billions of neurons, they can also create new neural pathways to other neurons. basicly its putting 2 cores on a die, give them no way of seeing eachother, and they will dig a channel to send data.

i think that its better to compare the brain to the internet then just 1 computer. millions of cells (the individual servers and computers), connected by millions of kilometers of pathways (optic fibre baby!) and TCP/IP that can find its way from a cell to a cell by going another way if one craps out.

i'd almost say, if the internet found a way to "grow new cables from and to servers", and finds a way to have all that information stored on it work together instead of just single units, it would work like a human brain.

 

[Gannon]

Originally posted by: pinion9
I have written AI programs before, so I have some real life knowledge of this. However, let get something straight about the computers beating man at chess.

Deep Blue beat the greatest chess player on earth. However, Deep Blue was a super computer with specialized hardware that was made specifically for playing chess. On top of that, Deep Blue had the storage capacity to actually store every possible permutation of the board at a certain level (say, 14 pieces left on the board or something like that.) Therefore, Deep Blue new very early on which states could lead to a win, and would try to force the board to a winning state, even if that state was 100 moves away from the end of the game. Deep Blue played a perfect game, analyzing every game board, and could look tens of turns into the future to see how a single move could be affected if the alternate player played perfectly.

In other words, before you even have a couple of pieces of the board, deep blue could have led you to a winning state and there is no human on earth that can make the sort of computations to tell that in two minutes. The algorithm simply searched for a winning board (marked as such by the programers) and the computer searched for ways to get to that winning board. Not that impressive.


Neural Networks, or genetic algorithms, are the current hype in AI. Essentially you create decision trees with many inputs and many layers. There are weighting functions between each layer of these nodes, and thresholds. Inputs are taken and passed from node to node, with essentially random weighting functions altering the values between nodes. Eventually you get 1 answer out of the entire network.

I built one for a game called mancala. I won't go into details of the game, but essentially there were about 12 different values that I used as inputs to my neural network. My weighting functions were random at first, and I made essentially 100 players with random weighting functions that would play against one another. The top 50 winning players would have their weighting functions altered very slightly, and then repeat. I did this for 50 generations of players over the course of 1 week on my home PC. The result? The first generations of players played randomly and were trivial to beat. The 50th generation of players were incredibly hard and I *never* beat the number one player at the game. My sister in law, who kicks my butt rotuinely, beat the number 1 player once out of 5 games. Amazing results.

The idea is that these inputs matter, but how much? How do they have an affect on one another? The neural network allows them to interact with each other, and the weighting functions allow them to interact at different levels. For example, if a node had a weighting function of .00000001, then its input obviously matters very little in the final outcome.

Edited for spelling crap

Nice post but I'd like to add winning and losing is complicated because it's determined by how the environment effects the players and decisions made, there may in fact be too many combinations to compute or not enough, so you get random wins and losses since total knowledge of certain systems may be practically impossible, so you get probabilities of winning or losing.

1) Is the game perfectly symmetric? (equal chance of winning and losing?) Since the ideal game with omniscient AI would end in a draw every time, unless the game was somehow flawed in that there was a "first movers" advantage, or some sort of environental or statistical loophole.

2) Is there a dominant strategy? (i.e. many early strategy games focused on either rushing, or how fast you could accumulate one powerful unit, or simply on the speed of resource acquisition and production of units)

3) Resouces, how much storage, recombination and computational resources is available to the person and the computer?

4) Human beings are self-aware, we do not understand what causes something to go from a 'machine' just sending bits and bytes back and forth, to a being that senses, sefl-reflects and is aware of itself.

 

[Maximilian]

Well, the OS would be like the mind of the computer, not the CPU, as the CPU is a tool of the OS, it wont do anything unless the OS tells it to. Windows ME computers can be considered retarded in this respect in comparison to more advanced operating systems

Also, i dont see how storage space can be compared, hdd's use multiples of 1024 (sort of)4? nibbles in a bit, 8 bits in a byte 1024 bytes in a kilobyte etc. What does the brain use? NTFS? somthing else? who knows, therefore id go with the "it cant be filled option". If you did fill it, then it would probably overwrite the oldest stuff, and if thats overwritten how do you know youve forgotten it? I personally believe a macintosh classic has more space than me, but thats just me my memory sucks.

Numbers cant be given for the FSB of the nervous system or w/e but a generalisation might be possible:
FSB of nervous system - very fast, i have excellent reflexs
GB brain capacity - inifinite?
Cache (short term memory) - next to nothing, im comparable to a celeron 128kb
GHZ of brain - very fast for non numeric calculations.

 

[herm0016]

digital vs. analog. . . .

 

[adwilk]

Interesting stuff, anybody know of a program i can download to reformat my brain? I really do think that that is a legitmate question, maybe someday. What if that were possible, what if we could develop a system to "program" the brain. If it has infinite "storage", what if? Imagine how much information could be uploaded to it if this ever came possible by means other than life experiences, what if the brain could learn instantly... just a thought...

 

[the splat in the hat]

Originally posted by: pinion9
I have written AI programs before, so I have some real life knowledge of this. However, let get something straight about the computers beating man at chess.

