Good.is: Building the “Everything Machine” Monday, Nov 30 2009
futurism and nanotechnology and singularity 11:59 am
My latest article (#3) in the Singularity series on Good.is is up, a piece that describes exponential manufacturing titled Building “The Everything Machine”. Meanwhile, Roko’s article on “Why the Fuss About Intelligence?” is the 2nd most discussed article on the site in the last week. I will repost my article here for further discussion, but I also encourage you to register on the site and comment there. Here it is:
Building the “Everything Machine”
Nanotechnology and exponential manufacturing could help us make whatever humanity needs, atom by atom.
Part three in a GOOD miniseries on the singularity by Michael Anissimov and Roko. New posts every Monday from November 16 to January 23.
Last week, Roko talked about how human intelligence made civilization possible, and how genuinely smarter-than-human intelligence—what some call “superintelligence”—would change everything, by magnifying nearly all of our capabilities.
It is important to note that organizations or countries are not smarter-than-human intelligences any more than a tribe of chimps is a smarter-than-chimp intelligence. We are talking about thinkers with fundamentally improved cognitive architectures, either through brain-computer interfacing or the creation of creative, flexible, brilliant artificial intelligence. Engineered intelligences with greater memory, creativity, pattern-matching capabilities, decision-making skills, self-transparency, and self-modification abilities.
This category of enhanced intelligences may not be as far away as you think. MIT scientists are already working on optically-triggered brain-computer interfaces that could link up many thousands of neurons to computers in the near future. Ed Boyden, who works at the MIT Media Lab, has called for the creation of an “exocortex” that assists our natural brains with an external, artificial cognitive assistant, also called a “co-processor.” We may even discover drugs or gene therapies that qualitatively improve intelligence by increasing the speed at which neurons can communicate, as was recently done with a rat, Hobbie-J.
When discussions of superintelligence crop up, a common question that is asked is, “okay, these entities are smarter-than-human, but wouldn’t they still be very limited by their environment and the intelligence of humans they have to work with?” Couldn’t we just pull the plug on a very clever artificial intelligence? Wouldn’t an enhanced human intelligence be limited by the slower people around it?
Not necessarily. One way superintelligent entities could leapfrog human industrial infrastructure and communication time lag would be by creating self-replicating manufacturing units, which might be based on synthetic biology or just sophisticated robotics. There already exists a self-replicating manufacturing unit today: RepRap (short for Replicating Rapid-prototyper), developed by a team at the University of Bath in Britain. It just requires human assistance for assembly—from there, the machine can print out practically all of its own parts, except for a few standard parts like computer chips. Completely autonomous self-replication is on the horizon.
The ultimate self-replicating manufacturing unit would be based on nanoscale fabrication—the rapid manipulation of individual atoms to build large products from raw materials. In 1959, the legendary physicist Richard Feynman gave a talk to the American Physical Society called “There’s Plenty of Room at the Bottom.” During the talk, he said “The principles of physics, as far as I can see, do not speak against the possibility of maneuvering things atom by atom.” Since Feynman’s talk, we have made leaps and bounds towards the goal of bottom-up manufacturing, building tiny robotic arms that can manipulate single atoms, molecular switches, gears, “nanocars,” even a nanoscale walking biped.
If we could design and fabricate the appropriate nanoscale machines and put them into a system capable of building all its own parts, we’d have something called a nanofactory, or to put it another way, an “everything machine.” The earliest nanofactories might only build products out of a couple types of atoms, say carbon and hydrogen, but they would have a tremendous impact because they would be automated by necessity, could self-replicate, and would be capable of building almost any chemically stable structure (as long as it used atoms the machine could handle) with atomic precision. Powered by the Sun and using purified natural gas for feedstock molecules, these nanofactories could quickly and easily build huge numbers of residences, greenhouses, appliances, medical equipment, water purification equipment, and much more, at a cost thousands of times lower than the manufacturing technology of today.
Humans are making progress towards nanofactories today, but I’ll bet that smarter-than-human intelligences could make much more rapid progress. In fact, it’s possible that the most direct route to nanofactories is through smarter-than-human intelligence.
And if you combine a smarter-than-human intelligence with self-replication and nanoscale production, it’s difficult to put a limit on how quickly superintelligence could change the world.
