What magical trick makes us intelligent? The trick is that there is no trick. The power of intelligence stems from our vast diversity, not from any single, perfect principle. – Marvin Minsky, The Society of Mind, p. 308At the time this was an astounding theory, almost heresy. Many AI researchers were still trying to build larger and larger "supercomputers" and more and more complex operational programs, but running out of steam in that direction. There are some very real limits in how much you can cram into one "computer", based on the speed of light, among other things, that we are unlikely to change.
Minsky, and others, finally realized that the key was not to make super-Rambo single computers, but to build networks of much smaller, simpler computers. And the key was not to have one, immense, incredibly complicated program of logic and reasoning, but to have instead hundreds or thousands or more of very simple rules that just worked together. That removed the limits, and the power of "computers" took off again.
Except, to paraphrase John Gall's law, when you scale up something, the larger thing of the same name is not really the same thing as the smaller thing of the same name. What was still called "a computer" is really many computers in the same box.
Initially, this took simple forms in commercial computers. Apple's "computer" was something like 10 computers in a box, one handling sound, one handling video, one handling communications, and one doing the coordination and "thinking." Soon we started noticing "computers" with "dual processors" or "quad processors", with multiple separate computers on chips interacting and doing the thinking together.
IBM's new "supercomputer" has something like 860,000 separate computers-on-chips all talking to each other. And the trend in distributed "grid" computing is to link millions of computers together to "think" about a problem together, like the ones all working on the Search for Extra-terrestrial intelligence (SETI), or recently, a search for a lost airplane.
So, even without VM, or virtual machines, the line between "one" and "many" in computing had to be crossed for the field to advance. Many-acting-like-one was clearly the best and most powerful solution. It "scaled-up" without being immensely complicated to build parts for or to "program". If you didn't have enough "power", more computers could just come over and join the party.
Now, finally, with Virtual Machines, the edge of consciousness of a "machine" no longer had to correspond to the edge of the physical "machine". IBM came out with the VM370 operating system for mainframes back in the 1970's, mostly so each user's "virtual machine" on one piece of hardware could think it was the only program running on the box. The problem with sharing the space and having multiple programs running "at once" on a mainframe computer with multiple "users" was that if one program "crashed", it could often take down the entire box. No one liked that.
With Virtual Machines, each program was deceived into thinking it had the whole box to itself, which it did for very small slices of time. The programs became essentially invisible to each other, each one seeing itself as the solitary occupant, even though many of them were running on the same hardware at the same time. It was very successful. Again, this spread to the world of personal computers -- which are now each much more powerful than IBM's mainframes were in 1970. With VMware, as I discussed in an earlier post, you can have essentially a Windows computer, a Macintosh, and an Linux computer all running on the same box and invisible to each other, mostly.
But on the server end, you could also slice and dice, and virtual machines didn't care how many physical machines they occupied - part of one, one, or many. They became decoupled, so a room full of left over computers could become, instead, and act like, one big computer, or several smaller computers that crossed hardware lines. If one box died, things just ran more slowly, as the consciousness shifted slightly in space, until the box could be replaced.
All of these are perfectly good models of intelligence that actually are used every day in the products that run all around us. Even laptops come standard now with "dual processors".
Two working as one was a pretty popular model for "parents" for "babies" in the biological world as well. The coupling between the two separate people in marriage was described in the Christian Bible as "two becoming one flesh" and, from an information processing and control loop view, that was literally true. Many processes that used to "run" in one person, now spanned two people.
So, as my last post suggested, it is not clear how far this design has spread to the rest of the physical thingie we call human beings or life on earth, or, with the Gaia model, Life on earth or maybe even LIFE on Earth, capitalizing the parts that are separately "alive" and conscious.
Now that we've proven that many can act as "one", in this physical world, using computers as building blocks, and that, as Minsky said in the quote above, the power comes from "diversity", the question comes as to the extent to which multiple people can "work together" as "one".
