Do you think that 50,000 years ago, as Cro-Magnons first began to filter into Europe, the resident Neanderthals scratched their bulky brows and wondered "gee, could these guys replace us?"
I strongly suspect that today, Homo sapiens is about to be replaced. But these next-gen sentients won't wander in from Africa, as the gracile Cro-Magnons did: They'll roll through the doors of artificial intelligence labs.
We're inventing our successors. And as a modest sideshow in this dramatic development, an IBM computer with the come-hither moniker "Watson" will be crossing swords with two accomplished humans this week on the popular television quiz program, Jeopardy.
Depending on your personal philosophies -- or maybe your susceptibility to the last blog post to cross your screen -- you're either betting on the vast, perfectly reliable memory banks and microsecond reaction times of Watson, or on the supple synapses of its human opponents, Ken Jennings and Brad Rutter. Most people, I assume, will bet on their own species.
But the big picture is this: It doesn't matter who wins the Jeopardy cash. This is a skirmish -- a Bull Run battle that won't change the outcome of the war. The machines are getting better, and despite the opinion you have of your kids, we're not.
In 1830, in a demonstration for the fledgling B&O Railroad, a small locomotive named Tom Thumb inadvertently got into a race with a horse -- and lost. This has become a famous story only because it was the last time that horses had a chance. Jeopardy may not be the last time that humans can hope to outwit machines, but that day will come, and it will probably come within this century.
Now a few points to note: This is not the first time that IBM computers have taken on protoplasmic adversaries. In 1996 and 1997, IBM machines played chess against Gary Kasparov, losing the first time, but winning (barely) the second. Of course, chess is an activity whose rigid rules allow mindless computation to eventually overwhelm skill and strategy.
Winning at Jeopardy requires more than computation, and indeed, it demands more than simply a phalanx of spun-up hard drives, loaded with facts. If the contest were merely a matter of information look-up, the odds on Watson would be short indeed. The real challenge for the computer will not be knowing the answer, but understanding the question.
Humans can unravel syntax and context far better than machines, at least so far. But it takes years to develop that skill in children, and computers are often asked to do so right out of the box. That's why many of the most interesting experiments in artificial intelligence involve devices that can interact and learn. Machines that can self-improve.
But games aside, some people -- actually, quite a few -- figure that no synthetic sentient will ever be able to do things humans take for granted, such as performing stand-up comedy, writing meaningful poetry, or simply knowing when it's safe to jaywalk. These skeptics are perfectly willing to admit that we might engineer a synthetic heart or kidney, but the three pound organ sitting between their ears -- well, that's sacred, man. Nothing could replicate its functionality, they aver.
All right, but consider the following. The processing power of computers is currently doubling every two years. Within a half-generation, the raw reckoning ability of your laptop will lap yours. In addition, human thought happens at the speed of neurons, with the signals ambling along at hundreds of feet per second, at most. Computer signals move at the speed of light, more than a million times faster.
The physicist Philip Morrison once described our brains as "slow-speed computers, operating in salt water." Harsh, but accurate.
In addition, your gray matter is boxed in by a brain case whose size is dictated by the sustainable body plan of a hominid. It weighs three pounds and operates at the same power level as a fridge light. A thinking machine, of course, needn't be hemmed in or energy starved.
Still and all, there's no denying the obvious: Neither Watson nor its chess-playing predecessors really think. Despite decades of work in artificial intelligence, no machine can unembarrassedly make that claim. But as the AI researchers say, don't confuse lack of success with lack of progress.
In a recent article in the Atlantic, Brian Christian describes how he recently took part in an annual contest between humans and computers (based on the well-known Turing test), and came out on top. In the battle between carbon and silicon sentience, he believes Homo sapiens will always be able to outfox the machines.
I'd like to think he's right, but somewhere deep in my soft, squishy cerebral cortex, I know he's whistling in the dark.
Clay Farris Naff: In the Watson Era, Will the Computer Be Servant, Master, or Savior?
Wray Herbert: Jeopardy! IBM Challenge Spotlights Cognitive Anchoring on National TV
Accumulating traits in evolution before the advent of genetic engineering was random and unconscious, only purposely executed with intelligent design after that. Perpetual cloning will complement on demand recombinant DNA and Synthetic Genomics technology for alien environments, as future robotically seeding spaceships colonise other astrogeological niches stemming from Earth. The day we create robots that are more intelligent than us they will see themselves as the latest appendage to an evolutionary long line of interconnected descen. In doing so will they perceive present human management of the Earth ecology as inefficient and cull a percentage of our numbers?
editor- Yea, just 'program' some 'emotion' in it...
I feel bad for any significant other in your life.
George- tell that to the Tibetans. Don't waste your time leaving pithy comments late at night to make yourself feel better. Try picking up the pipe instead. Then maybe your mind will open up just enough to realize science isn't a continuous buildup of knowledge but rather a particular thought-paradigm applied to a set of information (try reading The Structure of Scientific Revolutions). The current paradigms have reached the end of their usefulness due to a host of irreconcilable issues. One example: Einstein's relativity can't account for the existence of Larange points. Your seeming assumption that western science 'has it all figured out' and there's just a couple loose ends to tie up reveals how little you understand this fact.
