I’m trying to help a dear friend who would like to work on FAI research, to overcome a strong fear that arises when thinking about unfavorable outcomes involving AI. Thinking about either the possibility that he’ll die, or the possibility that an x-risk like UFAI will wipe us out, tends to strongly trigger him, leaving him depressed, scared, and sad. Just reading the recent LW article about how a computer beat a professional Go player triggered him quite strongly.
I’ve suggested trying to desensitize him via gradual exposure; the approach would be similar to the way in which people who are afraid of snakes can lose their fear of snakes by handling rope (which looks like a snake) until handling rope is no longer scary, and then looking at pictures of snakes until such pictures are no longer scary, and then finally handling a snake when they are ready. However, we’ve been struggling to think of what a sufficiently easy and non-scary first step might be for my friend; everything I’ve come up with as a first step akin to handling rope has been too scary for him to want to attempt so far.
I don’t think that I’ll even be able to convince my friend that desensitization training will be worth it at all—he’s afraid that the training might trigger him, and leave him in a depression too deep for him to climb out of. At the same time, he’s so incredibly nice, and he really wants to help with FAI research, and maybe even work for MIRI in the “unlikely” (according to him) event that he is able to overcome his fears. Are there reasonable alternatives to, say, desensitization therapy? Are there any really easy and non-scary first steps he might be okay with trying if he can be convinced to try desensitization therapy? Is there any other advice that might be helpful to him?
This sounds like someone who’s salient feature is math anxiety from high school asking how to be a research director at CERN. It’s not just that the salient feature seems at odds with the task, it’s that the task isn’t exactly something you just walk into, while you sound like you’re talking about helping someone overcome a social phobia by taking a part-time job at supermarket checkout. Is your friend someone who wins International Math Olympiads?
He sounds like someone with a phobia of fire wanting to be a fireman. Why does he want to work on FAI? Would not going anywhere near the subject work for him instead?
He wants to work on FAI for EA/utilitarian reasons—and also because he already has many of the relevant skills. He’s also of the opinion that working on FAI is of much higher value than, say, working on other x-risks or other EA causes.
If someone has anxiety about a topic, I suggest they go after all the normal anxiety treating methods. SSC has a post about Things that Sometimes Work If You Have Anxeity, though actually going to see a therapist and getting professional help would likely help more.
If he wants to try exposure therapy, good results have apparently recently occurred from doing that while on propranalol.
Working on AI research and working on FAI research aren’t the same thing. I think it’s likely a bad idea to not distinguish between the two when talking with a friend who wants to go into research and fears that UFAI will wipe us out.
More to the core of the issue desensitization training is a slow way to deal with fears and I’m not sure that it even works in this context.
A good therapist or coach has tools to help people deal with fears.
FAI is only a problem because of AI. The imminence of the problem depends on where AI is now and how rapidly it is progressing. To know these things, one must know how AI (real, current and past AI, not future, hypothetical AI, still less speculative, magical AI) is done, and to know this in technical terms, not fluff.
I don’t know how much your friend knows already, but perhaps a crash course in Russell and Norvig, plus technical papers on developments since then (i.e. Deep Learning) would be appropriate.
There’s no such thing, any more than there is research into alien flying saucers with nanolasers of doom. There’s a lot of fiction and armchair speculation, but that’s not research.
Any reason he’s not trying to fix his phobia by conventional means?
What I mean is that he’d be interested in working for MIRI, but not, say, OpenAI, or even a startup where he’d be doing lots of deep learning, if he overcomes his phobia.
I’m trying to help a dear friend who would like to work on FAI research, to overcome a strong fear that arises when thinking about unfavorable outcomes involving AI. Thinking about either the possibility that he’ll die, or the possibility that an x-risk like UFAI will wipe us out, tends to strongly trigger him, leaving him depressed, scared, and sad. Just reading the recent LW article about how a computer beat a professional Go player triggered him quite strongly.
I’ve suggested trying to desensitize him via gradual exposure; the approach would be similar to the way in which people who are afraid of snakes can lose their fear of snakes by handling rope (which looks like a snake) until handling rope is no longer scary, and then looking at pictures of snakes until such pictures are no longer scary, and then finally handling a snake when they are ready. However, we’ve been struggling to think of what a sufficiently easy and non-scary first step might be for my friend; everything I’ve come up with as a first step akin to handling rope has been too scary for him to want to attempt so far.
I don’t think that I’ll even be able to convince my friend that desensitization training will be worth it at all—he’s afraid that the training might trigger him, and leave him in a depression too deep for him to climb out of. At the same time, he’s so incredibly nice, and he really wants to help with FAI research, and maybe even work for MIRI in the “unlikely” (according to him) event that he is able to overcome his fears. Are there reasonable alternatives to, say, desensitization therapy? Are there any really easy and non-scary first steps he might be okay with trying if he can be convinced to try desensitization therapy? Is there any other advice that might be helpful to him?
This sounds like someone who’s salient feature is math anxiety from high school asking how to be a research director at CERN. It’s not just that the salient feature seems at odds with the task, it’s that the task isn’t exactly something you just walk into, while you sound like you’re talking about helping someone overcome a social phobia by taking a part-time job at supermarket checkout. Is your friend someone who wins International Math Olympiads?
He sounds like someone with a phobia of fire wanting to be a fireman. Why does he want to work on FAI? Would not going anywhere near the subject work for him instead?
He wants to work on FAI for EA/utilitarian reasons—and also because he already has many of the relevant skills. He’s also of the opinion that working on FAI is of much higher value than, say, working on other x-risks or other EA causes.
If someone has anxiety about a topic, I suggest they go after all the normal anxiety treating methods. SSC has a post about Things that Sometimes Work If You Have Anxeity, though actually going to see a therapist and getting professional help would likely help more.
If he wants to try exposure therapy, good results have apparently recently occurred from doing that while on propranalol.
Working on AI research and working on FAI research aren’t the same thing. I think it’s likely a bad idea to not distinguish between the two when talking with a friend who wants to go into research and fears that UFAI will wipe us out.
More to the core of the issue desensitization training is a slow way to deal with fears and I’m not sure that it even works in this context. A good therapist or coach has tools to help people deal with fears.
Oops. I’ve tried to clarify that he’s only interested in FAI research, not AI research on the whole.
I think that interest in AI research in general would help to demystify the whole topic a bit, it would make it look a little bit less like magic.
FAI is only a problem because of AI. The imminence of the problem depends on where AI is now and how rapidly it is progressing. To know these things, one must know how AI (real, current and past AI, not future, hypothetical AI, still less speculative, magical AI) is done, and to know this in technical terms, not fluff.
I don’t know how much your friend knows already, but perhaps a crash course in Russell and Norvig, plus technical papers on developments since then (i.e. Deep Learning) would be appropriate.
There’s no such thing, any more than there is research into alien flying saucers with nanolasers of doom. There’s a lot of fiction and armchair speculation, but that’s not research.
Any reason he’s not trying to fix his phobia by conventional means?
What I mean is that he’d be interested in working for MIRI, but not, say, OpenAI, or even a startup where he’d be doing lots of deep learning, if he overcomes his phobia.
It might be that his interest in FAI is tied to his phobia so if the phobia goes away, so may the interest...