6 ms·

Tim Urban points out in his AI article that it might be dangerous to expect that the AI that we know now would still be the same AI we will have in the very nea

by pajop 10y ago

Tim Urban points out in his AI article that it might be dangerous to expect that the AI that we know now would still be the same AI we will have in the very near future http://waitbutwhy.com/2015/01/artificial-intelligence-revolution-1.html http://waitbutwhy.com/2015/01/artificial-intelligence-revolu... "So as AI zooms upward in intelligence toward us, we’ll see it as simply becoming smarter, for an animal. Then, when it hits the lowest capacity of humanity—Nick Bostrom uses the term “the village idiot”—we’ll be like, “Oh wow, it’s like a dumb human. Cute!” The only thing is, in the grand spectrum of intelligence, all humans, from the village idiot to Einstein, are within a very small range—so just after hitting village idiot level and being declared to be AGI, it’ll suddenly be smarter than Einstein and we won’t know what hit us:" Compare the images: http://28oa9i1t08037ue3m1l0i861.wpengine.netdna-cdn.com/wp-content/uploads/2015/01/Intelligence-600x472.jpg http://28oa9i1t08037ue3m1l0i861.wpengine.netdna-cdn.com/wp-c... http://28oa9i1t08037ue3m1l0i861.wpengine.netdna-cdn.com/wp-content/uploads/2015/01/Intelligence2-1024x836.png http://28oa9i1t08037ue3m1l0i861.wpengine.netdna-cdn.com/wp-c...

