Bloviating

vivianimbriotis | Sept. 16, 2022, 2 p.m.

“One of the things I strongly dislike about myself is my need to say banal and uninteresting things in an obtuse and bloviating way.” That’s Socrates. He is sick, remember? He’s sitting in a blocky foam chair, his hospital bed behind him, the view of Athens revealed through the floor-to-ceiling window to his left. It is drizzling comfortingly outside, droplets lazily tickling the glass.

 

“I’ll stop you if what you say is too obvious,” says Diogenes. He’s bedraggled, for a doctor, and wears a name-tag that just says ‘dog’. “That is what I did when your student brought out the chicken, after all."

"Okay. Well I can talk, but only so long as what I'm saying is interesting."

"I wouldn't worry, Mr Socrates," says the doctor, smiling. "After all, it only needs to be interesting to us."

Part one: Second-order preferences

The chair creaks as Socrates shifts in it. "So I want various things. And I cannot have those things."

"Are you certain that you cannot have them?" Diogenes is sitting crouched, and now you can see why his name-tag says "dog" because he crouches like one.

"I could have them, possibly, but I would need to do things I am not willing to do, "

"May I have an example?"

"Mm. I would like to look like an Olympian - the sportspeople, not the gods," Socrates says, then adds parenthetically, "Although perhaps looking like a god would not be so bad."

Socrates, of course, believes in a very attractive god indeed. He continues, "But I have only finite time, and am not willing to make the temporal sacrifices I would have to make in order for that to happen." 

Socrates is not conventionally attractive, neither conventionally unattractive - his body is simply not the notable part about him, not the thing that leaves an impression.

"You could always try anabolic steroids," stage-whispers Diogenes.

"Those haven't been invented yet."

"Oh yeah, good point. Bad side effects, too. Attached to your liver?"

"Not quite sure what it does yet, but Pliny tells me I can perk it back up with some ground-up wolf's liver if mine stops ticking along."

"I think we're awaiting a proper RCT on that one."

"Anyway, if I could look like sculpture, that would be wonderful, but I cannot."

"Sure."

"So my new wish is this: if I cannot have that, I want to simply not want it."

"You wish to change your preferences? You have preferences about what your preferences ought to be?"

"I have preferences that I wish to quash, yes."

Part two: Unstable utility functions; Inner and outer alignment

"So I've been doing some reading about those thinking-machines that this Alan Turing guy is going to make," says Plato. His visitor badge is huge, bigger than his head. He sits awkwardly on the edge of the bed, and in stark contrast to Diogenes his toga is clean and pressed, his hands and face soft and clear. 

"Is this going to be an obtuse way to say something obvious?" says Socrates.

"There's this problem in artificial intelligence called 'inner alignment'," continues Plato, ignoring him. "It goes like this:

At first you want X, you want it exclusively

You perform actions to take it conclusively

 If an apple you want then an apple you eat

If a hot bod you need then a steroid tastes sweet

But one day you think, it would be efficient

To make something else that is apple-deficient

It eats apples for you, at first without flaw

But soon runs right into Goodhart's cursed law. 

When you create a mesa-optimizer

You had better hope it does not devise a

Reward structure different than exactly what it heard

Else distributional shift will have occurred."

"Gross," says Socrates. "Terrible rhythmic meter."

Part three: The need to be loved

Look - it's evening now, Plato has left, and Diogenes is back. Does he ever leave this place?

"I'm very lonely, you see," says Socrates. "I want few things, and one of them is a relationship. I do not have one, and it doesn't seem like that is going to change."

"It could change," and that's Diogenes, scratching his dirty hair with his stethoscope.

"It could. But it could have changed for a long time now. I have been hovering between acceptance and hope, and neither can flourish in the shadow of the other."

"Okay. So how can I help?"

"Well, I have this desire. It is a big desire, and somewhat outside my control. I wish to no longer have it."

"I'm not sure I can accomplish that."

"I mean, you must have a scalpel here somewhere - can't you cut it out, leaving the rest of me untouched?"

"No. I can't. You would be a different person then."