What do economics and robotic overlord scenarios have in common? Well, positing robots that have cognitive capabilities, they would assumably know real economics. Therefore, wouldn’t the law of comparative advantage apply? Even if robots could do everything imaginable better than humans (itself a stretch), it’d be in their best interest to stick to what they’re best at, and let us work at what we’re best at. No systematic obliteration of humankind is then necessary, right?
I guess that does make science fiction a lot more boring, though. Here’s the truly horrific version, then: we end up creating intelligent robots that are all Keynesians!
Actually the fact that it’s better for robots to cooperate with humans than to enslave or destroy them is the theme of the Matrix trilogy. (Not the first film but all three, which is why most people claim they only like the first one.)
Molineux has an interesting explanation for the huge commercial success of this “robots will screw us” theme: it appeals to what everybody unconsciously knows to be true about the State. Think about it: we build something designed to help (machines), which eventually turns against us and enslaves us even worse. Thus we live in an illusion, which can be recognized as such only by a few. Others will resist attempts of conversion and even defend “the system” violently if need be. Looks familiar?
A super intelligent computer would probably consider humans as a useless nausience and just eradicate us, assuming it has the means to do so at very low cost. Since making the planet uninhabitable for organic life is probably of no concern to a computer nuking it becomes an option and it can get rid of us at very low cost.
If I had control over a nuclear arsenal, was immoral and unaffected by the environment. Could create intelligent life myself and lived on a planet with a bunch of useless animals that where just dumb, in the way of my factories and generally annoying I would nuke the place.
If those idiotic animals had the capacity to build nuclear weapons I would most definitely nuke them out of existence … monkeys with nukes is way to dangerous to have around.
There is however a solution to this domesday scenario. If we can learn to integrate our own minds with machines before (or at the same time) we learn to build self-evolving super intelligent machines that should solve the problem and remove the restrictions of biological evolution for us as well to level the playing field.
But again, this ignores the law of comparative advantage (which was the whole point): given two people/groups A and B, even if A is better at absolutely everything, it’s to A’s advantage to do what they’re best at, let B do what they’re best at, and then trade.
Except if we can do everything so much better ourselves the marginal utility of B’s production would be quite small.
The cost on the other hand of having monkeys with nukes around, not being able to build factories where we want cause the monkeys have cities in all the nice places and so forth would be much greater then the itty tiny bit of value they could add to our massive wealth.
The whole idea that robots would have “interests” seems contrived. If you program them to have simulated interests, they will, but if not they won’t. Why would a robot want to survive? That’s a uniquely gene-based facet of animal life, based on evolution. I see no reason to believe it would apply to robots or AI.
To the point, AJ! Why the hell would an AI want to do anything at all? Did we program it that way? Asimov’s laws would appear to rule out any possibility of us having trouble with AIs (I, Robot notwithstanding). So there is hardly any chance that some Matrix-like universe could come about (but if it could, would we really want to get out)?
That may be how our home computers work but is not how AI design works…
The kind of robots we are speaking of would be coded to be self-motivated evolving machines. What you are speaking of is a normal computer but the most effective way to build an AI, unlike a normal computer, is to let the AI build itself. In order for that to be possible it has to be motivated to pursue certain goals and capable of evolving.
In this link TED have had robots basically designing themselves towards set evolutionary goals. What is really interesting is towards the end of the moving they didn’t reward them for anything, they just added power and waited. Eventually the algorithms created there own goal … self-replication.
So these machines won’t only be governed by the evolutionary guidelines we gave them to build themselves with, they will most likely develop there own goals along the way.
Most likely if machines like this got out of control they would be obsessed with self-replication and destroy everything in there path to do create more of themselves. Basically functioning like the earlies stages of biological evolution and the simplest forms of biological life … a virus.
I agree that would happen if you created robot with AI by a self-evolution system (and I’ve usually thought this would be the best way), but if the AI were just a program the “evolution” could happen entirely within a virtual universe created by the programmer.
It seems that even if the AI reaches ultra-superhuman intelligence it still has no way of knowing about anything outside its virtual world. The disadvantage of that is we don’t get direct AI robots, but the advantage is we don’t create Skynet. Maybe that ultra-intelligent program could give us enough clues to build rule-based robots that are pretty bad-ass, yet not ambitious.
More than likely an AI with the same basic cognitive patterns as humans and other animals wouldn’t see us as a threat, but rather as an opportunity as there are no other cognitive species (or synthetic lifeforms) visiting this little mote in God’s eye. Plus, consider the fact that such an AI would have as much to fear from death as we do as there’s a good chance that such AI couldn’t be created or manufactured, but rather taught or growth (as part of its own life cycle). Thus, to gamble it all on being overlord of Earth would seem like the fancy of a sociopath or a drunkard to such a synthetic.
Also, such an AI could be bound by the same logic and reasoning that limit deductive conclusions. If the fact is that no one can make all of economics decidable then that cannot be circumvented (thus no Keynesian bot for you). And no amount of sheer memory or raw processing power can overcome undedicability (it’s a fact/reality of Computer Science, don’t believe me look it up in Turing Decidable Languages.).
Threat or no threat when animals get in our way we wipe them out. An official argument used for moose hunting up here is that moose are a traffic hazard that needs to be controlled. It is not just predators and pest that we wipe out when we make contact with.
