Is The Singularity Near?

Writing in today's NYT.com, neuroscientists Sandra Aamodt and Sam Wang discount the premise of Ray Kurzweil's 2005 manifesto “The Singularity Is Near,”

Kurzweil caused quite a stir in technocratic circles by predicting that the point when computers would equal the thinking ability of humans was only a few decades away.

Actually, science fiction had been tackling that question for decades, but typically for the VC-computer industry, when one of their own posed the question, they deemed it worth considering. 

Picture is from Colossus: The Forbin Project and links to the movie's description on IMDB. In the 1970's Forbin Project supercomputers designed to protect humanity from nuclear holocaust decide we can't be trusted to run things and take over.

Now it is routine for characters on SciFi shows to ominously intone, "When the singularity occurs." 

Aanodt and Wang discount the idea, or at least its imminence, because power consumption will make it difficult to pack enough power into the size of a brain. A single brain, they estimate, holds about a petabyte of information, which equals a full third of all data on the Internet and a computer of that capacity would consume a full gigawatt, enough power to run Washington, D.C. .

That's an informative take, from the biological perspective, but to me, is beside the point. There are so many issues here, but just to tackle their approach ... 

The authors', if I understand them, seem stuck on the brain as a form factor. Putting that processing ability in a 3 lb gelatinous blob to emulate the brain is relevant only if you're building a small robot. A gigawatt would require a 1,000-acre solar farm, assuming your wanted your AI green. With the cost of storage rapidly going to zero, a petabyte is feasible. Since we only use a fraction, say 20%, of our brains then take that down to 200 terabytes. Both doable today. Granted neither will fit in a robot's head, but why is that important?

WIth distributed computing and widely-networked computers, whether the simple computing power can be packed into a small package doesn't seem a major issue; the computer doesn't even have to be in a single building. Having human-level computing power wouldn't require being packaged in humanoid form, even though androids and cyborgs make for better imagery. (If you're a geek, yes I'm assuming the Smoking Hairy Golfball problem doesn't surface because you don't need super calculating speed to be smart. Slow but logical would beat the U.S. Congress any day.)

Start by moving the logic, the software to a large computer. Then you have the Max Headroom model. Next, like Max, the TV character, the software can move from computer to computer. Finally, make our Max really flexible and break him up into modules that talk to each other. Then Max is a distributed computer; he has no physical home, but roams about the Internet; he's a creature of the cloud, as the Internet is referred to.

Models for distributed computing range from the benign such as the Seti project to the nefarious Conficker worm. Add the intelligence and self-awareness the authors are seeking and our amorphous, AI being could be distributed all over the Internet, moving redundant pieces itself from computer to computer, repairing, replacing or upgrading itself like a virus that re-writes itself. 

Need memory? Computer power? Borrow some from any computer in the world. MacAfee is no match for our AI being. The SciFi spins are interesting: Imagine generals saying, 'We have to blow the creature up" or "pull the computer's plug" while scientists explain that it doesn't exist in physical space but spans the cloud.

Personally, I don't understand the issues of how we would judge whether a computer thought like a human or not. Yes, there is the Turing Test and its counter the Chinese Room, but how do you really tell? (Both arguments strike me as over-intellectualized ways to say: If it quacks like a duck and walks like a duck ... ) A future computer might have much greater sheer storage and processing power than a human brain and yet not remotely process thought using what we recognize as human logic. Science fiction has explored this for decades: Will these future computers be friend or foe? Serve us or compete with us? Represent evolution to a higher form? Or just be tools? What happens when computers are better at designing their successors than humans; what will they create? Why are we arrogant enough to think their goal would merely be copying humans? How does humanity define itself when it is no longer the most intelligent species? It will take more than a few petabytes to figure that out. Or at least more than my solitary one.

Comments

[Page]

Another reason to develop alternative energy sources (or maybe not!). I'd hate to see The Terminator come to life...