Learning Machines - Question for the Experts

[Metryq]

We've had many threads on AI (artificial intelligence). I suppose one might define AI as the ability to produce a new, and perhaps unexpected, third from any two givens. And by "new" I mean something completely novel, not merely a randomly output datum already in the system, sequitur to the givens or not.

From my own reading and various opinions expressed in this forum the consensus is that we are a long way from achieving AI, and some believe it may never be possible.

With proper programming, a machine need not be independently "intelligent" to be useful. Collation and analysis of data, often drudge work, can accelerate scientific and technical developments by putting the right info in front of a mind that does have the "spark" for those "this bath water is too hot*" moments.

AI aside, I'm curious about the current state of "learning machines." As I understand the concept, such a machine is like a self-adjusting clutch, refining itself for optimum efficiency at a defined task. Nothing truly new ever comes out of the system, but the machine's actions "evolve" with the environment.

Now the curve ball. I've read many documented accounts of "micro-evolutionary" changes. The effect is quite real. However, we seem to be missing a few pages on "macro-evolutionary" changes. Carefully controlled and documented scientific attempts (with fruit flies, for example) to produce a totally new species have failed. For centuries many breeders have found that given organisms will "stretch" to a certain point before "snapping back" to the basic pattern, like a rubber band. Micro-evolutionary, but not macro-evolutionary.

This is not an argument for "intelligent design"; "god" is an assumption that explains nothing and opens up more questions. Occam's razor cuts it off. However, I do believe Darwinian evolution is flawed. We're still missing some major pieces of the model.

The analogy here is that observed micro-evolutionary changes are like learning machines. We understand the mechanisms of both well enough, but we're still unable to explain ourselves, or "replicate" our intelligence with AI.

Thoughts?


* "'Eureka' is Greek for 'this bath water is too hot.'"
—Doctor Who

.

 

[USS Triumphant]

I think they've been on the right path with exploring and modeling all of the ways that we know our brains work into the machines. Up until recently. I think now, if we really want to get to machines that can think like us, we need to start modeling the ways our brains DON'T work into them. The way a synapse will fire but end up at the wrong destination because the neuron it was bound for has died and not been replaced by one with the correct information, or by one that had a copying error and now contains slightly *wrong* information, or because somehow it just flat out *missed*.

We won't have real, sentient AI until we build a computer capable of not remembering why in the hell it just came in the kitchen - metaphorically speaking. I think that is where true self-awareness and creativity comes from.

 

[PurpleBuddha]

Now the curve ball. I've read many documented accounts of "micro-evolutionary" changes. The effect is quite real. However, we seem to be missing a few pages on "macro-evolutionary" changes. Carefully controlled and documented scientific attempts (with fruit flies, for example) to produce a totally new species have failed. For centuries many breeders have found that given organisms will "stretch" to a certain point before "snapping back" to the basic pattern, like a rubber band. Micro-evolutionary, but not macro-evolutionary.

You sure?

http://www.pbs.org/wgbh/evolution/library/05/2/l_052_05.html

And while I don't want to nitpick too much, biologists say that the process for micro and macro evolution is identical. They are exactly the same thing with time being the only difference.

If we really could make a system that was able to learn on its own, and then also simulate a selection mechanism, perhaps we could evolve an intelligence some day. Who knows. I am certainly not an expert.

 

[Metryq]

And while I don't want to nitpick too much, biologists say that the process for micro and macro evolution is identical. They are exactly the same thing with time being the only difference.

So we are repeatedly told. And some futurists have imagined that computers are just like the human brain and will simply "wake up" one day when enough connections are formed. But the Internet hasn't spontaneously started writing poetry or wondering about its place in the universe.

Salamanders unable to breed with one another are still salamanders. The fact that human biologists have called them different species strikes me as mere semantics. Fruit fly genealogies "longer" than the age of the Earth have been exposed to every imaginable environmental stressor and mutagen, yet have failed to produce, say, a mosquito.

DNA is a complex informational code, not mere puzzle pieces tossed in the air and coming down in a different order. Fruit flies didn't mutate into mosquitoes because they contained only the information to make fruit flies. Where does the new information come from?

If we really could make a system that was able to learn on its own, and then also simulate a selection mechanism, perhaps we could evolve an intelligence some day.

Hand-wavium. "Selection" is merely verbal shorthand for an organism better adapted to its environment, but it does not tell us where information comes from. Existing learning machines already perform a kind of selection, for example favoring chess moves more likely survive the environment (a game). If anything, this reduces the machine's adaptability in the same way sex tends to suppress mutations. (Counter-evolutionary.)

