Via Insty, did the Singularity just happen on Jeopardy? Good question, but here’s the bit that caught my eye:
Great story from CNN. The human champion describes Watson as a Terminator. It never gets tired, bored, has stage fright, gets cocky, or intimidated. It just keeps coming… answering general knowledge questions better than the best Jeopardy players on the planet. Scary right?
I thought back immediately to 1997, and a column Charles Krauthammer wrote on the occasion of chess grand master Garry Kasparov’s loss to IBM’s Deep Blue:
You might think it is a little early for fear. Well, Garry Kasparov doesn’t think so. “I’m not afraid to admit that I’m afraid”, said perhaps the most fearless player in the history of chess when asked about his tentative play. When it was all over, he confessed why: “I’m a human being, you know. When I see something thing that is well beyond my understanding, I’m scared.”
OK, I understand that one really shouldn’t draw conclusions from two data points. But we can certainly speculate!
In both cases, the computer seemed …unhuman… to its human opponent. That much we can attribute to the persons, and not the machine. But — Kasparov didn’t actually use the word “Terminator,” but you could slip it into his statement above and it would fit just fine. The machine just keeps… coming… at you.
The machine is remorseless, it seeks to win. And then it does win. That’s something we need to keep in mind, because right now, the cold science facts look eerily like the dystopian science fiction.
Listen, and understand. That terminator is out there. It can’t be bargained with. It can’t be reasoned with. It doesn’t feel pity, or remorse, or fear. And it absolutely will not stop, ever, until you lose on Jeopardy.
So, the question becomes now, does Weird Al update his song for this? http://www.youtube.com/watch?v=BvUZijEuNDQ
Heck, he didn’t update it for Alex Trebek, why should he do it for IBM’s tin can?
Yes – I remember how smug my “Speak N Spell” sounded back in the early 80s when I misspelled a word. Metal (well, mostly plastic) bastards!
There’s actually a group dedicated to making sure that the first greater-than-human intelligence is friendly, and not a terminator: http://singinst.org.
Kasparov has been less afraid more recently.
How soon will we see Advanced Jeopardy analogous to Advanced Chess?
The Terminator has been among us for some time, in the disguise of industrial robots. Over the last 20 to 30 years, industrial output has doubled, but so-called worker productivity (more robots) has tripled. This has lead to substantial job losses in manufacturing. The jobs didn’t go to Mexico or China, both of which are also losing manufacturing jobs. The robots took them.
Where is Summer Glau when you need her?
She’s off fighting the genetically-engineered zombie horde.
No one seems to address what I think is a pretty major point in Watson’s success, the buzzer.
Every Jeopardy contestant says that much of the win/lose in the game comes from how good your buzzer timing is. Whatever Watson’s knowledge skills, I’m sure it can model EXACTLY when to buzz in, which if you are bad, makes you good, and if you are good, makes you great.
Especially if your opponents are nervous over playing a machine.
Pretty much this. It’s all about the buzzer. Clearly the two humans know the answer just as well, they’re just not as quick.
If I was a law student, I’d be shaking in my Nikes because what Watson does for trivia can easily be applied to a large percentage of what low-level lawyers do for those oh-so-expensive large legal partnerships.
I was on Jeopardy in the mid-1990′s and Doug is right: it’s all about the timing. I’d been practicing at home clicking in with a pen and timing off of Alex’s voice. In the studio there are two lights that come on when it’s clear to buzz in and at first I tried to adjust my timing to react to those — too late. When I switched back to cuing off Alex, I started running categories like mad.
My stint (I won two games and then lost in Final Jeopardy) was when the show was shot at Sony Pictures Television in Culver City, the old Columbia lot. One of the goals was to last through the morning shoot and so get to have lunch at the commissary. I managed to do so. Besides the fun of being on the lot and seeing how the pros did it (I produced videos for education at that time), it was fun to find out how much I had in common with my fellow contestants and how similarly our minds worked.
Watson can have “his” wins. He’ll never have the human Jeopardy experience.
And John Henry was a steel drivin’ man.
Yes, AI is relentless and man is weak.
But recall that we evolved as an approximating solution to the “iterated prisoners’ dilemma.” Simply put, that means there is no analytic solution about when to cooperate, when to cheat. We learned to read the “other” only after great culling by the hand of Darwin.
AI will act tirelessly towards the goals with which it was programmed. But then what? Like an agile matador, step aside, let the AI get there, and watch it turn itself off.
( Interesting if flawed and pessimistic treatment at http://www.acceleratingfuture.com/michael/blog/2011/01/yes-the-singularity-is-the-biggest-threat-to-humanity/ )
Robert Arvanitis;
Since I’m the guy who wrote this critical response to the Michael Anissimov blog post you link, I think it only fair to point out that, while I concur with your general judgement, Michael actually said that once AI can improve their own design capabilities faster than humans can evolve their performance (intellectually or otherwise), then AI/AGI entities become an existential threat to humans if they should come to regard humans as competition rather than companion. Given the fundamental competitive nature of current human existence, precisely how we go about avoiding the former is, in Michaels opinion, a matter of extreme importance in this early stage of AI development.
