Dateline: 03/23/97
IF I replaced your brain, bit by bit, with holographic chips where soft, mushy gray matter is supposed to be, would you end up being merely artificially intelligent?
This is not a trick question, neither does it arise directly from anything anyone is actually doing today. But it could arise indirectly in a couple of decades, out of work to construct a "bionic brain" already underway at the University of Southern California.
This is not The Six Million Dollar Man (the name of an American TV series from the 1970s, in which a severely mutilated hero is rebuilt from bionic parts. I can’t recall whether the hero’s brain was left intact, but I do recall he was given bionic eyes that could act like a powerful telescope).
For one thing, the price tag will be closer to $100 million; maybe lots more. Add the salaries and benefits of USC Professor Theodore Berger and a seven-member team of scientists—experts in everything from semiconductors to neurophysiology; multiply the total by the 20 years or so the project is likely to last; add the $10 million cost of their chip-making facility in downtown LA; and then triple everything for the many additional scientists and newer chip fabrication plants that will doubtless be needed. And you start to get the picture.
And for another thing, Professor Berger is not out to build a bionic person, or even a bionic brain in the sense of a "superbrain" capable of extra-ordinary mental capacity. If his research pays off, it will be in the form of restoring "ordinary" functionality to damaged brains. According to Samuel Greengard, a Burbank, California sci-tech writer, Prof. Berger’s goal is to "create a parallel-processing network that could function as a brain implant. Such a device would restore physical and mental functions lost to stroke, head trauma, Alzheimer's, epilepsy, and an array of other maladies."
The idea is really quite simple. A lump of brain tissue is basically just a parallel processing network that happens to be constructed from, well, tissue. Carbon-based, organic stuff; but still basically just a parallel processing network. And guess what! We (or at least, folks like Prof. Berger) can build parallel processing networks out of silicon-based stuff! So, cut out a lump of damaged brain tissue, plop in a chip, and the brain is back in business. There are just a few minor details to take care of.
Learning
The first is: How will the implant know what it is supposed to do? Most of Prof. Berger’s own time is consumed with finding the answer to this question, by "identifying and cataloging predictable and repeatable neural electrical patterns" (Greengard). In other words, "mapping" the patterns of activity in the brain, presumably in order that the appropriate patterns can then be imprinted or pre-programmed into the neural chips. From my understanding of the nature of neural networks, I’m not sure that this is either possible or necessary.
A central feature of a neural network is that we can know what goes in and what comes out, but we cannot know exactly what goes on in between. For instance, in the 1960s American neurophysiologist J. M. R. Delgado implanted tiny electrodes in animal brains. Inducing a high-voltage current in the electrodes induced changes in emotional, sexual, and social behavior. So, Delgado was able to know and manipulate the input (the electric current) and to observe and manipulate the output (behavior), but he still could not say what specific patterns of neural activity went on between the input and the output.
I suspect that what may be discovered is that the implanted neural chip will be self-configuring, in the sense that its "natural" state will be to process signals in just the same way that a transplanted heart processes bloodflow in its recipient’s body. Which leads us to the second issue.
Connecting the nerves
The fundamental issue with a heart transplant (besides such issues as tissue rejection) is to connect the veins (inputs) and arteries (outputs) correctly to the heart. But whereas with the heart we are talking about a handful of connections that can be eyeballed by the surgeon, with a lump of brain tissue (even just a thimbleful) we are dealing with billions or even trillions of ever so tiny and fragile connections between the synapses (made up of dendrites and axons which can be thought of, loosely, as the "wires" that join neurons together) and their equivalent connectors on the neural chip. Even just ten percent of the brain has up to a billion neurons and up to a trillion synapses!
Which leads to the third minor inconvenience.
Size of a truck
"You'd need an implant device the size of a pickup to get anywhere near the processing power of the brain," writes Greengard. Using current chip manufacturing technology, our thimble-sized implant would actually be closer to the size of a pickup truck’s engine block.
Enter Dr. Armand R. Tanguay Jr., director of the Center for Neural Engineering at USC and a member of Berger’s team, who is building tiny parallel-processing networks in that $10 million fabrication plant we mentioned earlier, using holographic techniques. Holography allows for signal transmission using light (photons) instead of electricity (electrons), which eliminates the need for wiring and, as side benefits, the friction, heat, and potential for short-circuits of electrical wiring.
We’re talking inside the chip, at this point. To connect the outside of the chip with the surrounding brain tissue, instead of the practically impossible task of manufacturing billions of tiny pins and wiring them to nearby synapses, the plan is instead to leave tiny bare metal spots (which, I imagine, will need to be of nanometer—billionths of a meter—scale) on the outside of the chip, where electrical signals can be exchanged with nearby synapses.
What’s this got to do with AI?
I was afraid you were going to ask; because to me the answer’s not all that clear. Obviously, the worlds of AI (especially connectionist AI—artificial neural networks) and the neurosciences are closely intertwined within Prof. Berger’s domain. I don’t mean that kind of relationship.
To me, this fascinating project holds out promise of answering the fundamental question of the nature of intelligence itself and the nature of "artificial" intelligence itself. We have to assume, for purposes of this discussion, that the Berger team will achieve its goals—and that is by no means a sure thing. But assuming it does succeed, what happens when 90%, 99%, and finally 100% of neural tissue (and why not include neural tissue in the rest of the body, not just the brain) is replaced by Berger chips? Does the transplantee cease being intelligent, becoming, instead, artificially intelligent? And is there a difference?
There are other issues, too. For instance, I remember hearing three or four years ago that Jaron Lanier, cyberpunk father of virtual reality technology, was spending somebody’s hard-earned cash on research to build a neural implant chip that would be used to connect our nervous system directly to a computer—to "jack in" directly to cyberspace, bypassing the keyboard, mouse, and monitor. There’s no reason why Prof. Berger could not build a "real" computer, complete with (shudder) Windows 95, into the neural chip.
Others, too, are into neural chip building. Dr. Hugo de Garis, staff, and international collaborators of Japan's Advanced Telecommunications Research (ATR) Institute are building an artificial brain ("CAM" Brain) with a billion artificial neurons, with evolved cellular automata-based neural circuits, to be completed by the year 2001. They say they already have 10 million neurons. It is unclear from their Web site whether ATR has the chip making facility, however.
There. I knew it had something to do with AI! Hope you enjoyed this piece, and be sure to (1) write to me if you know the answer, and (2) read Greengard’s excellent article in HotWired if you want more detail on the work of Prof. Berger and his colleagues.
Until
next week,
NEXT WEEK: The Field of AI. As you can see from the above, there’s more to AI than computer science. Much more. Just what are the aspects of human inquiry and endeavor that go into making up this field? We’ll review these aspects, and explain how the AI site at The Mining Company is uniquely organized to reflect them, and why it is your one-stop shop for AI news and information.