I'm no expert, either, but repetition of these "truths" does not make them any truer. Dogmatism will not help our theories evolve towards greater understanding, nor will learning machines "somehow" become AI if AIs are actually mosquitoes.

If a creature, such as a human, is well adapted to its environment, then that environment changes, the human can adapt more rapidly through intelligence than physical mutation. What does that tell you about evolution?

 

[Robert Maxwell]

Now the curve ball. I've read many documented accounts of "micro-evolutionary" changes. The effect is quite real. However, we seem to be missing a few pages on "macro-evolutionary" changes. Carefully controlled and documented scientific attempts (with fruit flies, for example) to produce a totally new species have failed. For centuries many breeders have found that given organisms will "stretch" to a certain point before "snapping back" to the basic pattern, like a rubber band. Micro-evolutionary, but not macro-evolutionary.

This is not an argument for "intelligent design"; "god" is an assumption that explains nothing and opens up more questions. Occam's razor cuts it off. However, I do believe Darwinian evolution is flawed. We're still missing some major pieces of the model.

Macroevolution has been observed. Speciation has been observed. Historical evidence for both abounds.

So, what in the world are you talking about?

 

[Awesome Possum]

And while I don't want to nitpick too much, biologists say that the process for micro and macro evolution is identical. They are exactly the same thing with time being the only difference.

So we are repeatedly told. And some futurists have imagined that computers are just like the human brain and will simply "wake up" one day when enough connections are formed. But the Internet hasn't spontaneously started writing poetry or wondering about its place in the universe.

Salamanders unable to breed with one another are still salamanders. The fact that human biologists have called them different species strikes me as mere semantics. Fruit fly genealogies "longer" than the age of the Earth have been exposed to every imaginable environmental stressor and mutagen, yet have failed to produce, say, a mosquito.

DNA is a complex informational code, not mere puzzle pieces tossed in the air and coming down in a different order. Fruit flies didn't mutate into mosquitoes because they contained only the information to make fruit flies. Where does the new information come from?

If we really could make a system that was able to learn on its own, and then also simulate a selection mechanism, perhaps we could evolve an intelligence some day.

Hand-wavium. "Selection" is merely verbal shorthand for an organism better adapted to its environment, but it does not tell us where information comes from. Existing learning machines already perform a kind of selection, for example favoring chess moves more likely survive the environment (a game). If anything, this reduces the machine's adaptability in the same way sex tends to suppress mutations. (Counter-evolutionary.)

I'm no expert, either, but repetition of these "truths" does not make them any truer. Dogmatism will not help our theories evolve towards greater understanding, nor will learning machines "somehow" become AI if AIs are actually mosquitoes.

If a creature, such as a human, is well adapted to its environment, then that environment changes, the human can adapt more rapidly through intelligence than physical mutation. What does that tell you about evolution?

Our intelligence is an adaptation, one that has made us extremely successful. So it shows that evolution works, it's not a gift from an imaginary magic man in the sky. But given how our intelligence has also lead us to polluting our environment and creating weapons that could wipe us out, it could cause our extinction. Intelligence could ultimately be an evolutionary dead-end.

Also putting scares quotes around truth and comparing science to dogma doesn't make your point valid. You might as well say you hate the facts, so you're going to ignore them.

 

[CorporalCaptain]

Our intelligence is an adaptation, one that has made us extremely successful. So it shows that evolution works, it's not a gift from an imaginary magic man in the sky. But given how our intelligence has also lead us to polluting our environment and creating weapons that could wipe us out, it could cause our extinction. Intelligence could ultimately be an evolutionary dead-end.

Like all adaptations, our intelligence is, so to speak, a work in progress.

That said, I don't believe that our intelligence is solely responsible for the mess we're making of the planet. Greed, short-sightedness, and a general failure to apply the full potential of human intelligence are major factors that indicate that humans are otherwise maladapted to cope with the technology that our intelligence brings us. That's not necessarily the fault of our intelligence, but rather possibly of other human weaknesses that are able to dominate our behavior so that we do stupid things.

 

[Awesome Possum]

Our intelligence is an adaptation, one that has made us extremely successful. So it shows that evolution works, it's not a gift from an imaginary magic man in the sky. But given how our intelligence has also lead us to polluting our environment and creating weapons that could wipe us out, it could cause our extinction. Intelligence could ultimately be an evolutionary dead-end.

Like all adaptations, our intelligence is, so to speak, a work in progress.

