Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[zchris13]

A. The Singularity is a crock of shit, the most it'll be able to do is catalogue data and shitpost, which I guess is useful. But computers/AI are only as intelligent as the data we give them access to.
B. Cthulhu, you just made apocalyptic climate change sound like an episode of Doctor Who.
C. One day, the Yellowstone supervolcano will erupt and kill us all and there is literally nothing we can do about that.

[Angle]

You're both assuming that the singularity will take the AI SuperIntelligence Explosion route. There are other ways it could happen. The other big one would be the human enhancement route, where human beings enhance their capabilities repeatedly. I think we actually have several bay12ers working on that one.

[Baffler]

What is it going to do then? "We'll create AI and it'll fix all our problems!" is repeated often enough, but the more I hear people talk about it the more it just sounds like a pipe dream. Messianic thinking from a people with no God.

[Angle]

Well, assuming we could create a true superintelligent AI and get it to do what we want it to, any of a billion things. Invent more or less anything we want, answer any question, etc. I think that creating such a thing is going to be more difficult than most people think, and I worry a great deal about getting it to do what we want it to do. That's why I'd rather pursue the Human Enhancement approach.

[Baffler]

Even then, I'm not sure it could do much better than reinvent on its own all of the solutions to problems that are already being tossed around. It's too simplistic to think there's one solution to the broad sorts of problems like poverty and climate change it seems to be expected to 'solve'. The absolute best I can see it doing is coming up with a long list of comprehensive options (it can do that because it can keep track of more information by itself) for people to fiddle with and argue between. It'd be an improvement, but not all that much of one.

And all that's assuming it's even worth listening to. The idea that it needs to learn implies that someone has to teach it, and there's not any way I can imagine to do that that isn't basically guaranteed to produce wide blind spots, enormous biases, or both.

[Cthulhu]

What is it going to do then? "We'll create AI and it'll fix all our problems!" is repeated often enough, but the more I hear people talk about it the more it just sounds like a pipe dream. Messianic thinking from a people with no God.

That's exactly what it is.  99% of singularity talk I've seen you could replace "AI singularity" with "second coming" without changing the meaning at all.  They also have about the same odds of happening.

[Angle]

Uh, no? I don't think it'll happen that easily or that fast, and I'm not entirely sure it'll work out well for us, but I don't think such predictions are that far out there.

[Zrk2]

I still don't buy that a true AI is even possible. Does anyone have a link to a solid essay that might prove me wrong?

[Angle]

What do you mean by AI? We've done most of the important bits already. Hell, we were doing them before I was born: http://aliciapatterson.org/stories/eurisko-computer-mind-its-own

[Reelya]

I still don't buy that a true AI is even possible. Does anyone have a link to a solid essay that might prove me wrong?

True AI? What's a person then? Unless you believe in mystical soul stuff , "True AI" is generated every day from nothing. It's in the same sphere as flight. One might have said heavier than air flight is physically impossible, yet we see birds flying and they are heavier than air. The proof that it is possible, is the fact that it already exists. You don't need a solid essay to point out that if something already exists, then it is, by definition, in the realm of the possible.

EDIT: And even if conscious AI isn't possible for us to build. Guess what? The usefulness of an AI to everyone else doesn't require it to be conscious, just to be able to read information and make reasoned decisions on that data. The singularity is basically a given once we make a bot that can read and absorb college-level textbooks. At that point, you just feed all the information into it in a huge infodump. And a thing that reads books doesn't need to be conscious to be useful.

So, we can definitely narrow down stuff needed for an AI "singularity". Consciousness is definitely not a required trait. What is required is something that can correctly interpret not only syntax but semantic meaning of text. And those projects are definitely happening. Cyc is one example of a long running project aimed at this, and it can correctly apply common sense in analysing quite a few types of data. Branches of the Cyc project are used to detect financial fraud and computer security flaws. In these cases, the software does not have to be programmed at a low level specifically for the domain. You tell it basic things about the problem domain in English, then it works out the missing contexts itself. One example application of Cyc is feeding a database straight into it, without it knowing anything about the problem domain. The thing reads the column titles then applies semantic analysis to work out what the data "means", converting the entire table into logical predicates (with correct semantic links inferred). It's able to detect anomalies in this type of data without being told anything about the problem domain.

[Jelle]

What do you mean by AI? We've done most of the important bits already. Hell, we were doing them before I was born: http://aliciapatterson.org/stories/eurisko-computer-mind-its-own

Games are logical problems to be solved, making a program to solve it is not a big a deal. Especially when the rules are neatly defined such as in a game, throwing all those rules (or predicates) into prolog for example should let you solve it easily enough. While the example is impressive for its sheer processing power, it's no AI.

[Reelya]

Not for N complete problems. Read the article again. The sheer number of variables make brute force search impossible. You can't throw e.g the travelling salesman problem into prolog and get an answer, unless you're prepared to wait longer than the universe will exist. Solving problems in a short which by definition will take supercomputers trillions of years to solve naively is exactly what AI is about.

To design just one vessel, some fifty factors must be taken into account: how thick to make the armor, how much fuel to carry, what type of weapons, engines, and computer guidance system to use. Each decision is a tradeoff: a powerful engine will make a ship faster, but it might require carrying more fuel; increased armor provides protection but adds weight and reduces maneuverability.

Since a fleet may have as many as 100 ships -exactly how many is one more question to decide -the number of ways that variables can be juxtaposed is overwhelming, even for a digital computer. Mechanically generating and testing every possible fleet configuration might, of course, eventually produce a winner, but most of the computer's time would be spent blindly considering designs that are nonsense. Exploring Traveller's vast "search space," as mathematicians call it, require the ability to learn from experience, developing heuristics -rules of thumb -about which paths are most likely to yield reasonable solutions.

That's not just throwing some predicates into a language and getting a canned output. We've already established in this example that throwing the rules into prolog is not viable. Prolog brute forces the solution and it's not very fast. Solving problems of this level is actual AI.

If you think you can just throw the rules of a game into prolog and get a perfect answer, try doing so with chess. That should be simple enough right, and involve no AI? After all the rules of chess are much simpler than the wargame purchase phase and battle rules in the described game.

In the article, the predicates themselves were generated and evolved using genetic algorithms. None of that was known before. The fact that it beat all the human players by a wide margin is testament that there were no known predicates to build into the system, the computer innovated new strategies.

[Jelle]

Not for N complete problems. Read the article again. The sheer number of variables make brute force search impossible. You can't throw e.g the travelling salesman problem into prolog and get an answer, unless you're prepared to wait longer than the universe will exist. Solving problems in a short which by definition will take supercomputers trillions of years to solve naively is exactly what AI is about.

Indeed you're right, shouldn't have just skimmed only the first part of the article. Seems like it used some kind of genetic algorithms? Very interesting, should give it a thorough read later.

[zchris13]

Like I said, useful but not messianic.
AI is cool.

[Cthulhu]

Uh, no? I don't think it'll happen that easily or that fast, and I'm not entirely sure it'll work out well for us, but I don't think such predictions are that far out there.

I'm referring to it as a rabbit we pull out of the hat to solve all our problems so we don't have to worry about mortality and the state of our species and environment.  That will never happen.

I don't see true AI as impossible but I don't really see "true" AI as a thing, but that's cause I don't really believe in the things we assume to distinguish "true" AI from a regular computer program.  It doesn't have free will, consciousness, or subjective experience, but I'm not sure we do either.