How tightly can we be coupled together in a dynamic "team", and are the equations behind that behavior the same flavor as those behind massive grid computing, that is , is this coupling an unlimited process upwards?
All of this suggests, again, that the "team" thingie we are building out of ourselves has a "life of its own" and is, on it's own level, a separate consciousness, and, ultimately other emergent properties such as a separate awareness of itself and surroundings, and a separate volition.
Here we have crossed into xenobiology, or Theoretical Biology, and are asking if one "self", like VM, can cross multiple "bodies". And, for that matter, can "we", what we thought of as our selves, simultaneously be an emergent property of the ten trillion cells that make up our body, and part of a larger emergent consciousness that we are only dimly aware of indirectly?
There is no reason in control system theory, or physics, or information theory, or artificial intelligence, or the field called "artificial life" for that not to happen. It's a rather disturbing prospect for two groups -- researchers, who have to start with a fresh piece of paper analyzing the active agents behind what's causing what, and most ethics and religions, that may have an issue with what defines "a person" or "life", and certainly have a stake in certain definitions.
So, let me say that nothing I just described interferes with classical religious concepts of "self", that I can see. These are descriptions of "HOW" life operates, not, on a deeper level, of "WHAT" is behind it. And everything in real life has multiple meanings and roles, so just because we've found one meaning of something doesn't mean we've fully described it. Humans are composed of elemental atoms, but are far more than just a heap of atoms. That doesn't change the way we live our lives.
But, if humans are interlocked on the behavioral or spiritual level, it does change the way we should be thinking about health, teamwork, medicine, law, etc.
So, this becomes a testable hypothesis, or a model, that by definition over-simplifies the description of something by trying to capture simply the essence of some facet of it. Many models can fit reality simultaneously, which doesn't "prove" anything or conflict. Different models have different purposes, and give good results for some uses, and terrible results for other purposes when stretched beyond their limits.
Statisticians and scientists know the rule "All models are wrong but some models are useful."
Whether it is ultimately "right" or not, does the model of consciousness and control being decoupled or decouple-able from individuals and crossing the gap between them lead to other things we could predict or test or look for that we would have never thought to look for before? If so, and it's relatively inexpensive, we should go look and see if those things are present and if they "explain" of "fit" areas where scientific theories so far haven't been very successful at predicting consequences of interventions.
One prediction I mention above, and discussed in prior posts, is that "teamwork" or interactions, are basically unbounded upwards. There is no limit, in the scale we operate in, to how strong a team could be made, or how large a team could be made, if this model is correct. What we've seen to date as successful "teams" is just a glimpse into what is possible when human beings interact over time and take down the barriers between them.
As I discussed before, many times, this would be really good news, because our social problems have become more complex than any individual human can even begin to comprehend, and we need to change gears to some more powerful way of dealing with, basically, the problems we create for ourselves.
Here's the picture I made that I keep on posting, in the hope that it will sink it, orignally published as "Houston, we have another problem."
What I'm trying to convey here is that, like our computers, we need to not only start with multidisciplinary research and teams, we need to crash through those starting primitive steps, take advantage of instant messenger, Twitter and other hive-consciousness tools, and consciously put our minds and spirits together in a much more organic whole than we've imagined doing before, because it appears possible, and because we need the result.
We need the outcome. We need to synthesize "Meta-being" teams that seamlessly act as one and that start taking on the problems around us that are way too big for humans to solve, and solving them for us.
I have to get in a jab here and point out that "seamlessly acting as one" does not describe the behavior of various ways we have assembled people into governing bodies, such as the US Congress, that can't seem to agree on much of anything anymore. That model of collaboration or competition or whatever the heck it is seems to have truly run out of steam, now that our social problems have become so complex and tangled and inter-related so that you can no longer "prioritize" or "horse-trade" and "just" solve a few high-profile issues and call that adequate.
