https://en.wikipedia.org/wiki/David_Lewis_(philosopher) interesting guy. modal realist. also influential on "radical interpretation", a theory that we can interpret behavior (including speech) to infer things like the beliefs, desires, and intentions of the agent, without knowing anything about internal cognitive mechanisms (like brain states). radical interpretation (some claim) is the philosophical theory you want to use if you want the results of alignment experiments to mean something nontrivial about the beliefs/intentions/goals/desires/deceptions/etc of AIs.
when can you fairly say that you know something about an AI's beliefs or desires? People throw around implications of these proxy metrics way too casually (eg "scheming evals" presupposes that getting a certain score on a certain test means the AIs have an intent to deceive and seek power) and it's bad for the discourse.
philosophy can help a bit! there are certain desiderata we'd definitely like to see before we decide an entity has a "belief" or "desire." we can make our evals and experiments less terrible by checking for these desiderata.
beliefs should be held (pretty much) consistently. they should be (approximately) rational and coherent. an entity should not believe a thing and its opposite, for instance. and beliefs should motivate behavior.
desires should be (mostly) transitive. they should motivate behavior. they should (mostly) follow means-ends reasoning -- if you desire A, and believe B is necessary for A, then you should also desire B.
i appreciate and am troubled by the critique in the comments that this is illiberal. someone like JS Mill would say that you can only know the truth if you have exposed yourself to the best arguments from all sides; the "liberal subject" is assumed to be able to think critically. the frame in Johnson's tweet where by default we are going to be fooled, manipulated, hypnotized, triggered, etc by communication, is one where few people can participate in the liberal "marketplace of ideas" thing, and it sits uncomfortably with ideas like "more information is always good" or "we learn from disagreement and putting our ideas to the test" or "rationality requires facing facts." is it willful blindness to be concerned about "security" against manipulative or upsetting content?
on the other hand: Johnson is talking about tactics, drawn from therapy or philosophy, for how to distance yourself emotionally from speech/text. I think "liberal subjects" or "critical thinkers" are already doing this, actually! Any time you evaluate something critically, you are evaluating it at a remove. It isn't hijacking your amygdala, it isn't evoking a whole host of intense sensory/emotional associations that you can't get out of your head, or at least you are able to put such effects aside from your verbal/intellectual assessments. For people who don't do that easily, zooming in on "ok, well, how do people do that?" seems helpful.
if i imagine reversing this advice, that would be like "actually you should always be totally absorbent, hypnotized by every influence that comes your way", and that sounds obviously terrible!
Is it bad to "look away" from stuff that your nervous system can't handle?
well, the thing is, maybe it's "bad" in the sense that people ought to respect you less, or discount your views, if you're easily overwhelmed.
it's clearly impractical to say "people who are easily overwhelmed should constantly expose themselves to things that overwhelm them." that doesn't work. they will just be overwhelmed all the time. there is such a thing as learning to cope with distress/challenge, or getting desensitized to things, but i don't think it works like "if you flay your mind with the worst stuff on the internet, you will become a master of resilience".
i'm not sure what works and what doesn't when it comes to "desensitization" or "getting over" things.
"practice makes permanent" seems to be a thing. if you practice a behavior, it will come to feel more natural. "don't react in a dysfunctional way when you see X" is a habit you can build.
"self-soothing" also seems to be a thing. doing what you need to do to become comfortable, rather than forcing yourself to remain in maximum discomfort. in my experience, i find i actually have to invest in "being nice to myself", "finding my comfort zone", "protecting my peace", etc, and it makes a huge difference in my everyday effectiveness. is it right or wrong? who knows. it works. people who do it don't regret it.