Sequoia- I'm glad you admit you know nothing of quantum mechanics. I'm glad you can admit that even quantum physicists don't actually know what they're talking about. Go interact with your 'friends' on Facebook or something. Bra.
Unless, of course, you can explain to me how entangled particles react to one another faster than it would take a beam of light to travel between them? After all, it must be your overwhelming knowledge of quantum mechanics that triggered your comment and not
Elijah Rising
So, at 189 years old, I don't think it's bother me much.
But you're right about how machines will only get better and better as time goes on. If they get to be too good, perhaps we'll do what was done in Asimov's book, "Bicentennial Man", which was to make sure that when a machine was discovered to create brilliant works of art and to be capable of independent thought and desire, all further manufacture of that type of machine was stopped. Instead (in that book) people made machines that were capable of doing brilliantly in specializations such as surgery (and were capable of only that specialization). When I read about things like Google's driverless car, it reminds me of that book and what the future will hold.
Does anyone know if they input the question to Watson at the beginning or end of when it's read?
In my post above, I referred to Asimov's book, "The Bicentennial Man". He redid the book and published it in 1993 as "The Positronic Man", which is the version I read, not the earlier one.
If Messrs. Vinge and Kurzweil are to be believed, we'll do so in the next twenty years or so, but even if the modeling problem turns out to be way more complex than they assume (and there's research to suggest that might be the case), eventually Moore's Law will drag us all kicking and screaming to the edge of the abyss. Or the event horizon of the Singularity. Pick your favorite metaphor.
Still, I say, what's the big deal? Sure, Neanderthals are not around anymore, but recent studies suggest some of their DNA is ... in us. I'm not advocating any hanky panky with Rosie the Household Robot here, but what if the advances in other technologies just keep, well, advancing. The Other that you see in the not-yet-existing AI might turn out to be nothing more than us, with maybe a few improvements here and there.
With luck, improved eloquence in article comments will accompany the intelligence boost of Homo Sapiens Mark II. Still, while I'm still a Mark I, I'll do what I can and hope I've made a case for chillin' and seeing what comes next.
Schrodinger's cat was a concious observer.
How do you define intelligence? Or conciouness? How does one replicate something they can't explain?
AI might happen. Not at the rate Kurtzweileans think for sure.
Machine-enhanced human conciousness, on the other hand, is virtually already here.
A reality check: we need commodities, oil, food, cars, energy. Computers do very little for that. We are likely to starve, not afford cars, but have terrific tech toys. Don't overstate the value of computer productivity advances. I'd look at solar energy, much more important.
Unfortunately, there's not much to back up Mr. Shostak's beliefs. It's been over fifty years since the first computers began doing things no human - or team of humans - could ever achieve. Simple things like converging on solution sets that require millions of calculations (and are written by college freshmen every day) were suddenly possible. Were humans replaced? Hardly.
Similarly, Watson may hail a new era of human/software interaction, but little more. While it may be incredibly useful, remember that nobody has yet programmed anything that is even in the ballpark of what people think of as "artificial intelligence." If you're hopeful (or worried) that this device will soon take over human thinking functions, you're going to be disappointed (or gratified). It's an interface barrier, nothing more.
Finally, I'll point out that Mr Shostak's analogies are inept. Neurons can transmit data a lot faster than a few hundred feet per second, and the power analogy is just silly - watts is a measure that doesn't mean anything about intelligence (watts = joules/second = newtons*meter/second - no relation to intelligence or processing).
What do you expect from a guy who looks for aliens for a living?
In other words, the "Jeopardy" competition is rather like Lindbergh's crossing of the Atlantic, 84 years ago. It didn't change the future of aviation much ... but it did foretell it in a dramatic, and easily comprehensible fashion. T-Haight will consider Lindebergh another poorly chosen analogy, but I think that readers will get my point.
Everyone keeps trotting out Moore's Law as if it were Gospel. To argue the opposite, I would suggest that history shows us that each new technology starts, then changes, then changes more rapidly.
Then it matures. It keeps advancing, but to an asymptotic limit.
For example, look at airplane airframes and engines and the history of their development. The Wright brothers' airplanes were barely faster than a running man. By the end of WWI, planes could fly between 100-150mph. By WWI, twenty years later, it was 400mph. By the end of that, 600mph, and less than ten years later, 1500mph.
Now, compare the general design configuration and top speed of the F4 Phantom of 1958 with the fastest operational fighter today. Compare the the general design configuration and top speed of the 707 with the not-even-certified-yet 787. Commercial and military aircraft performance is not that higher than it was 50 years ago!
You can see similar limits reached with other technologies; trains, ships, cars.
Sure, all of these things have gotten better, but the rate of progress has slowed.
So, you can't expect simply to extrapolate the current line on the chart and predict from that.