All 200 holes in the bottom of the boat need to be dealt with, or the 12 we dealt with won't matter. The entire algorithm of "prioritizing" to "focus on the big issues" has run out of steam, and as the graph illustrates, that's not coming back again if we just wait for it.
We've given ourselves no choice but to face the kicking and screaming and figure out how to work together and think together, in real-time, in a way a thousand times more powerful than we've ever done before.
It's fascinating, exciting, and scary, all at once. But it's not optional reading, and there will be a test on the material. It's coming up, and it's pass-fail for all of us.
Like Toyota figuring out at last that auto model changeovers could be done in 6 minutes, not 6 weeks, or biofeedback doing in 5 minutes what Zen masters took decades to accomplish, we need to look at how technology can interact with society to make this serious industrial-grade teamwork thing work.
Toyota has demonstrated that a car company built along massive cooperation and holistic awareness can run circles around established car companies. But they took 50 years to get there, and we don't have 50 more years at the rate things are breaking down now.
So, we need some focused research efforts on ways to get this working in 5 weeks, not 50 years, somehow. Nothing else seems to be working.
I've come at this in a somewhat soft way, but the question isn't really a mystical question requiring incense and eerie music in the background, holding hands and singing Kum-ba-ya.
This is a very practice, very straight-forward computational question, of the type that "evolutionary computing" people deal with regularly. What very simple set of guidelines and rules, if followed by everyone, would result in the desired "emergent" behavior we are after?
This is the same problem IBM faces with its supercomputer and it's million separate computers it needs to operate as one. No one can "program" such a beast -- we can only search for guidelines for behavior and very simple rules that, when multiplied together and interacting, result in the effect we're after. Artificial Intelligence has about 25 years experience now tackling such problems, as does the field "Artificial Life" or "a-life", of the type studied at the Santa Fe Institute.
John Holland was still teaching the course in Complex Adaptive Systems when I was an AI grad student at the University of Michigan, and he was instrumental in getting the Santa Fe Institute going, and I looked into that work then. It's amazing. It's where we need to go.
The problem could be conceptualized as one of extreme "technology-mediated collaboration" as well, a specialty of the University of Michigan School of Information.
Given the problems of trying to collaborate across 24 time zones and different cultures, with deep contexts, I do think that maybe the way to do this is with some sort of virtual reality such as Linden Labs "Second Life Grid", so that people have at least one constant, reliable, safe, shared deep context they can get together and work in, regardless where their physical bodies are located at that moment. That gives a home-base people can invest in, regardless whether they have a stable job or tenure, or their dot-com is about to fold, that can learn and discover and evolve the coordination and collaboration tools that really help this fly.
The ability to embed your audience in a 3-D living context that dynamically shifts to provide the correct meaning to the content of your talk is intriguing to me, and something Second Life artists and musicians are doing already. This is "image processing" bandwidth brought to the problem of expressing complex concepts in accessible ways, dense with interaction and animation that persist long after the lecturer has left for the day, and filled with people from around the globe who shared an interest in staying to delve into the subject more deeply, who might be only one or two per company or country, but add up to a critical mass with instant global travel to a shared virtual exploratorium and university.
Virtual reality could be the missing "VM" or virtual machine catalyst to making this fly, a universal language and translator that manages both content and context in high-bandwidth front-channels and unlimited back-channels that our very clever brains, or those of our children, will figure out how to master and fly effortlessly in an amazingly short period of time.
On the other end of the age spectrum, such a world would let us start recruiting a larger fraction of senior citizens who are still bright, with deep experience but little mobility or budget or tolerance for what travel involves these days, and get them interacting as part of the solution, instead of retiring to sunny spots and taking their expertise with them. There would be immense public health benefits to that step alone, as connectivity to a larger social world is about the single most powerful thing one can do to a senior citizen to improve their "biomedical" prognosis. "Second Life" could have adaptive technology for vision or sound built in easily, and ability to "replay" a conversation could slow it down if need be to get the content correctly understood before proceeding without exhausting the patience of a conversational partner.
No comments:
Post a Comment