Deep Blue beat the greatest chess player on earth. However, Deep Blue was a super computer with specialized hardware that was made specifically for playing chess. On top of that, Deep Blue had the storage capacity to actually store every possible permutation of the board at a certain level (say, 14 pieces left on the board or something like that.) Therefore, Deep Blue new very early on which states could lead to a win, and would try to force the board to a winning state, even if that state was 100 moves away from the end of the game. Deep Blue played a perfect game, analyzing every game board, and could look tens of turns into the future to see how a single move could be affected if the alternate player played perfectly.

In other words, before you even have a couple of pieces of the board, deep blue could have led you to a winning state and there is no human on earth that can make the sort of computations to tell that in two minutes. The algorithm simply searched for a winning board (marked as such by the programers) and the computer searched for ways to get to that winning board. Not that impressive.


Neural Networks, or genetic algorithms, are the current hype in AI. Essentially you create decision trees with many inputs and many layers. There are weighting functions between each layer of these nodes, and thresholds. Inputs are taken and passed from node to node, with essentially random weighting functions altering the values between nodes. Eventually you get 1 answer out of the entire network.

I built one for a game called mancala. I won't go into details of the game, but essentially there were about 12 different values that I used as inputs to my neural network. My weighting functions were random at first, and I made essentially 100 players with random weighting functions that would play against one another. The top 50 winning players would have their weighting functions altered very slightly, and then repeat. I did this for 50 generations of players over the course of 1 week on my home PC. The result? The first generations of players played randomly and were trivial to beat. The 50th generation of players were incredibly hard and I *never* beat the number one player at the game. My sister in law, who kicks my butt rotuinely, beat the number 1 player once out of 5 games. Amazing results.

The idea is that these inputs matter, but how much? How do they have an affect on one another? The neural network allows them to interact with each other, and the weighting functions allow them to interact at different levels. For example, if a node had a weighting function of .00000001, then its input obviously matters very little in the final outcome.

Edited for spelling crap

yea like l said the computor can focus all its energy on 1 task , how was l incorrect about the chess v man , what are you clearing up ? That ok the computor only need find a winning map , put the human to the same task and time it, make it fair, see whats faster focused. But you are right in the detail , l was just making a piont on human brain v computors.
l would say when / if they biuld a computor that has the potential to fundamentally be smarter , to mimic a human , l would say that would be a highly specialised computor, and the piont is with all its energy focused it beat a mans brain .Regardless of short cut programmers , its pure power to crunch single quasion problems is massive , who knows humans may have a micro chip embedded into our brains in the future to increase our ability to focus on large problem / equasion. 1 can only hope !

 

[sdifox]

Think of it this way, computers are great in terms of simple tasks done billions of times. Your brain is very good at doing very complex thing, sometimes many of them, all at the same time. Also, computer can not think, we can.

 

[Sparky19692]

Being a simple robotics engineer I always find this subject simply amazing!
How ignorant and simplistic people become when comparing these two.
All of us sit hear marveling at a simple and logical piece of hardware at the same time what is your minding doing?
Well your breathing air how many muscles and billions of calculations does your brain do to accomplish that.
If your muscles do not contract properly you lose the rhythm and ca no longer breath. Hence death. Glad Bill gates is not involved here?

Your are reaaading, waht is that? How can your eyes amd brian inturprate this?

Wait a minute I can walk. Do you know that when you take your first step on a stairway yes that is singular,
your brain recalculates every proceeding step for height and width so you don?t have to fall on your face like
you did when you where 2 years old. What about climbing over that snow bank,
in front of my car with the ice in front of it.

I just wish I could get my Darn furnace to keep the house at 72.75°F no matter what.

You mean you can type??? Even after some one has moved the keyboard WOW

This list goes on forever people have NO appreciation for how vastly complex and adaptable the human mind
is until they try to replace it with robotics/computers.
Will computers ever get there maybe Never is a long time.

How many times did you blink during this??? Think about it.

 

[aerialcombat]

wow...

i'm just so marveled at how smart some of you people are...

i just wanted to give you guys props for enlightening me a bit more.

I, too, am planning to study AI in grad school, but I just realized I've got a looooooong way to go. Thanks.

 

[dunno99]

Since I only took one semester of AI at university, my comments aren't going to be worth much other than a very rough bound.

It's hard to directly calculate what's the equivalent supercomputer to a brain...instead, let's approximate what kind of neural net would be required to simulate the brain and what kind of computing is required for that neural net to operate at the same "frequency" as the brain. The brain is about equivalent to a 100 billion node neural net, with each node hooked up to about an average of 200 adjacent nodes (this neural net would be about 20 levels deep). So, we have about 10 trillion connections (we're not going to double count the connections). In order to calculate the activation function, a weighted sum is computed over all the adjacent nodes. Let's assume that 100 of the 200 connections are used for input, so that means we need to compute a weighted sum over 100 inputs, which is 100 multiplies and 99 adds (let's just say 100 adds...there's also the threshold activation comparison, but we'll ignore that).

So now, we need a 1 quadrillion adds and 1 quadrillion multiplies at about 30 Hz (we're guessing the "clock" of the brain here). We will assume that we can approximate everything with 64-bit floats, that we can ignore the initial 5 cycle delay of computing a multiply, have infinite bandwidth, customized hardware, and that everything is perfectly pipelined without worry to delays and synchronization. We will then need a 6 FPU CPU running at 10 PHz (PetaHertz). In the roughest sense, you can also think of that as 215 BlueGene/L supercomputers (each has about 280 TFLOPS) working in perfect harmony over an infinitely fast interconnect.