Is everything a religion?

[Elok]

My intended point was that we're treating this as a performance challenge--accomplishing X tasks--when what would be necessary would not be meeting such a challenge but getting one system to mimic the behavior of a radically different, already existing system. It's reinventing the wheel, out of an unsuitable material, in a world where wheels grow on trees. You know what else has surpassed humans in a variety of ways? A cinderblock. Humans have substantially lower weight-bearing potential, require more resources to sustain themselves, and can't be left exposed outdoors for three months without harm. Yet I am not afraid of the cinderblock singularity.

[Jon Miller]

My problem is that AI has been around 0 years.

What is often called AI now can be replaced by a series of gears if desired. I wouldn't call Deep Learning (what created the Go beating 'AI') to be any step toward AI.

It is true that if we started making steps toward AI that the skills the AI had at 'memory' and 'pattern recognition' might be much better than humans. But that is the same way as saying that since dolphins have been taught to jump higher and higher through hoops that they are making amazing progress towards sentience and will quickly surpass humans in the next few years.

Basically: there is no obvious relation between what happens in the field of machine learning with intelligence as seen in humans or even animals.

JM

[Kidlicious]

Life on Earth has been evolving for billions of years. Of course it's really good at what it does. AI has been around for .00000005 billion years and has already surpassed humans in a wide variety of ways. Declaring at this stage that a lack of total success is indicative of a hard limit on AI seems a little premature.

AI is not life.

[Jon Miller]

I think that Neural Networks are very misnamed, and the recent advances that most people refer to have no analogy in biological systems.

It is pretty clear to me that that NNs only had biological systems as a muse and that the muse wasn't really the biological system but rather our 1980s/1970s understanding of the biological system.

JM

[Lorizael]

My problem is that AI has been around 0 years.

What is often called AI now can be replaced by a series of gears if desired. I wouldn't call Deep Learning (what created the Go beating 'AI') to be any step toward AI.

It is true that if we started making steps toward AI that the skills the AI had at 'memory' and 'pattern recognition' might be much better than humans. But that is the same way as saying that since dolphins have been taught to jump higher and higher through hoops that they are making amazing progress towards sentience and will quickly surpass humans in the next few years.

Basically: there is no obvious relation between what happens in the field of machine learning with intelligence as seen in humans or even animals.

JM

Try to come up with a solid, non-self-serving definition of sentience that also verifiably applies to other humans whose subjective inner life you have no access to but couldn't possibly be applied to machines.

[Jon Miller]

Why?

I am saying that our current progress in machine 'learning' are not steps toward creating even a dog level intelligence much less a human level one.

The misunderstanding is due what we happened to call the field a few decades ago.

It is like saying that engines were steps towards a super machine intelligence... yes, they are strong machine systems but that doesn't have anything to do with intelligence.

JM

[Lorizael]

Right. That means you're making the claim that what we're doing is not related to sentience/intelligence, despite AI doing things that we at least casually relate to smart people things (playing Chess, etc.). So you need a good definition that is also clearly the kind of thing that happens in other humans/animals. Coming up with such a definition is more difficult than you might first believe. That is, it's not easy to make the case that humans/animals other than yourself are sentient/intelligent but that nothing else is.

[Jon Miller]

the 'AI' is not playing chess

the human is playing chess using a prosthetic, like a hearing aid...

this (sort of thing) is the best argument that we are doing anything in real AI http://www.artificialbrains.com/openworm

JM

[Jon Miller]

Question:

If there was some trip wire and a shotgun and someone came and tripped the wire and got shot, is this an example of AI killing humans. If the human who set it up died long before, would it change the nature (AI killing humans or just a prosthetic of the human)?

I am not claiming that AI is impossible, I am claiming that what is done in the fields of Machine Learning and Artificial Intelligence (including with any type of neural network ) has nothing to do with real Artificial Intelligence. It is only doing things like creating a better speaker or engine or etc.

Basically in your belief system, we are just working on the eye... which does connect into the brain at a deep level... we aren't working with the brain.

JM

[Lorizael]

Humans don't play chess. The Earth does, by using humans.

This is clearly not a comparable situation, but unless you can give a good description of what the activity of chess-playing is and show that humans other than yourself obviously engage in it but that machines clearly don't...

[Elok]

Not exactly following you guys here, but it's not like various tools haven't been surpassing human capabilities for a very long time. That is the whole reason we have tools, after all; a fairly simple calculator leaves the mathematical abilities of most humans in the dust. But AFAIK we haven't invented anything that shows initiative, unpredictability, emotions, or learning. I don't think we know how to even begin to do such a thing. I suppose it's possible that that could change, but I don't see it happening real soon.

[Lorizael]

If problem-solving machines were predictable, we wouldn't need them.

[Jon Miller]

I start playing chess.

You could argue that my decision to start playing chess is the same sort of thing as the chess algorithm's 'decision' to move the pawn to K3, but it isn't. The Chess algorithm is evaluating a loss function when it 'decides' to move the pawn to K3, and I (am likely) not doing so when I decide to play chess. This is because (all evidence) points towards me not evaluating a loss function when I make most decisions.

This isn't an inefficiency with Nature (Nature after all is very efficient) and this is not "shorts for some, pants for others".... this is a fundamental and crucial difference.

It is because what I (and even dogs) do is what you need to be able to do in an open environment (like what we experience in nature) and evaluating a loss function is applicable only for the constrained space of models/etc (which is where the chess application or go application is).

Even a 'self-driving car' is still working in a extremely constrained space of models and not in the open space which exists in nature.

JM

[Jon Miller]

If problem-solving machines were predictable, we wouldn't need them.

They are 100% predictable and could be replaced by a system of gears, or a human with X amount of time and paper.

JM

[Lorizael]

They are 100% predictable and could be replaced by a system of gears, or a human with X amount of time and paper.

JM

For some problems that machines are able to solve, X amount of time might be so long as to be meaningless. Humans, being made of nothing more than goo, are also in principle predictable, even if we are currently extremely bad at predicting what humans will do, because 100 billion neurons is a lot and we understand very little about how human intelligence works at present.