Slashdot Mirror
Brain Prosthesis Ready For Testing
jhouserizer writes "New Scientist is reporting that an artificial hippocampus is ready to undergo testing. The leader of the team of scientists is Theodore Berger of the University of Southern California in Los Angeles. They hope these artificial hippocampuses can replace damaged (stroke, Alzheimer's, etc.) portions of your brain. I wonder what portions of 'you' would be noticeably different to your family & friends? I wonder how long it will be before we can have HUDs, such as in this story by Cory Doctorow?"
All the creatures will die, And all the things will be broken. That's the law of samurai. (Jubai, 1605)
consider:
- the hippocampus is a black box (they can't even see the "object code," if you will
- complex systems are notoriously difficult to debug
I find the claim that the scientists have considered every possible behavior and simulated it in firmware to be suspect.
How can they be sure they have considered every possible input/output? How can they be sure that what they observed was "correct" behavior?
Any biologists or neuroscientists care to elucidate? Also, how similar is the human hippocampus to the rat's? Couldn't the behavior differences require complete regression testing? It seems like this increase in precision in medicine demands a commensurate increase in the precision of testing.
Why not both?
I remember those kinds of debates, and it always seemed to me that people got very hung up on the idea that only human experiences count for anything. There was this assumption that AI's goal is to become human is the sense that it actually experiences mental states identical to those of humans. But - what's wrong with having sophisticated mental states that aren't human mental states?
It will be really interesting once this sort of prostetic brain surgery happens - to be able to interview the patients and see if they really feel as if their mental states are different as a result of the new "tissue".
Isn't that the premise behind the Turing test?
I agree it acts exactly like a mind would, but it's not a mind.
I've seen the cams and pistons and bore and stroke and valves and shafts and spark plugs, coils, compression, explosion, expansion etc. I agree it acts like an internal combustion engine, but how do I know it's an actual internal combustion engine, and not just acting like one?
The point being, we judge our own mind solely by how it acts, depite knowing it's just simple electrical impules and synaptical thresholds, so we can only just an artificial mind on the same basis.
A good book that adresses all these issues (not as much with AI, but with mind) is "The Minds I", by Douglas Hofstadter and Danial C Dennett.
Basically, it seems they sent an input to it, got the output, and repeated that enough times to make a chip that produces the right response for each input. I see some problems with this:
First, isn't the brain a dynamic thing? This doesn't sound like something that can adapt.
Second, does this method work at all? If I say, sent bytes to a router, analyzed the output, and made a chip that produced that output with the input I sent, would it work? It just sounds way too simplistic.
Maybe somebody has better info on this?
Don't get your hopes too high for this invention. The process overall is very, very cool, but the fact that they don't understand how the hippocampus works, they just worked out a neural net model of imputs and outputs in rats, leads one to believe there will be a lot of bumps down this road
You're right, but just as a thought experiment, replace the word "hippocampus" with "neural tissue" or "brain cell" or "brain cell molecule".
My point being, there is always a lower level which you might not understand completely. That doesn't mean you can't successfully mimick the behaviour and achieve the same result.
That's assuming that the power of attorney is given to a family lawyer. Often guardianship/power of attorney is obtained by family members who care for a person who is no longer able to take care of financial/medical decisions.
As a person in that situation (my wife had a stroke 3 years ago that left her with communication/cognition difficulties) I'd be willing to see what something like this would do, and given the choice of "would you like to be like you were before" I'm fairly certain she would agree. I'm not sure what would qualify as "living well otherwise" with some forms of brain damage.
Unfortunatly, things like this are still a long way off, but here's hoping.
unless ($Brogen) { $fixit = ''; }
Is a wooden box a box or is a metal box a box? Boiware, silicon, data, what's the difference?
;-)
Sounds more like a theology discussion.
Quack, quack.
You raise a very interesting point. Recording. With the chip designed to be external, there is definitely the ability to have your memories recorded onto chips. Or even more interesting, have people upload their memories to a computer, to where you can download them and store them in your head.
This could be a boon for training. Imagine being able to pull down a file from the net, jacking into a usb port, and after a while, being able to speak chinese. Or have an intimate knowlege of physics. Wow.
On the other hand, it would make the term "knowledge transfer" more insidious. Law enforcement would love this. Suspicious spouses too. Having an interface like this would end the last private place in your existence: your own head.
But this is only just come out of it's conceptual stage. It'll be interesting to see where the technology takes it.
Who knows what changes in thought patterns might occur with completely fresh neurons in a brain?
No need to wonder; look at how "fresh neurons" behave in real life. In other words, look at newborn babies. The answer is "not much".
Neural weights only really have meaning in highly specific contexts. Even if you could "copy & paste" neurons in your brain, the new location would render the neurons effectively noise, having no coherent effect, and thus having effectively no effect at all.
Again, you can partially see this in the real world. We've watch people's brains adapt to losing vision and going to sound for their primary input, converting vision brain area to sound brain area in the process. It's not magical; the old vision stuff is effectively useless and completely re-purposed. Cognitive-level concepts are far, far, far higher then neural weights. So the old neurons are effectively full of garbage.
That's the reason this is so impressive to me. We've more-or-less decoded how the ear transmits sound to the brain, and have devices that can do this now, albiet not quite as well as real ears yet. We've started with ocular implants, though I don't know if that uses direct ocular nerve stimulation. This is because there are reasonably rational patterns that the sense data is transmitted in.
But once you're inside the brain, the nerve impulses have no objective meaning. "Thought transmission", if it is ever acheived by technology, won't be as simple as replaying neural impulses from one brain into another; there's no one-to-one correspondence between neurons, and certainly no corresponence to neural weights. (Odds are, we'd have to learn to use it, and it would 'just another' line of communication, not 'mind reading' as it was portrayed in past literature. Of course, if too much information is transmitted skilled "telepaths" might still get more information then the sender intended, just as reading body language can tell you more then the speaker intended.)
To acheive any success with an internal brain structure, understood or otherwise, is (IMHO, this is subjective of course) orders of magnitude more interesting then the ocular implants, which were pretty impressive themselves.
Again, I emphasize: This isn't magic. This is droll reality. Out of context, a neuron is nearly useless.
Visual pathway prosthetics work either by stimulating the retinal ganglion cells in the eye, or by stimulating in primary visual cortex.
Neither way has yet proved useful enough to deploy on a large scale. It is a little tougher than a cochlear implant, because you have to seal the device inside the eye, and provide a power source that can stimulate a bunch of microelectrodes.
Just because we don't understand something now doesn't mean it cannot be replicated in the future. There was a time, about 30 years ago, when simulating the function of the human ear was unheard of. Now, patients get cochlear implants and can understand speech. Artificial hearts are in use. The brain is a matter of time, the retina will come relatively quickly, next will be implants that couple motor cortex to external devices, there are already stimulating electrodes that modulate the motor system...
where we will be in 30 more years is pretty cool.
And nowhere do they ever claim, even in a best-case scenario, that they will be able to replicate the function of the hippocampus completely and accurately. But that isn't the point of a prosthesis. The point is to help, as much as possible, even if it's only a mere shadow of the functioning organ. Given a choice between encoding 1% of the normal number of long-term memories and 0%, which would you choose? You seem to be saying that this research should be considered a failure if it doesn't approach 100%.
Or maybe you're right; maybe the brute-force approach won't work at all. But right now it's the only way to even begin to tackle the problem of a working prosthesis. And whatever else happens, we will learn something from it. But the point is, even if it eventually fails, it is up to THEM to tell us that it failed, AFTER they do the research. Not for you to make your sweeping pronouncements when the research is in its infancy.