su3su2u1 is a pseudonymous Internet author who posts to many places, most notably Tumblr. He has argued, at great length, that MIRI is not a real research organization and that Eliezer Yudkowsky is a crackpot. Many have written responses, including me and Scott. Instead of writing yet more replies to su3su2u1’s claims about MIRI, I’d like to explain why everyone arguing with him should stop wasting their time.

EDIT: I should emphasize that the point of this post isn’t to criticize su3su2u1 just for the sake of it, but to save people from wasting their time arguing with him. Since this is my own advice, I will follow it, and not comment further on su3su2u1’s writings after today.

Although Eliezer is not a crackpot, I think everyone must admit that he (though not MIRI’s other researchers) has several apparent signs of crackpottery. These include not having a formal education; writing mostly for his own websites, instead of peer-reviewed journals; not having had an established reputation within AI, when he first started writing about AI safety; and not co-authoring his papers with well-known AI researchers. su3su2u1 frequently criticizes MIRI and Eliezer on these grounds. su3su2u1’s stated theory is that he criticizes MIRI and Eliezer because of these signs of crackpottery. Others have argued that these are merely excuses, and that su3su2u1 just criticizes because he dislikes Eliezer personally. If the “crackpot signs” all went away, under this alternative theory, su3su2u1 wouldn’t change his mind; he’d just make up new reasons for claiming Eliezer/MIRI are crackpots.

To test these theories, we could, if you will, imagine an alternate-universe Eliezer – an Eliezer-Prime – who has unconventional ideas, but none of the “crackpot signs”. For example, we could say that:

– While Eliezer himself was mostly self-educated, Eliezer-Prime got his degrees from MIT, one of the top technical universities in the world.

– When he first thought of “Friendly AI”, Eliezer wrote about it on his own website. But when Eliezer-Prime got his big idea, he instead first published it in the Proceedings of the National Academy of Sciences, one of the top few most respected scientific journals.

– su3su2u1 has said that, if he were running MIRI, his first priority would be to make everyone get a PhD. We can say that Eliezer-Prime got a PhD. For extra impressiveness, let’s say that he also got his PhD from MIT.

– Of course, a PhD might not mean that much, if it’s in an unrelated subject. If you’re an expert on X, you might still be a crackpot on Y. So we’ll specify that Eliezer-Prime was actually awarded his MIT PhD for his “crazy ideas”. His PhD thesis, of course, would have been approved by his doctoral committee, all MIT professors. We’ll pick the most renowned professors we can find for his committee, like Marvin Minsky and Gerald Sussman.

– Eliezer’s publications on Friendliness mostly aren’t peer-reviewed. But Eliezer-Prime’s are, of course. Eliezer-Prime has a track record of many relevant, technical, highly-cited and peer-reviewed publications, in respectable scientific journals.

– MIRI, which employs Eliezer, is relatively new and not that prestigious. Instead, we’ll give Eliezer-Prime a job at, say, Oxford, the oldest and arguably most prestigious university in the Anglosphere, with previous positions at other top universities like Stanford.

su3su2u1 might not agree with Eliezer-Prime, but hopefully, he wouldn’t just dismiss him as a crackpot. If he disagreed, he’d treat it like a serious discussion with a well-respected researcher in the field, and back up his points with technical, peer-reviewed sources.

Alas, it is not to be. Eliezer-Prime is real – his name is Dr. Eric Drexler, the founder of the field of nanotechnology – and su3su2u1 commented thus:

“The whole thing [molecular manufacturing] is pseudoscience. The founder, Drexler, is a crackpot himself.” [EDIT: su3su2u1 says that this quote is from a different person with the same username. However, he’s also said that he does in fact endorse this quote, so I am not misrepresenting him. See discussion below – my apologies for any confusion.]

In his dismissal of Drexler, su3su2u1 included no math, no equations, and no technical work. He based his arguments on loose verbal analogies, and (unsourced) claims that Drexler was unaware of even the basics of the field he invented, like scaling laws (which his MIT PhD thesis spends an entire chapter on). He cited only one source, an article by chemist Richard Smalley, which wasn’t technical and wasn’t even peer-reviewed. Rather, it was a two-page pop sci magazine piece, centered around a silly analogy comparing molecular chemistry to romance. Via Tumblr, I politely asked su3su2u1 for links to technical, peer-reviewed sources that rebut Drexler’s ideas; he has so far declined to reply. If he ever does, I’d be happy to post them below. But until then, it seems safe to say that the “making up excuses” theory is vindicated, and that trying to change su3su2u1’s mind just isn’t going to happen, no matter how many of his arguments are proven false.

EDIT: su3su2u1 has written a response, with two major points.

The first is that he continues to dismiss Drexler, and continues to not provide any technical arguments or peer-reviewed sources as a basis for his dismissal. He says that, “I believe you can find similar physicists making the same argument by walking into a material science department and asking any physicist about it.” But if that’s true, where are the peer-reviewed sources? Per Google Scholar, Drexler’s book Nanosystems (an edited version of his PhD thesis) has been cited over 1,700 times. His nontechnical book, Engines of Creation, has been cited over 2,200 times. His original 1981 paper in PNAS has been cited over 500 times. If su3su2u1’s opinions are common among physicists, surely there’s a peer-reviewed source which discusses them, somewhere in all those thousands of cites.

The second is that he says su3su2u1 is a common username, and one of the sources I cite is not actually him, but a different person using the same handle. [EDIT: This discussion has been moved to the comments, per Douglas Knight’s recommendation.]