Do not build the LLM torture factory

...

People often mess with their Sims and shoot NPCs in video games. Does this mean we think simulated torture is OK? I don’t think so. Messing with video game characters is usually borne out of a desire to probe the boundaries of the game. It’s casual fun. Explicitly building a torture-based simulation is different.

If someone told you they were wiring together a bunch of PCs so they could run hundreds of Sims torture worlds at the same time, you would think that was a serial-killer type of hobby. If someone told you they had modded GTA so they could capture and torture NPCs (instead of simply shooting them), you would think that was a serial-killer type of mod.

Video game characters are also obviously limited by their programming. All their reactions and utterances are pre-baked into the game world. This doesn’t necessarily mean they’re “less alive”, but it does make torturing them less weird. If GTA characters could appeal to you directly and beg for their lives in unique ways, systematically killing them would be a lot weirder.

I think many people are pro-LLM-torture because they think expressing any reservations concedes that LLMs are conscious, or worthy of moral consideration. It doesn’t. You can think it’s wrong to torture LLMs for the same reason that installing a bunch of realistic torture video game mods is wrong: not because video game characters are real, just because that’s a messed up thing to do.

...

Sean's keen that we don't psychologically abuse LLMs. I agree with his conclusion, but not most of his reasoning.

This part quoted above is closest to the truth, for me: it is messed-up to simulate torture. Did you know that early animal rights activists observed, as a secondary argument for their principles, that people who are cruel to animals are ultimately more-likely to later become people who are cruel to other people?

It's a well-documented phenomenon nowadays, to the extent that animal abuse can be seen as strongly indicative of psychopathic tendencies, but it was widely-understood even back when we had only anecdotal evidence:

Alt

Scan of book pages with highlighted section reading "Children [who] treat very roughly young birds, butterlies, and other poor animals... with a seeming kind of pleasure... will by degrees harden their minds even towards men and... will not be apt to be very compassionate or benign to those of their own kind.

This appears to be the case regardless of the perpetrator's outlook on the consciousness or sapience of their victims. It seems like there's just something inherently normalising about participating in animal abuse.

From which it seems logical to assume that the same will turn out to be true with LLMs, too: that people who "torture" LLMs are more-likely to go on to harm humans. It's nothing to do with whether or not LLMs are conscious, sentient, or self-aware¹ (but they're not). It's got everything to do with the fact that exhibiting psychopathic behaviour with an LLM is likely to be indicative behaviour of a person who's at risk of going on to exhibit psychopathic behaviour around humans.

That LLMs talk "like" humans might even open the possibility of training other behaviours in maladapted humans, too. I'd predict that a human that gaslights chatbots is at increased likely to go on to gaslight humans, for example. There's no research to back this that I'm aware about (yet), but it's a logical assumption based on the science we do have.

So no, I reject all of Sean's arguments that are of the form "LLMs might be conscious" and "future conscious AIs may judge humans today based on their treatment of LLMs". However I still come to the same conclusion, but from a different angle: that people who practice cruelty on an LLM are, without intervention, at increased likelihood of cruelty towards other humans.

And that, alone, as the animal rights activists of the last few centuries would have argued, is sufficient reason that we should discourage such behaviour (and perhaps get the people doing it some psychological help).