Random Access - Tuesday, June 29, 1999Who wrote this column?
When your PC gets smarter than youby Chris Gulker
Television, wrote Jules Verne at the turn of the last century, would be a great boon to people living in the future. He figured it would only take 1000 years to perfect.That the first televisions were demonstrated some 40 years later, should serve as a cautionary note to those in the field of prediction, though some might say TV is far from "perfected".
Nevertheless, one can draw a point from this tale. One of the world's great visionaries, living in a period that saw rapid invention and change, missed the mark by a huge margin. What gives?
Indeed as global communications improve, the rate of change speeds up. Inventions and ideas feed off each other, and the better the communications infrastructure, the more efficient the process of cross-polination.. A network's utility goes up at the rate of the square of its nodes, so the Internet alone is adding hundreds of millions of ways for ideas and idea-makers to inform each other.
Ergo, predictability's horizon is getting very short. Things we thought were way out there may, in fact, be right around the corner. Yet, if one were to predict that a computer would surpass 20th-Century human in intelligence by 2020, there would probably be few takers.
The reason is that one prediction that came up very short was that of artificial intelligence. Stanley Kubrick's landmark film 2001 featured a computer sufficiently artificially intelligent to mimic the kind of unenlightened self-interest that's all too human.
But a creature like Kubrick's HAL has yet to see "life", Artificial Intelligence, or AI as the field is often called, turned out to be a lot harder than originally thought. AI has staged a quiet comeback, with genetic and fuzzy and other AI-inspired algorithms at work daily in applications as diverse as autofocus cameras, automated telephone operators, automobile engines and stock exchange computers. Otherwise, there's precious little sign of intelligent artificial life on this particular planet, Scotty. Beam me up! Indeed, the thinking computer is still a distant dream. Authors like George Dyson say that machine intelligence is a Darwinian inevitability, utterly outside of human control. Given a substrate on which to exist, artificial live will evolve as surely as carbon-based life did.
Of course, some pundits would point to the glut of less-than-informed content on the Internet and say it proves that computer intelligence isn't going up, rather computer use is forcing human intelligence down.
But, glancing through a recent book by Ray Kurzweil called The Age of Spiritual Machines, I came across some figures that were very intriguing.
Kurzweil posits that AI's slow start is rooted in the difference between the computational ability of the brain and of current Silicon-based hardware. The brain, Kurzweil contends, is capable of some 20 million billion calculations per second, where as even a supercomputer like IBM's chess-playing Deep Blue can only manage about 10 trillion computations per second.
Interestingly enough, in 1988 Kurzweil predicted that a computer would beat a human grandmaster in chess in 1998. Deep Blue bested Kasparov in 1997, so Kurzweil's reckoning may not be all that bad. He was certainly closer to the mark than Verne's TV prediction, though even the more modest chess prediction came sooner, rather than later.
But, if you apply Moore's law going forward, and assume that answers will be found for the challenges facing the ever-shrinking features of modern chips, then even $1000 PCs will have brain-like capabilities in the not so distant future.
In fact Kurzweil predicts that by 2020 a $1000 PC will run calculations at the rate of 20 million billion. He further posits that enough RAM to contain the brain's store of 100 trillion synaptic strengths - some million billion bits - will come down from it's current cost of about $200 million (if purchased at my local Fry's outlet) to about $1000 by about 2020.
Extending the curve produces even more interesting results. The same $1000 that could buy a computer that would have the computational capacity of a human brain by 2020, would buy a machine equaling the brains of a small village by 2030, the whole population of the US by 2048 and that of a trillion brains by 2060. By 2099, a penny's worth of computing power will be a billion times greater than that mustered by all the projected 10 billion inhabitants of earth.
Now, it's certainly true that merely having sufficient horsepower is not, all by itself, enough to do the job. The invention of, for example, powerful engines did not immediately lead to devices like locomotives and airplanes, but it was an important first step.
It's also true that supercomputers, those hugely expensive denizens of university laboratories and shady government departments, are often 10 years ahead of their desktop counterparts. If that continues to be the case, then sometime as soon as 2010, a computer could exist that would rival a human brain's computational ability. One of IBM's proposed machines would hit 100 teraflops in 2003, only about 200 times less than a brain, and about equal to the average newspaper columnist.
Likewise, a lot of work needs to be done before raw computational power will translate into even rudimentary intelligence. But, I find myself wondering if the jump will come sooner than we expect, a la Monsieur Verne.
In which case you may begin to wonder who wrote this column, me or my iMac...
Random Access | www.gulker.com | Help/Info
editor@gulker.com This page was last built with Frontier on a Macintosh on Tue, Jun 29, 1999 at 7:57:41 AM.