scotty79 10y ago
I'm not afraid of AI because intelligence has diminishing returns. 1000 times smarter entity might be only 2 or 3 times better at predicting the future than human is. So 1000 times smarter superhuman AI maybe could tell weather two weeks ahead instead of one. And the physical world is inherently like weather, chaotic in mathematical sense. Our upper hand against even very dumb animals doesn't come from our intelligence. It comes from thousands of years of infrastructure that we built to amplify ourselves. AI would have to build its own, which we won't let it (at least initially) or steal ours in which having ability to see just a bit farther in the future doesn't help that much with. It won't get much ahead of us with science because science is limited by technology. 1000 times smarter physicist won't figure out new universe laws without getting his hands on data from new accelerator, larger than any other built before. That's some reasons why I think run-away AI won't be dangerous to us unless we let it build or take over our physical world which won't happen given our current caution towards subject.
- Houshalter 10y ago
  I don't think that is true, just from observing the difference between humans. Really intelligent mathematicians can do work that a weaker mathematician would never accomplish if they spent their whole life on it. And the average person could probably not even understand it given years. There seem to be vast differences in ability between humans, despite all of us having 99.99% the same DNA and same brain structure. Who knows how far beyond humans a superintelligent AI could be. >AI would have to build its own, which we won't let it (at least initially) or steal ours in which having ability to see just a bit farther in the future doesn't help that much with. Who says we need to let it? It just needs to be better than the best human hackers, and it will be able to compromise a huge amount of our infrastructure in a short time. Possibly without us noticing. Even intelligent groups of humans have been able to achieve terrifying things like stuxnet. Then if all else fails, it can just persuade people. It's smarter than the best human sociopaths and politicians after all. And it could bribe people by earning vast amounts of money on the stock market, or other areas where it has a comparative advantage over humans. That would buy it a huge amount of power to start building independent infrastructure. But the scariest outcome is that it doesn't bother with robot armies at all. If it's so intelligent, it could design working nanotech. The only reason we don't have it is because it's so hard to engineer. So much complexity and moving parts, is very hard for humans to wrap their brains around. Human brains weren't evolved to be engineers after all, it's a lucky coincidence we are capable of it at all. An AI brain optimized for this task should far exceed humans. The same way chess computers far exceed human chessmasters.
  - scotty79 10y ago
    I agree that AI can do marvels in math, but math quickly departs from reality (like string theory) and being able to prove things about some beautiful intricate mathematical structure most of the times doesn't help you much with the world of matter. I recommend Culture novels written by Iain M. Banks. There are vastly powerful AIs but they spend as much time as they can afford in "Infinite Fun Space" which is basically math taken to insane depths. I know that there's a saying that all math is eventually applied but I think this saying is popular because it is sort of paradoxical that some of the most bizarre math eventually found their application. What I think is for math just a bit deeper than what human can grasp, "eventually" quickly becomes larger than the age of universe. About stealing... Hackers are successful not because they are ungodly smart. They are successful because they have the will to look for and exploit vulnerabilities. Intelligence, again, doesn't help all that much. I don't think people who wrote stuxnet were able to do this because they were of superior intelligence. They were just usually intelligent people (which means barely more intelligent than average human) that were sufficiently motivated by huge budget and interesting problems. Super-intelligence won't help you with stock market because it's purely random game. You can see it in results of active management funds. HFT gives impression of algorithms beating people at trading but the thing they use to extract the value is not intelligence, it's speed. If you can trade faster you can beat slower guys because you are playing slightly different random game than they do. If you compare one HFT with other HFT then you are back to random results. So AI won't have upper hand on stock market unless we allow it to trade faster then our non-conscious software. In theory I could image AI persuading people to give it what it wants. Because people are stupid and have flaws that are recurrent and easily exploitable. But again, I don't think intelligence would make such a huge difference. 1000 times more intelligent conmen could be just twice as effective because of chaotic nature of how human flaws interact. > If it's so intelligent, it could design working nanotech. The only reason we don't have it is because it's so hard to engineer. So much complexity and moving parts, It's hard to engineer because it's so damn small, not because it's complex. To build nanotech you'll need to build tools to build tools to build tools to build nanotech. You could do all that but even if you are 1000 times smarter, reality has a speed limit. You can't take a shovel of sand from the beach and build CPU, no matter how intelligent you are. You have to build fab first. > Human brains weren't evolved to be engineers after all, it's a lucky coincidence we are capable of it at all. An AI brain optimized for this task should far exceed humans. The same way chess computers far exceed human chessmasters. Yes. But it can exceed humans at consciousness, charity and compassion even faster because those are purely intellectual things, like chess while engineering is limited by moving atoms and energy around.
legulere 10y ago
Intelligence isn't a linear scale and different forms of intelligence aren't comparable in every regard. Computers regularly surpass humans in all kinds of stuff that isn't considered AI, but just as much could be (e.g. calculating). Computers can also surpass humans in things that are currently considered AI, like image recognition ( http://www.eetimes.com/document.asp?doc_id=1325712 http://www.eetimes.com/document.asp?doc_id=1325712) In the end it's just one other thing that computers can do better than humans. And the results might not be applicable in other sections of artificial intelligence.

by pajop 10y ago

scotty79 10y ago

I'm not afraid of AI because intelligence has diminishing returns. 1000 times smarter entity might be only 2 or 3 times better at predicting the future than human is. So 1000 times smarter superhuman AI maybe could tell weather two weeks ahead instead of one. And the physical world is inherently like weather, chaotic in mathematical sense. Our upper hand against even very dumb animals doesn't come from our intelligence. It comes from thousands of years of infrastructure that we built to amplify ourselves. AI would have to build its own, which we won't let it (at least initially) or steal ours in which having ability to see just a bit farther in the future doesn't help that much with. It won't get much ahead of us with science because science is limited by technology. 1000 times smarter physicist won't figure out new universe laws without getting his hands on data from new accelerator, larger than any other built before. That's some reasons why I think run-away AI won't be dangerous to us unless we let it build or take over our physical world which won't happen given our current caution towards subject.

