World's top Go player loses second match of five against Google AI

[aleph naught]

But there are already gears.  Where would a computer get an idea of a wheel.  Maybe the idea of the wheel came from seeing a stone rolling down a hill.  Would a computer come up with ideas in such a way?

From examples, just like us. Computers learn the exact same way as we do: you show them a picture of a wheel and tell them it's a wheel, and after showing them enough different wheels they start being able to pick out wheels themselves.

In artificial neural networks this works through an algorithm called backpropegation, and if you know just a little bit of calculus you'd be able to understand it

[Nelvan]

But what about the first person who came up with the idea of the wheel?  There had to be a first person.  It is a different story to improve upon a wheel but what exactly is the first wheel improving upon?

[TheBigOhMan]

But there are already gears.  Where would a computer get an idea of a wheel.  Maybe the idea of the wheel came from seeing a stone rolling down a hill.  Would a computer come up with ideas in such a way?

From examples, just like us. Computers learn the exact same way as we do: you show them a picture of a wheel and tell them it's a wheel, and after showing them enough different wheels they start being able to pick out wheels themselves.

In artificial neural networks this works through an algorithm called backpropegation, and if you know just a little bit of calculus you'd be able to understand it

Calculus is the work of the devil.

[Lion IRC]

Meanwhile, in other news;

A Ferrari beats Usain Bolt in the 100 metre sprint.

[Trinity]

Can it make coffee?

[wonderer]

Strong AI would be able to express why move X is better than move Y.

Why think that?

Do you think that you could verbally express a description of your mother's face that would unambiguously describe your mother to anyone who read the description?  Of course not.  Yet you easily recognize your mother's face without being able to give a detailed account of how.

The knowledge in a neural network can be a lot like that intuitive recognition of your mother.  You can't really explain it.  The recognition is something that just happens subconsciously in your brain.

[Nelvan]

Wonderer,

I could say that my mother doesn't look Asian.  I could say she looks more female than male.  I could say she doesn't look like a toddler.  I can find many creative ways to kep eliminating.

I could also say my mother looks like so and so actress.  Or I could say she looks like an owl because she has big round eyes, a round face and small mouth and tiny nose.  I can add that she looks like a worried owl because her forehead is wrinkled and her eyes slightly droopy.  I can say that she looks like a cross between an owl and a gold fish and so on.

[wonderer]

Wonderer,

I could say that my mother doesn't look Asian.  I could say she looks more female than male.  I could say she doesn't look like a toddler.  I can find many creative ways to kep eliminating.

I could also say my mother looks like so and so actress.  Or I could say she looks like an owl because she has big round eyes, a round face and small mouth and tiny nose.  I can add that she looks like a worried owl because her forehead is wrinkled and her eyes slightly droopy.  I can say that she looks like a cross between an owl and a gold fish and so on.

Yes, you can say all those things, but they have virtually nothing to do with the nearly instant process of recognizing your mother which occurs.subconsciously in your brain.

That stuff above might be like a story you make up about recognizing your mother, but you don't actually use an item by item checklist that you can explain when actually recognizing your mother.

Similarly there is no good reason to expect a strong AI based on neural networks to be able explain all the processes leading to the AI's own choice of actions. 

[Nelvan]

Wonderer,

I don't think that is the point.  I mean I know the color red or green when I see it.  I don't know how I know it.  I don't think about it.  It seems automatic.
I'm sure an AI can also tell the difference between green and red but does a computer know what color is?  Can it know any abstract concept like that?   

[kurros]

Wonderer,

I don't think that is the point.  I mean I know the color red or green when I see it.  I don't know how I know it.  I don't think about it.  It seems automatic.
I'm sure an AI can also tell the difference between green and red but does a computer know what color is?  Can it know any abstract concept like that?

If you asked that Jeopardy! AI, Watson, what color something was, it would probably tell you the correct answer, despite not having any visual sensors . So they can certainly categorise information into abstract classes like that, yes.

And as for being able to explain their choices, that is a crazy requirement. Why must an AI be smarter than a human and fluent in English to be a strong AI? If we made an AI that was as smart as a bird at general problem solving then I would say we had done it, but like birds it could not tell us why it was doing what it was doing.

[kurros]

Meanwhile, in other news;

A Ferrari beats Usain Bolt in the 100 metre sprint.

If you don't think this is a significant milestone in AI then you should go read about it more, because you are wrong.

[Nelvan]

Watson AI simply has the definition of color memorized.  It still has no idea what color is.

As far as birds, actually they do communicate in their own way.  As far as communicating how they build their nest, for example, that is not what they do.  Maybe AI is not meant to explain anything but in that case, it remains soft AI.

[kurros]

Watson AI simply has the definition of color memorized.  It still has no idea what color is.

As far as birds, actually they do communicate in their own way.  As far as communicating how they build their nest, for example, that is not what they do.  Maybe AI is not meant to explain anything but in that case, it remains soft AI.

AI can communicate too, not sure what your point is. And are you really sure that *you* don't "simply" have the definition of colours memorised?

[kurros]

Just a little more on the scale of the task:

Here's how much computing power Google DeepMind needed to beat Lee Sedol at Go

In short:  1,202 CPUs and 176 GPUs

[Lion IRC]

Meanwhile, in other news;

A Ferrari beats Usain Bolt in the 100 metre sprint.

If you don't think this is a significant milestone in AI then you should go read about it more, because you are wrong.

I don't like being told I'm wrong. *sniff*


In my opinion a Ferrari actually CAN go faster than Usain Bolt. But you seem to disagree.

I'd be impressed if the Go software was written by a non-human.
Self-sufficient, autonomous AI writing its own AI software script. THAT would send shivers up my spine.

But for the time being, all we have is the same old boring "if/then" method merely sped up faster and faster so it does more 'stuff' in the same time - thereby achieving exactly what we would expect.

Fast doesn't equal 'intelligent'.