For people not directly involved with SIAI, is there specific evidence that it isn’t run by a group of genuine sociopaths with the goal of taking over the universe to fulfill (a compromise of) their own personal goals, which are fundamentally at odds with those of humanity at large?
Humans have built-in adaptions for lie detection, but betting a decision like this on the chance of my sense motive roll beating the bluff roll of a person with both higher INT and CHA than myself seems quite risky.
Published writings about moral integrity and ethical injunctions count for little in this regard, because they may have been written with the specific intent to deceive people into supporting SIAI financially. The fundamental issue seems rather similar to the AIBox problem: You’re dealing with a potential deceiver more intelligent than yourself, so you can’t really trust anything they say.
I wouldn’t be asking this for positions that call for merely human responsibility, like being elected to the highest political office in a country, having direct control over a bunch of nuclear weapons, or anything along those lines; but FAI implementation calls for much more responsibility than that.
If the answer is “No. You’ll have to do with the base probability of any random human being a sociopath.”, that might be good enough. Still, I’d like to know if I’m missing specific evidence that would push the probability for “SIAI is capital-E Evil” lower than that.
1) You’ll have to do with the base probability of a highly intelligent human being a sociopath.
2) Elaborately deceptive sociopaths would probably fake something other than our own nerdery...? Even taking into account the whole “But that’s what we want you to think” thing.
3) All sorts of nasty things we could be doing and could probably get away with doing if we had exclusively sociopath core personnel, at least some of which would leave visible outside traces while still being the sort of thing we could manage to excuse away by talking fast enough.
4) Why are you asking me that? Shouldn’t you be asking, like, anyone else?
Re. 4, not for the way I asked the question. Obviously asking for a probability, or any empirical evidence I would have to take your word on, would have been silly.
But there might have been excellent public evidence against the Evil hypothesis I just wasn’t aware of (I couldn’t think of any likely candidates, but that might have been a failure of my imagination); in that case, you would likely be aware of such evidence, and would have a significant icentive to present it. It was a long shot.
I looked into the issue from statistical point of view. I would have to go with much higher than baseline probability of them being sociopaths on the basis of Bayesian reasoning starting with baseline probability (about 1%) as a prior and then updating on the criteria of things that sociopaths can not easily fake (such as e.g. previously inventing something that works).
Ultimately, the easy way to spot a sociopath is to look for the massive dis-balance of the observable signals towards those that sociopaths can easily fake. You don’t need to be smarter than sociopath to identify the sociopath. The spam filter is pretty good at filtering out the advance fee fraud and letting business correspondence through.
You just need to act like statistical prediction rule on a set of criteria, without allowing for verbal excuses of any kind, no matter how logical they sound. For instance the leaders of genuine research institutions are not HS dropouts; the leaders of cults are; you can find the ratio and build evidential Bayesian rule, with which you can use ‘is HS dropout’ evidence to adjust your probabilities.
The beauty of this method is that it is too expensive for sociopaths to fake honest signals—such as for example having spent years to make and perfect some invention that has improved lives of people, you can’t send this signal without doing immense lot of work—and so even as they are aware of this method there is literally nothing they can do about it, nor do they want to do anything about it as there are enough people who do not pay attention to certainly honest signals to fakeable signals ratio (gullible people), whom sociopaths can target instead, for a better reward to work ratio.
Ultimately, it boils down to the fact that genuine world saving leader is rather unlikely to have never before invented anything that did demonstrably benefit the mankind, while a sociopath is pretty likely (close to 1) to have never before invented anything that did demonstrably benefit the mankind. You update on this, and ignore verbal excuses, and you have yourself a (nearly)non-exploitable decision mechanism.
For people not directly involved with SIAI, is there specific evidence that it isn’t run by a group of genuine sociopaths with the goal of taking over the universe to fulfill (a compromise of) their own personal goals, which are fundamentally at odds with those of humanity at large?
Humans have built-in adaptions for lie detection, but betting a decision like this on the chance of my sense motive roll beating the bluff roll of a person with both higher INT and CHA than myself seems quite risky.
Published writings about moral integrity and ethical injunctions count for little in this regard, because they may have been written with the specific intent to deceive people into supporting SIAI financially. The fundamental issue seems rather similar to the AIBox problem: You’re dealing with a potential deceiver more intelligent than yourself, so you can’t really trust anything they say.
I wouldn’t be asking this for positions that call for merely human responsibility, like being elected to the highest political office in a country, having direct control over a bunch of nuclear weapons, or anything along those lines; but FAI implementation calls for much more responsibility than that.
If the answer is “No. You’ll have to do with the base probability of any random human being a sociopath.”, that might be good enough. Still, I’d like to know if I’m missing specific evidence that would push the probability for “SIAI is capital-E Evil” lower than that.
Posted pseudo-anonymously because I’m a coward.
I guess my main answers would be, in order:
1) You’ll have to do with the base probability of a highly intelligent human being a sociopath.
2) Elaborately deceptive sociopaths would probably fake something other than our own nerdery...? Even taking into account the whole “But that’s what we want you to think” thing.
3) All sorts of nasty things we could be doing and could probably get away with doing if we had exclusively sociopath core personnel, at least some of which would leave visible outside traces while still being the sort of thing we could manage to excuse away by talking fast enough.
4) Why are you asking me that? Shouldn’t you be asking, like, anyone else?
Re. 4, not for the way I asked the question. Obviously asking for a probability, or any empirical evidence I would have to take your word on, would have been silly. But there might have been excellent public evidence against the Evil hypothesis I just wasn’t aware of (I couldn’t think of any likely candidates, but that might have been a failure of my imagination); in that case, you would likely be aware of such evidence, and would have a significant icentive to present it. It was a long shot.
I looked into the issue from statistical point of view. I would have to go with much higher than baseline probability of them being sociopaths on the basis of Bayesian reasoning starting with baseline probability (about 1%) as a prior and then updating on the criteria of things that sociopaths can not easily fake (such as e.g. previously inventing something that works).
Ultimately, the easy way to spot a sociopath is to look for the massive dis-balance of the observable signals towards those that sociopaths can easily fake. You don’t need to be smarter than sociopath to identify the sociopath. The spam filter is pretty good at filtering out the advance fee fraud and letting business correspondence through.
You just need to act like statistical prediction rule on a set of criteria, without allowing for verbal excuses of any kind, no matter how logical they sound. For instance the leaders of genuine research institutions are not HS dropouts; the leaders of cults are; you can find the ratio and build evidential Bayesian rule, with which you can use ‘is HS dropout’ evidence to adjust your probabilities.
The beauty of this method is that it is too expensive for sociopaths to fake honest signals—such as for example having spent years to make and perfect some invention that has improved lives of people, you can’t send this signal without doing immense lot of work—and so even as they are aware of this method there is literally nothing they can do about it, nor do they want to do anything about it as there are enough people who do not pay attention to certainly honest signals to fakeable signals ratio (gullible people), whom sociopaths can target instead, for a better reward to work ratio.
Ultimately, it boils down to the fact that genuine world saving leader is rather unlikely to have never before invented anything that did demonstrably benefit the mankind, while a sociopath is pretty likely (close to 1) to have never before invented anything that did demonstrably benefit the mankind. You update on this, and ignore verbal excuses, and you have yourself a (nearly)non-exploitable decision mechanism.
What would be the best way of producing such evidence? Presumably, organisational transparency—though that could - in principle—be faked.
I’m not sure they will go for that—citing the same reasons previously given for not planning to open-source everything.