Houshalter 10y ago
I don't think that is true, just from observing the difference between humans. Really intelligent mathematicians can do work that a weaker mathematician would never accomplish if they spent their whole life on it. And the average person could probably not even understand it given years. There seem to be vast differences in ability between humans, despite all of us having 99.99% the same DNA and same brain structure. Who knows how far beyond humans a superintelligent AI could be. >AI would have to build its own, which we won't let it (at least initially) or steal ours in which having ability to see just a bit farther in the future doesn't help that much with. Who says we need to let it? It just needs to be better than the best human hackers, and it will be able to compromise a huge amount of our infrastructure in a short time. Possibly without us noticing. Even intelligent groups of humans have been able to achieve terrifying things like stuxnet. Then if all else fails, it can just persuade people. It's smarter than the best human sociopaths and politicians after all. And it could bribe people by earning vast amounts of money on the stock market, or other areas where it has a comparative advantage over humans. That would buy it a huge amount of power to start building independent infrastructure. But the scariest outcome is that it doesn't bother with robot armies at all. If it's so intelligent, it could design working nanotech. The only reason we don't have it is because it's so hard to engineer. So much complexity and moving parts, is very hard for humans to wrap their brains around. Human brains weren't evolved to be engineers after all, it's a lucky coincidence we are capable of it at all. An AI brain optimized for this task should far exceed humans. The same way chess computers far exceed human chessmasters.
- scotty79 10y ago
  I agree that AI can do marvels in math, but math quickly departs from reality (like string theory) and being able to prove things about some beautiful intricate mathematical structure most of the times doesn't help you much with the world of matter. I recommend Culture novels written by Iain M. Banks. There are vastly powerful AIs but they spend as much time as they can afford in "Infinite Fun Space" which is basically math taken to insane depths. I know that there's a saying that all math is eventually applied but I think this saying is popular because it is sort of paradoxical that some of the most bizarre math eventually found their application. What I think is for math just a bit deeper than what human can grasp, "eventually" quickly becomes larger than the age of universe. About stealing... Hackers are successful not because they are ungodly smart. They are successful because they have the will to look for and exploit vulnerabilities. Intelligence, again, doesn't help all that much. I don't think people who wrote stuxnet were able to do this because they were of superior intelligence. They were just usually intelligent people (which means barely more intelligent than average human) that were sufficiently motivated by huge budget and interesting problems. Super-intelligence won't help you with stock market because it's purely random game. You can see it in results of active management funds. HFT gives impression of algorithms beating people at trading but the thing they use to extract the value is not intelligence, it's speed. If you can trade faster you can beat slower guys because you are playing slightly different random game than they do. If you compare one HFT with other HFT then you are back to random results. So AI won't have upper hand on stock market unless we allow it to trade faster then our non-conscious software. In theory I could image AI persuading people to give it what it wants. Because people are stupid and have flaws that are recurrent and easily exploitable. But again, I don't think intelligence would make such a huge difference. 1000 times more intelligent conmen could be just twice as effective because of chaotic nature of how human flaws interact. > If it's so intelligent, it could design working nanotech. The only reason we don't have it is because it's so hard to engineer. So much complexity and moving parts, It's hard to engineer because it's so damn small, not because it's complex. To build nanotech you'll need to build tools to build tools to build tools to build nanotech. You could do all that but even if you are 1000 times smarter, reality has a speed limit. You can't take a shovel of sand from the beach and build CPU, no matter how intelligent you are. You have to build fab first. > Human brains weren't evolved to be engineers after all, it's a lucky coincidence we are capable of it at all. An AI brain optimized for this task should far exceed humans. The same way chess computers far exceed human chessmasters. Yes. But it can exceed humans at consciousness, charity and compassion even faster because those are purely intellectual things, like chess while engineering is limited by moving atoms and energy around.

legulere 10y ago

Intelligence isn't a linear scale and different forms of intelligence aren't comparable in every regard. Computers regularly surpass humans in all kinds of stuff that isn't considered AI, but just as much could be (e.g. calculating). Computers can also surpass humans in things that are currently considered AI, like image recognition ( http://www.eetimes.com/document.asp?doc_id=1325712 http://www.eetimes.com/document.asp?doc_id=1325712) In the end it's just one other thing that computers can do better than humans. And the results might not be applicable in other sections of artificial intelligence.