That said, I don't believe that our intelligence is solely responsible for the mess we're making of the planet. Greed, short-sightedness, and a general failure to apply the full potential of human intelligence are major factors that indicate that humans are otherwise maladapted to cope with the technology that our intelligence brings us. That's not necessarily the fault of our intelligence, but rather possibly of other human weaknesses that are able to dominate our behavior so that we do stupid things.

I don't think it's the reason we did it, it just gave us the tools to do it.

But yeah, we're pretty happy to be as greedy and careless as we want and try to ignore the consequences or try to kick it down the road so we can deal with it later. We should try to be a little more concerned about future generations and not go out of our way to create problems for them.

 

[CorporalCaptain]

^ An important issue is that lots of people think in terms of social order. They see things solely through the lens of how people are affected, and they are blind to natural phenomena.

I'm fully willing to postulate that anthropocentric thinking is instinctive for people, and that overcoming the limitation of viewing the world exclusively that way is an important step in making human intelligence more rational. My present judgment is that anthropocentric thinking is largely and specifically responsible for a slew of current issues threatening human survival, from climate change denial to religious fundamentalism and the politicization of science education.

In dealing with anthropocentric thinking, I think it's important to acknowledge that it does represent the application of intelligence. From a logical standpoint, the fundamental problem with it is that the worldview is truncated to eliminate everything that contradicts anthropocentrism. Regarded that way, it's a problem of perception at least as much as one of cognition.

 

[Awesome Possum]

So you're saying we should dose the entire planet with mushrooms? Because that would change perception pretty quickly. It's going to be pretty hard to keep being concerned with making money at the expense of the rest of humanity when the corners of your house fly apart revealing the wireframe of the universe. The barriers around the mind collapse, time and individuality become meaningless. Where the idea of me being me and you being you are as absurd as a right and left hand being different beings and not parts of a greater whole.

Maybe then, we get our shit together and actually do something.

 

[PurpleBuddha]

And while I don't want to nitpick too much, biologists say that the process for micro and macro evolution is identical. They are exactly the same thing with time being the only difference.

So we are repeatedly told. And some futurists have imagined that computers are just like the human brain and will simply "wake up" one day when enough connections are formed. But the Internet hasn't spontaneously started writing poetry or wondering about its place in the universe.

The two things you are comparing are so different that I cannot see a point worth considering.

Salamanders unable to breed with one another are still salamanders. The fact that human biologists have called them different species strikes me as mere semantics. Fruit fly genealogies "longer" than the age of the Earth have been exposed to every imaginable environmental stressor and mutagen, yet have failed to produce, say, a mosquito.

DNA is a complex informational code, not mere puzzle pieces tossed in the air and coming down in a different order. Fruit flies didn't mutate into mosquitoes because they contained only the information to make fruit flies. Where does the new information come from?

Isn't it amazing how much change you and I can see even in our blink of an eye existence? You talk as if this process hasn't been ongoing for billions of years and this fraction of a moment is the only time worth considering when trying to understand the full scope of evolution. Do you believe in a young earth?

Hand-wavium. "Selection" is merely verbal shorthand for an organism better adapted to its environment, but it does not tell us where information comes from.

Yeah that is exactly what I was talking about. Some form of selection would be needed to have a system of evolution. What is "hand-wavium" about the concept of selection? Anyway as I said, I was just speculating as to how your idea may be true. I personally do not believe that an ai will be "evolved" in this manner.

I'm no expert, either, but repetition of these "truths" does not make them any truer. Dogmatism will not help our theories evolve towards greater understanding, nor will learning machines "somehow" become AI if AIs are actually mosquitoes.

You lost me. I thought my original post wasn't in conflict with yours for the most part, other than pointing out how you were wrong about marco and micro evolution. What dogma are you talking about? Natural selection? Fine, but good luck supporting your point.

If a creature, such as a human, is well adapted to its environment, then that environment changes, the human can adapt more rapidly through intelligence than physical mutation. What does that tell you about evolution?

Sure they could, by definition. What's your point?

 

[CorporalCaptain]

So you're saying we should dose the entire planet with mushrooms?

Oh, of course not. The obviousness aside, that unwilling participants shouldn't be dosed, it would involve all sorts of risks. For one thing, conditions for many, if not most, wouldn't be ideal. On that scale, even a small fraction of negative reactions could be disastrous, especially in the wrong places. For another, people would still, at least nominally, be in the same positions the day after that they were the day before, generally with prescribed pressures to maintain the status quo. Thirdly, I have no doubt that some would be shouting that it was the work of the Devil. Many would be falling over themselves to get in line with that, as that would provide the easy way out to get back to normal. A large-scale retrograde reaction wouldn't be out of the question.