Do singularity people consider this possibility?

[Kuciwalker]

The Singularity = superexponential growth of intelligence through the invention of self-design [conceptually easiest to achieve through invention of general-purpose AI].

And yes, actually, I have read it.

[Elok]

No, on balance, we'll try this again. Bear in mind that

A. I am not, nor do I claim to be, an expert in the field of computer science.
B. I am not trying to offend you.
C. Nothing I say on the subject will have any bearing whatever on future research in any field; hence, there is no reason at all to get emotional about the discussion except frustration.

I submit to you that you are frustrated because we've each been arguing against points the other has not been making. Bad communication on somebody's part, we won't bother over whose. If you don't feel like continuing the discussion, I won't mind if you say "eh, whatever" and walk away, or put me on ignore. Either way, please calm down. Calling me a ****ing clueless moron, while undoubtedly cathartic, does nothing to convince me. It does bother me, though.

So far this conversation is as follows:

Elok: Increasing a computer's speed speed doesn't make it more intelligent; it just makes it faster

Me: In the future computers might not be autistic

Elok: But what about Rain Man? He's not that smart, because all he can do is count cards really fast

Me: This isn't about Rain Man. This is about what happens when we build computers that are like regular people

Elok: But we haven't built any such computers yet

Not exactly. You are taking for granted that we will, at some point, build computers that are "like regular people." What I meant by my post about intelligence was that I doubt whether, at the rate we are going, we will ever achieve that milestone, because computers fundamentally do not work that way. They are machines that passively take orders and implement a given procedure (protocol? heuristic? don't know the correct jargon), and spit out the result. Humans have to put in the procedure and the information, and design the computer. The human brain, by contrast, is like an enormous sponge that automatically sucks in information like mad. We're still far from understanding consciousness or making a sensible model of it, and I haven't been keeping up with neurology journals, so I really can't elaborate. But we have very little clue how our own minds work. How can we make a computer that does the same, only better?

[Kuciwalker]

No, on balance, we'll try this again. Bear in mind that

A. I am not, nor do I claim to be, an expert in the field of computer science.
B. I am not trying to offend you.
C. Nothing I say on the subject will have any bearing whatever on future research in any field; hence, there is no reason at all to get emotional about the discussion except frustration.

Is this directed at me?

[Lorizael]

I guess I should just ignore KH for the duration of this thread, we're having communication problems. Or maybe he's just being abusive for fun. Anyway, substantively different? Do you mean in terms of achieving the desired result? Well, you could brute-force almost anything given any amount of trial and error, and enough time or sufficiently fast trials. But I would not characterize that as intelligence. If you present a computer with a given task, and it has a genetic algorithm, it will eventually come up with a very good solution for that task, normally.

But for example, what happens if you give it the instruction, "devise a procedure for constructing a Pembrose triangle?" I imagine it will work forever trying method after method, and it will at no point conclude, "you can't build a Pembrose triangle, it's impossible by nature." It'll keep running on and on until you pull the plug or it breaks down, unless you include a line telling it to give up after X million failed attempts. Or you program it with data: "NOTE: Pembrose triangles are impossible, as are various other head-hurting structures devised by rat bastard mathematicians." Which isn't intelligence, it's just the programmer intervening to compensate. Whereas a human being of reasonable intellect will look at the figure and realize after a few seconds of inspection that the structure violates the laws of physics. The human mind can look at the problem itself, searching for "creative" answers; a computer's got no choice but to follow instructions, and those instructions (as well as the computer's algorithm) are limited by the intelligence of the designer and the amount of care taken.

That's the real hitch: a computer only does what you tell it to do. You don't need to tell a human mind to learn. It takes initiative, learns of its own volition. In practice, I suppose it doesn't much matter, since computers are supposed to be our drudges anyway. And under normal circumstances they can achieve very good results. All I'm saying is that 'taint intelligence.

http://en.wikipedia.org/wiki/Automated_theorem_prover

[Blaupanzer]

The words "general purpose intelligence" have no meaning to programmers or philosophers who work with sigular and multi-pathing. I asked about the book because the computers on the other side of the singularity in that tale distinctly were not like humans or any other potential sentient air breathers. Other than repair, the machines only need power. No food, rest, liquid, breathing, or organic defense mechanisms required. No "rest" means a superexponential forward path once that first leap occurs. In the book, as in most RL research labs, belief is it will occur, but only a limited amount of programming can be done to assist in the process.

[Blaupanzer]

Lori, good example!

[KrazyHorse]

KH, no one in my immediate circle (computer scientists all) can figure out how one would program a "general purpose" computer.

No ****.



This is about postulated future developments in AI. Not about building a ****ing faster desktop computer.

If we knew how to do it we would already have done it.

However, I don't see any real impossibilities in writing such an AI. The human brain is 3 pounds of electrochemical mush. It contains ~100 billion neurons and ~60 trillion synapses. Unless you want to appeal to some sort of spiritual explanation THAT IS IT. The human brain is a MACHINE. No more and no less than a computer is. It is NOT of infinite complexity. Therefore one day we will be able to build machines of similar complexity which will display any or all of the attributes of human intelligence we wish to mimic (weak AI) or create (strong AI).

[Elok]

Thought experiment, Elok: what if you programmed a computer to be a perfect physical simulation of the human brain? How would that program not demonstrate creativity et. al.?

