Slashdot Mirror
Microprocessors With Living Brain Tissue
FurBurger writes: "Another interesting article from Discover.com on NeuroComputers . 'Although scientists have developed software that attempts to mimic the brain's learning process using only the yes-no binary logic of digital computers, all the connections in a personal computer are wired back at the factory. Breaking a single one of these connections usually crashes the computer.' (a la Windows =))" The promise of neuron-based computers is greater flexibility and fault tolerance, with components that require very little power. Or, as FurBurger puts, it, "Watch out, Transmeta!" Mike also points to a June article on the BBC about the same group and their "leech-ulator."
I hate M$ as much as the next guy, but I hate to see brain-dead digs like this one show up on the front page of Slashdot. What next, Bill Gate's Momma jokes? It makes us look stupid.
Cut it out.
If you're not wasted, the day is.
If you're not wasted, the day is.
Did anyone else have 790 pop in mind when reading this post? He's a character in the great sci-fi series *LEXX*, who happens to have a piece of human brain tissue at the core of his circuitry. Which explains how he, a robot head, could fall in love with the love slave Xev.
Sci fi and science have always played off one another. I wonder how many scientists were inspired growing up by the fantastic creations of the 1950s comic books, like aeroplanes that could fly into space, or by Asimov and others.
But, I digress. I just have to point out that it may be difficult to overclock human brain tissue, but...
"The more corrupt the state, the more numerous the laws."--Tacitus, *The Annals*
--
Imagine the Office paperclip a few years down the line if its capble of changing, learning and growing in strength
-- Oh Well
Breaking a single one of these connections usually crashes the computer.' (a la Windows =))
so, who can find me an operating system, open source or closed, that can withstand an electrical connection failure? redudant hardware is typically abstracted from the OS, so it stands to reason that any real hardware failure is gonna cause you a very real OS failure. GNU/Linux or GNU/Not.
Today Mr Random J. Hacker was banned from keeping computer equipment for life after being found guilty of cruelty to electronics after leaving his PDA on the dashboard of his car. Mr Hacker said "I only popped into Radio Shack for five minutes and I thought that it would be fine, left in the car". A spokes-terminal for the SPCEE (Society for Prevention of Cruelty to Electronic Equipment) said "The interior of a car can heat up rapidly, literarily cooking electronic devices to death. You should always try to take any devices with you when you get out the car, but if you must leave then inside make sure that you wind the window down and leave them a bowl of water".
I am absolute disagreement with your assessment of living neural tissue vs. silicon. Comparison of conduction rates is invalid and misguided for a host of reasons, an elctrical engineer or a neuroscientist could give you a trillion reasons why each. But leaving aside that technical point, I think you misunderstand the potential advantages of neural circuit over simple semiconductor technology. 1) Complexity of input, simplicity of output. Silicon semiconductors are on or off. CNS neurons are arranged in such a way that thousands of inputs synapse onto one neuron, which then either does or does not fire an action potential. This is an extremely elegant and flexible system. Each one of those thousand odd inputs is either inhibitory or excitatory, and also has a set strength realtive to other inputs. The beauty of this system is clear- it allows distillation of huge amouints of information into one action- the exact ability we are searching for in intelligent beings, whether natural or built by us. While this could be SIMULATED by a comparatively gigantic nmber of silicon transistors (in the form of a chip), it would never possess 2) Learning ability Neurons and synapses are plastic. The strength of individual inputs in the CNS is continually changing, being reinforced by certain actions and reduced by others. This is the cellular basis of learnt behaviour (ok, a bit simplified). Silicon can't do this. An OS running on the silicon could be programmed to SIMULATE this behaviour, but again in an artificial, memory hungry way. Your brain doesn't have an OS, it is an OS! That is the plain advantage of organoc neural computers. A chunk of memory doesn't need to be clogged up by instructions on HOW to artificially "learn", the whole thing is a learning machine! By the way, why are they using leech neurons? Surely they suck! (sorry, couldn't resist)
Tonight the sky is empty. But that is nothing new
I've worked with artificial neural networks to some extent in the past, so I hope that lends my words a bit of credibility. I don't call myself an expert, by any means, but I know a bit of what I'm talking about. (Tho' I'm first to admit that "a little knowledge is dangerous...") Anyways.
At some time in the past (I don't know exactly when, probably in the 50's), a group of computer scientists, excited by their new technology, tried throwing together a large number of analog "neuron" circuits to see if they will exhibit any kind of self-organization. It's similar as what these people are proposing to do with living tissue, except that it was done with electronics.
I don't know the details of what they tried, but the conclusion was simple. Nothing happened. It just sat there and did random stuff, from beginning to end.
I don't think self-organization in the brain is possible without having some kind of enforced organization at birth that gets the process going. To put it another way, the neurons have to be "pre-programmed," from the start, to organize themselves.
In artificial neural networks (simulated with digital computers), the problem is finding the right network topology, and the right learning algorithm to fit your problem. Maybe things have changed, but the last time I worked with it (about five years ago), this problem was still a black art. And not only do you have to get the network itself correct, you have to encode your problem in the right way, in order to get the best results. You have to do a lot of pre-programming (and maybe even some post-processing).
It goes to show that "self-organization" is not a magic bullet. The problem is that the whole system interacts. The operation of each neuron, the interactions between them, the format and encoding of the input data, and the format and encoding of the expected output data. It's all intertwined.
Will biological neural programming have the same problems? Or will the fact that real neurons are being used reduce the problem? Maybe it will actually compound the problem by making the whole pre-programming question heinously complex. After all, neuron interaction is more than just synapses: there's hormones, there's chemistry, and maybe there's stuff we haven't discovered yet.
DeWeerth says, "we might not have to understand [self-organization] to exploit it." I'm not about to argue against a person who no doubt knows his stuff (and I don't for a moment think he's unaware of the issues), but I must admit to being a little skeptical. Programming with zero effort has been a dream in A.I. circles for a long time. I can't help but feel that it's a pipe dream.
Accountability on the heads of the powerful.
Power in the hands of the accountable.
It is correct that the brain has some fantastci computing powers we cannot mimic yet. It is also correct that the brain rewires to an extent. It is also very robust in that it can sustain substantial damage and still continue to work.
But this has to do with the LOGIC of how the brain works and NOT the MATERIAL.
In order to make our silicon function as the brain we have to understand how the brain functions. And here we're talking about billions of very complex neurons working in parallell. (Even for insect we're talking tens of thousands).
When we understand the logic we can implement it using the best suited technology.
Living neurons are slow.In the human brain the maximum spikerate is 1000 Hz and the conduction velocity through the nerve-fibers are not that much either. (Don't remember the figures, but we're talking about metres per second.) This is much much slower than silicon.
The comparrison between a transistor (2 states) and a neuron (more or less analog) is stupid. We can pack a shitload of transistors into the same space used by a neuron. In addition we don't have to keep the silicon alive.
Silicon can never rewire, but the logic of
rewiring can be implemented.
While the article is interesting, it is not interesting to see a computer built from brain tissue. But the knowledge of creating a computer from brain tissue would probably enable us to build real smart silicon.