Michael Anissimov is a futurist and evangelist for friendly artificial intelligence. He writes a Technorati Top 100 Science blog, Accelerating Future. Michael currently serves as Media Director for the Singularity Institute for Artificial Intelligence (SIAI) and is a co-organizer of the annual Singularity Summit.
Michael:
I’ve been reading your excellent blog for years and fundamentally agree with all of your ideas, except the following:
“It is important to note that organizations or countries are not smarter-than-human intelligences any more than a tribe of chimps is a smarter-than-chimp intelligence.”
Anders Sandberg argues the opposite point here:
http://machineslikeus.com/news/anders-sandberg-cloud-superintelligence
Several months ago, I made the same argument in your comments: http://www.acceleratingfuture.com/michael/blog/2009/05/the-danger-of-ai-part-2/ Greg Egan has also argued along similar lines. http://metamagician3000.blogspot.com/2008/04/transhumanism-still-at-crossroads.html
Serious thinkers are divided on this issue, with SIAI affiliated scholars taking one position, and other thinkers (Robin Hanson, J. Stors Hall, Kurzweil, Brin, possibly Jeff Hawkins) having the opposite view. This is clearly not a matter of explaining a new concept to someone who is unfamiliar with it. I’ve read LOGI, Omohundro’s paper, Legg’s PhD thesis, EY’s Existential Risks paper and lots of OB posts, and I suspect others have as well.
If we apply the AIXI intelligence test, from Legg’s PhD thesis, to a blackbox consisting of a single human brain with a functioning corpus callosum, we’ll find that it is more intelligent than either brain hemisphere alone would be (ie. where the corpus callosum is severed). Whether you want to call a single brain a single intelligence, or the collective intelligence of two hemispheres (or of many brain modules), is a matter of semantics. If we apply it to a blackbox consisting of a society of well coordinated humans, we’ll find that it is more intelligent than a single human. I’ve always found it interesting that Hayek independently discovered Hebbian learning in the brain as well as the power of the collective intelligence of markets. Chimps lack the flexible general intelligence and language skills to form social groups with a considerable degree of collective intelligence, even if there are a billion chimps. On the other hand, compare a well coordinated society of a billion humans to one of only a hundred humans. An SI would differ in efficiency and speed, which will eventually add up to a difference in kind, but perhaps not as quickly as the hard ascent scenario envisions.
If you think this debate is worth pursuing, would you consider linking to Anders’ post and commenting on it. I would also like to know what the rest of your thoughtful readers think.
Thank you for the opportunity to post here.
Gus K.
I think this is an important point. Regardless of whether organizations can properly be considered more intelligent than individual humans, they indisputably have vast advantages in effectiveness in many situation. The narrative of superhuman intelligence necessarily dominating puny humans obscures interesting scenarios of developing AGIs held in check by networks of people and technology. Such futures might not be exciting as a singleton suddenly bursting onto the scene, but the deserve careful thought.
Thanks Gus. I appreciate your comment and have been thinking about it. I never saw the picture that Sandberg, Egan, Hawkins, Hall, and yourself all have a position on this distinctively different than the SIAI-associated writers.
Since I feel that this contention point is extremely important, I will be thinking about it carefully, and seeing if I can use scientific arguments to argue one way or the other. There are interesting arguments both for and against.
Benjamin, I just don’t rate those types of scenarios as very likely, so I’d rather explore the subtleties of hard takeoff scenarios. I am always interested in hearing what others have to say on the former type of scenario, though.
A case could be made that a football crowd is stupider than most humans – but in the case of most companies, they seem just obviously smarter than humans, due to the well-known phenomenon of “collective intelligence”. Is this really a point of contention? What counter-argument could there possibly be?
It is more conventional to say that companies – while smarter than humans – are not yet *so* much smarter as to qualify as superintelligences.
Tim, you could say that more people in the company simply lets them do more but does not let them achieve very many complex tasks that any group of, say, four of them couldn’t do on their own. The benefits of aggregation might be so low that they are useless on huge classes of complex problems. A room full of arithmetic students cannot cooperate to solve a calculus problem, for instance.
Ants cooperate too, but is a colony of ants “smarter” than a single ant? Maybe only in a very abstract sense. Humans might have better aggregation properties, but a Devil’s Advocate could say “not by much”.
Tramadol….
Tramadol 93. Tramadol tablets. Buy tramadol index. Taking tramadol while pregnant. Tramadol. Tramadol hcl 50mg….
What is xanax….
Pharmacy order generic xanax. Xanax….