Muggle Powers

It could function in what we would call an 'un-computer-like fashion' even if it weren't based on a different hardware architecture. That's the whole point of abstraction.

You're still projecting a variety of tenuously related parts of the system onto levels completely different from where they actually reside.

You don't make a creative AI by using a 'creativity opcode', any more than, say, a human botanist understands the growth of cherries on a tree using a 'cherry-understanding neuron'. You can take apart a botanist's brain into individual neurons, and look at every single one, and you will never find one with more than absolutely zero understanding of cherries. Projecting the limitations of neurons onto humans makes about the same amount of sense as projecting the limitations of machine code instructions onto AIs, for essentially the same reasons.

Join my forum game!

[quoteblock]It could function in what we would call an 'un-computer-like fashion' even if it weren't based on a different hardware architecture. That's the whole point of abstraction.[/quoteblock] If the AI isn't computer like you can't take advantage of computer like attributes like fast and efficient processing sp, or perfect memory. In this case you might as well replace it with a human.

[quoteblock]You don't make a creative AI by using a 'creativity opcode', any more than, say, a human botanist understands the growth of cherries on a tree using a 'cherry-understanding neuron'.[/quoteblock]

Actually, I was implying the opposite. An opcode is inherently controlable. You can turn on the function or turn it off. My supposition is that creativity comes from everywhere and nowhere. You can encourage it or discourage it but you certaintly can't be creative on command or stop having new ideas.

[quoteblock]Projecting the limitations of neurons onto humans makes about the same amount of sense as projecting the limitations of machine code instructions onto A.I.s, for essentially the same reasons.[/quoteblock]

But this doesn't help you're argument at all. First off, it's wrong because we can see the beginnings of creativity and problem solving in even small batches of nerves and second it's basically saying "We don't know how an AI works and theirfore it can do everything". Finally, you're equating a complicated idea to a fundamental attribute.

Hardly. You can still make it better than a human, at particular tasks it's optimized for, or even in general. If you have enough hardware power one year to make a human-equivalent AI, a few years down the road when hardware power has increased you can have a superhuman AI.

That makes more sense than assuming a bunch of arbitrary limitations that have more to do with our intuition than anything else, seeing as we already know how flawed that intuition is.

I don't think so. Creativity is more general than cherry-understanding, but calling it 'fundamental' is terribly misleading. Neither one has any more to do with machine code than it does with neurons.

Join my forum game!

Just one word. Chess playing computers.

We used to think that it was a skill that was very creative, unpredictible and chaotic for computers to beat humans. 20 years ago a computer beat the best of us (not easily) and Kasparov said that he thought he saw a human hand guiding the computer because he saw imagination.

Now, the top chess playing programs, running on a consumer level computers, don't even compare with humans. When the subject comes to if computers could beat the best human players, it's not a matter of if they will win, but how much of a handicap should the computer have, so humans could have a chance to win. It's crazy and it is not yet an AI that plays against us.

Hell, battlefield AI is not even that hard to program. We can do that with even todays computers and no AI.

If an AI civilisation has come to rule its own homeplanet (and have the means for faster than light travel, because if not, how do we even have this war? (Oooooh! This could be a nice handicap for the AI. Make it so FTL doesn't exist(because if it does exist the AI could invent it) and the robots have only a limited army to conquer earth!)) it means that it is creative enough to win a war against its creators.

edited 11th Jun '15 12:27:20 PM by DDentonas

We've become quite good at designing algorithms that solve very specific, narrow tasks, such as playing Chess, without actually being intelligent. To me, the test of a genuinely intelligent machine is versatility- the ability to not only solve problems, but to solve many different kinds of problems, to solve new kinds of problems it hasn't been specifically prepared for, and to spot non-obvious connections between different areas of thought, jumping between them and combining them in appropriately efficient yet imaginative ways. There seems to be no good reason why computers can't do this, given that humans (and even some animals) can, but at the same time, we still don't seem to be very close to figuring out how to write an algorithm with this sort of ability.

Join my forum game!

That's called "General Artificial Intelligence" or AGI. One of the problems we have is that we dont really know how to define AGI or measure it. There are many obstacles to advances in AGI, but a reliance on formal logic isnt one of them. Human intelligence does rely on heuristics, but there's nothing magical about it- we simply tag certain features of everyday experience with emotional intensity, and then default to a general script or template for dealing with that type of experience. When this happens in a problem solving or decision making context, we call it "bias" or "logical fallacy"; but that doesnt mean that the underlying cognitive process is somehow fundementally "alogical" whatever that would mean. Somewhere, down at the level of neuro-chemical connections, there is a sequence of "on-off" patterns that could certainly be replicated with a code of some kind. It's all information.

The problem is that the human brain contains 100 billion neurons, with an average of 10 thousand axons connecting them. That's a lot of complexity, and I suspect we just dont have the computational capacity to replicate that yet.

There have been some pretty impressive advances though. Like an AI that learned to play multiple different videogames with no-preprogramming given only a video input and their current score.

Blind Final Fantasy 6 Let's Play