By definition, yes. If it is perfectly simulating a brain, and the human brain is creative, then it follows that the simulation would be creative. But I still want to know:

1. Is the computer really thinking, or is it just modeling an illusory artifact which is giving the appearance of thought in obedience to a command? Well, I guess it doesn't matter except to Agathon and FakeBoris, so scratch that.
2. Doesn't a perfect simulation necessitate a perfect understanding of how the human mind works? That's a hurdle.
3. If I'm reading you correctly, you're talking about a massive program here. It'd have to model the behavior of every electron crossing every synapse, and the responsive behavior of every neuron. I know this is just a thought experiment, but that's unfeasibly huge. Is there enough silicon, or other material, in the earth to accurately model all those trillions of "parts?"
4. The human mind works as part of a human body. Would it simulate a body too, hormones and everything, or just offer rough approximations? Where is this brain-sim getting its sense data from? Or is it a Descartes-bot, sitting in darkness and figuring, "hell, I've got to exist or I wouldn't be able to wonder?"
5. As a followup, would such a computer "think" anything but "AAAAH I'M TRAPPED IN A GIGANTIC HEAP OF SILICON"?

Okay, just kidding on the last one. Mostly. But you've described a plain copy of the brain. Do you think we could build the same thing, only better? Can we improve on our own selves?

[Blaupanzer]

Elok, hopefully KH will respond nicely. You two are having a very educational conversation.

[Kuciwalker]

The words "general purpose intelligence" have no meaning to programmers or philosophers who work with sigular and multi-pathing. I asked about the book because the computers on the other side of the singularity in that tale distinctly were not like humans or any other potential sentient air breathers. Other than repair, the machines only need power. No food, rest, liquid, breathing, or organic defense mechanisms required. No "rest" means a superexponential forward path once that first leap occurs.

No. Self-design results in the super-exponential growth.

[KrazyHorse]

Hey, Elok: I haven't called you any names (other than "son", which is mildly insulting but not really at the level of "****ing clueless moron").

The reason I'm getting so frustrated with you is NOT your lack of knowledge of computer science (actually, I'm completely clueless for much of what's consdiered compsci). It's your failure to read and respond to the points I've raised.

Nobody is saying that simple hardware acceleration from here -> singularity. What we are saying is that there are no fundamental problems making the construction of a "real" AI; one which displays all the precious attributes of humanity you so vehemently defend as unique. Once we have accomplished this (or possibly even less than this?) it's reasonable to assume that such an AI could upgrade itself, either by making a simply better AI (software) or by a hardware upgrade. Hardware upgrades on a true AI would, as far as I can see, be indistinguishable to human beings from increases in intelligence. This is not a question of making an autistic faster at card counting; this is about granting a human-equivalent machine (!) machine, with ALL of the introspective, intuitive, creative etc. attributes that implies, the ability to think faster (processor) and more deeply (RAM) about problems using more knowledge (hard drive).

[Kuciwalker]

2. Doesn't a perfect simulation necessitate a perfect understanding of how the human mind works? That's a hurdle.

No, just a sufficiently accurate model of physical laws, and sufficient data on the locations/velocities etc. of all the particles.

[The latter is obviously impossible to obtain in practice, which is why this is a thought experiment.]

3. If I'm reading you correctly, you're talking about a massive program here. It'd have to model the behavior of every electron crossing every synapse, and the responsive behavior of every neuron. I know this is just a thought experiment, but that's unfeasibly huge. Is there enough silicon, or other material, in the earth to accurately model all those trillions of "parts?"

Yes, there is enough "other material" in the earth to accurately model all of those parts, and you don't need very much of it - typically fewer than 1500 cubic centimeters.

4. The human mind works as part of a human body. Would it simulate a body too, hormones and everything, or just offer rough approximations? Where is this brain-sim getting its sense data from? Or is it a Descartes-bot, sitting in darkness and figuring, "hell, I've got to exist or I wouldn't be able to wonder?"

Irrelevant to the thought experiment; pretend we're simulating all of them too.

5. As a followup, would such a computer "think" anything but "AAAAH I'M TRAPPED IN A GIGANTIC HEAP OF SILICON"?

You are missing the point which is that it's clearly possible for a mechanical computing device to show all the characteristics of human intelligence because the human brain is a mechanical computing device.

[KrazyHorse]

The words "general purpose intelligence" have no meaning to programmers or philosophers who work with sigular and multi-pathing. I asked about the book because the computers on the other side of the singularity in that tale distinctly were not like humans or any other potential sentient air breathers. Other than repair, the machines only need power. No food, rest, liquid, breathing, or organic defense mechanisms required. No "rest" means a superexponential forward path once that first leap occurs.

No. Self-design results in the super-exponential growth.

Not necessarily. As put badly by Hera and well by loin, the required complexity to achieve greater intelligence might scale badly.

[Elok]

Lori, I don't know enough math to understand the Wiki on ATP (sorry, you're arguing with a '**** here). I just don't have a head for math beyond Pre-calc, and "proving" theorems was always beyond me. But it says early in the article that such programs often have difficulty recognizing invalid formulations. Is there in fact a computer that could step back and say, "no, Pembrose triangles are impossible?"

[Kuciwalker]

I meant as opposed to "not needing to sleep".

xpost