I disagree with him largely because I think this is a worry better addressed as part of the developmental process rather than preemptively via conditioning or foundational software structure. Opinion’s differ obviously, but I do think the concern rates better than casual dismissal.
Will Brown
Yes. Take your point about self-improving AI as the challenge posited by Anissimov. In fact, your piece is how I first found Anssimov.
This being the comment section of the often-lighthearted Vodka Pundit (and thanks to our host for even making this dialogue possible!), I’d note that one man’s “casual dismissal” is another’s sprezzatura.
In a more fulsome mode, perhaps I can expand on the rather succinct initial post:
“John Henry…” Luddites fear change. The proper view is that technology is enabling, not threatening. Few actually manufacture today, even fewer farm. Indeed, the very nature of work is changing radically. See for example today’s Wall Street Journal //http://online.wsj.com/article/SB10001424052748703439504576116340050218236.html?mod=djemEditorialPage_h
“Relentless…” Like a thermostat. The furnace runs until the room temperature rises. Even if the windows are open…
“…approximating solution to the iterated prisoners’ dilemma…” For believers, we have a purpose beyond the engineering.
For non-believers, the engineering solution is a simple model. Imagine a table top with an assortment of marbles and dice. Tilt the table just a little. The marbles roll off and the dice remain. Shall we say the dice “survived?” Or perhaps that the marbles “escaped?” Either way there is a selection process independent of any interpretation that may be imposed. That said, we in present version are here and other versions are not.
For a putative self-improving AI, I note that the verb “improve” begs the question – improvement along what dimension? By what metric? Indeed, to what end?!
“…agile matador…” Once we infer on what such an AI may be optimizing, simply agree. And God forbid we EVER ask such an AI to care about humans. Therein lies all the mischief. In all our iterations of forethought known as science fiction, the disaster arises when (Forbin, Skynet, HAL, you name it) decides to protect us from ourselves.
Come to think of it, isn’t that the liberal agenda?
Whether Watson has a buzzer advantage or not (that is simply about winning the game), it is a great engineering accomplishment that a computer can even answer many of the questions. We will continue to knock on the door of the Singularity. Computers will do more and more, and we will continue to be pleased with how much better they can make our lives and moan about how much stress the technology also creates for us.
I think basic optimism about these technological developments is warranted. However, I don’t think unbridled cheerleading is. If the proponents of the Singularity are correct about the emergence of computing power, then we are certainly not smart enough to understand the consequences of the advancement of computers that have emergent properties of independent thought with far greater processing power than humans can ever posses.
When the category becomes “The Full Monty” I wonder if Watson will still be able to answer correctly.
Unless of course his database includes all the Monty Python episodes and movies.
I see the $100 answer, “Blue. No yel– Auuuuuuuugh!!” and wonder if the humans could answer “What is your favorite color?” while laughing.
The buzzer is a red herring. Watson was spoon-fed the “answers” in easy [for it] data packets. No cueing off, or being misled by, Alex’s nuance, no reading the boxes, no audio or video clips to disambiguate. They didn’t make him read. Sure, a computer can do that, but for some reason they chose not to. Nanoseconds were at stake, and IBM wasn’t going to take that chance on the front end.
Is there a way to make Watson understand where Deep Blue is now? No? Thought not.
Some singularity.
go ahead and laugh but i’m joining the second-place winner in welcoming our new computer overlords.
I want to see Watson take on Sid Meier in a game of Civilization V.
Yawn.
Watson is nothing but a data-mining program combined with syntax analysis. It parses the host’s statement, then uses that parsed information to search its database. Others have called it a web search engine on steroids. Zork did the same thing (albeit at a much less complex level) twenty-five years ago. On a 4.77Mz PC.
In either case, it is in no way intelligent, nor was Deep Blue. The latter was just a computer optimized for branch analysis. It could do that like gangbusters, but couldn’t balance a checkbook worth a darn; the code wasn’t there.
I’ll be worried when a computer can tell me who put the “bop” in the “bop-be-bop-be-bop.” I’ll be scared when it can also tell me who put the “ram” in the “ramma lamma ding dong.”
Just in case, don’t automate the pod bay doors.
For now I think the first worry is bots that can accomplish actions *without* having features of higher intelligence.
What happens when criminals are able to instruct their bots to go out and collect money for them and leave the money at a predetermined, hidden spot – using any means necessary including violence (or perhaps purposefully committing violence)? Or are we going to license bot makers, forbid hacking them for our purposes, so no do-it-yourselfs would be allowed with robotics and only large licensed companies could participate? Will I need to have protective bots with me at all times, a sort of arms race?
Easy to forget in all the technobabble that both Kasparov and the Jeopardy champs were playing against several people simultaneously in “digital” form. No matter whether standing shoulder to shoulder or embodied in a computer – the sum of the group intelligence (memory or recognition) will be quicker to a solution than a single person ie. “brainstorming”