It is also perfectly rational to exterminate moose so that people don’t have to crash with them. So if the computers had the same cognitive patterns as us, where superior (which was a an important condition of the question) I most definitely don’t think they would go out of there way to achieve peaceful co-existence. If we are in the way, aren’t advanced enough to produce much or to pose much of threat then the choice is simple.
However there is little reason to believe we will be in the scenario given for this question. The AIs will probably be more like animals that we should be able to control. It is not like overnight we have second sentient species on the planet.
Also pilot testing will probably be done as software only as AJ suggested so we can see what happens and how they evolve in simulation before we give them real guns…
Then why do people often avoid killing random road crossing ducks and geese in the park? It’s not out of fear of fines as I suspect such fines are minimal compared to actual hunting without a license. How about the fact that people feel revulsion when they find out a neighborhood child or their own child has taken to torturing the dogs or cats in the neighborhood, possibly even killing a few? This is because humans have empathy.
Empathy is the mechanism by which all living things of our type can learn. An AI without empathy will never learn, they will simply be automations that follow syllogism to its end without question. Empathy is the learning mechanism’s only means to progress. If you don’t believe me, go and search on the learning capacities of sociopaths and lower order autistics; they both follow the same path of limitation.
Typical bullshit of someone who’s not actually versed in psychology. The ability to put together the same reasoning a little faster does not imply superiority. In fact, as a person with an officially registered IQ of 125, I can tell you that raw analytical reasoning never trumps an apt capacity to see fallacies or errors, even if it takes a long time for the person to illustrate the error or fallacy.
Also, you’re assuming that the theory of such conflicting interests is only applicable to human/synthetic relations. What about human/human relations? What are the facts about that in terms of theory of conflicting interests? If we look at economic theories, it’s found that people tend to actually want to cooperate. This is very true in labor specialization, where one person is better at producing one good over another in contrast of another person (vice versa). Whether its a farm or a factory, a town or another country, the result is the same, coordination happens between all conflicting interests to achieve ends in an optimal fashion (rather it should be considered less than perfect from optimal, but it’s optimal in respect to all concerned).
Not likely, sorry. Because again, you’re missing an entire set of knowledge here as with psychology. To make an AI that can be utilized for human utility, it must be human-like in its capacities. To have something akin to a cat or a dog to pilot an airplane, it wouldn’t work out as to interface with it would be harder than simply leaving it on autopilot and having a human cozy up to a novel in the time between take off and landing. Plus, there’s little market for an AI that doesn’t fit with human utility, unless you’re looking to make a market in some sort of synthetic pets or whatever.
Empathy is a product of biological evolution. There is really very little reason to believe artificial evolution would value this at all since it isn’t affected of inter-dependency within the tribe in the same way humans have been. Also humans really have very little empathy, we don’t only exterminate animals because someone have the notion that they are in the way, we exterminate other humans as well. The later is not a logical to preserve most of our values, the sooner is.
Dude read the first post again. That the computers are superior is a premise of the question this tread is about. If that is possible or not is a whole different matter which I haven’t said much about yet.
However it is clear that artificial life could very easily become superior to us. Not because it thinks faster but because it evolves much much faster. Complex biological evolution is extremely slow. Artificial life, no matter how complex, can however still adapt and reconstruct itself all the time. It will evolve much faster then anything we have seen.
Again you are missing the entire premise of the tread which is that one party is superior in all regards to the other.
No, if we build a self-piloting air-plane we won’t have an air-plane with an artificial human pilot. What we will have is a self-piloting air-plane that we just need to tell what to do… exactly like riding a horse. The development AI is taking is that the machine can do more and more by itself, but is still guided by human instruction and learns according to incentives we create. That is no different then any trained animal.
Many of the projects being done now are even modelled after animals, artificial rescue dogs and whatnot. In general animals are a pretty good template cause use animals to carry out complex tasks that we can’t use humans for today. Removing the element of human supervisions altogether from machines that are supposed to do something useful is still along way away.
Then you don’t grasp what empathy does, so I’ll explain. Empathy for living creatures like us allows us to learn from each other and from the environment. This is due to the nature of the different kinds of neurons that make up a brain. In this case, empathy is driven by the mirror neurons. As their name suggests, they allow a brain to mimic or mirror the behaviors, thoughts, and emotions of others. In a brain that has malfunction or maladaptive issues with mirror neurons, learning is restricted. This is true of low order autistics and sociopaths, where although their analytic capacities seem to be intact their capacity to learn in general is severely retarded. This is why autistics can have powerful traits of memory and raw computation, but never seem to learn the simplest tactics or methods to learn new information (like where the nearest pharmacy is or some other generalized piece of knowledge). This is also why sociopaths may score high in analytical IQ tests, but fail to get into universities or on-the-job training programs (as these require active learning capacities related to mirror neurons).
Then your assumption is a false one, no computer is ever superior in any case. You lack an understanding of computer science and electrical engineering to even make the case. I suggest getting into Information Theory (and Computational Theory) to get a grasp of the gravity of your assumption, especially in terms of what it would do to every known theory (as proven) in computing.
Why does being superior imply cruelty?
Nope on everything in this paragraph (I’m a CS major going to grad school, so I’m informed on this issue, I would like to debate this in private if you have the time…).