it is more a matter of “Anyone who thinks that an AI cannot be built using utility functions is a crackpot.” It is a core belief in the sense that it is not allowed to be false. It is unthinkable, so rather than try to defend it, those who deny it have to be personally attacked.
Won’t comment about past affairs, but these days at least part of MIRI seems more open to the possibility. E.g. this thread where So8res (Nate Soares, now Executive Director of MIRI) lists some possible reasons for why it might be necessary to move beyond utility functions. (He is pretty skeptical of most, but at least he seems to be seriously considering the possibility, and gives a ~15% chance “that VNM won’t cut it”.)
The day that I get invited as a guest speaker by either MIRI or FHI will mark the point at which they start to respect and take seriously alternative viewpoints.
Won’t comment about past affairs, but these days at least part of MIRI seems more open to the possibility. E.g. this thread where So8res (Nate Soares, now Executive Director of MIRI) lists some possible reasons for why it might be necessary to move beyond utility functions. (He is pretty skeptical of most, but at least he seems to be seriously considering the possibility, and gives a ~15% chance “that VNM won’t cut it”.)
The day that I get invited as a guest speaker by either MIRI or FHI will mark the point at which they start to respect and take seriously alternative viewpoints.