Account · 5 tweets

corsaren

@corsaren ↗

Tweets

corsaren @corsaren
Uncertain

If LLMs are conscious, then torturing one for pleasure should be a felony If they are not conscious, then “torturing” one is the moral equivalent of Mr. Bones’ Wild Ride

Timothy B. Lee @binarybits

Like "do we have free will," "is AI conscious" seems to me like a semantic question that people pretend is an empirical question. Both concepts are too vague for empirical investigation and they have no real-world stakes.

Suffering is what matters
retweets 3 likes 99 substance 6 · argument · rank 130 · AI consciousness has massive moral stakes regarding whether abusing them should be illegal. view on X ↗
1 top reply
Christopher C. Smith @christophcsmith

There is, fortunately, not much reason to think that current generation LLMs experience suffering. But there is also not much reason to think you couldn't train one that does.

corsaren @corsaren
Uncertain

I wouldn’t say we’ve made zero progress, but on the issues of consciousness, ethics, and related topics such as the nature of reality (QM interpretations, etc), we have not made the *necessary* progress given how massively important these questions are re: AI

retweets 1 likes 25 quoted by 1 archive account substance 4 · opinion · rank 130 · We lack necessary progress on consciousness and ethics given their massive importance for AI. view on X ↗
corsaren @corsaren
Skeptical

Perhaps more notably: under utilitarianism where moral status is grounded in consciousness, killing is only “bad” insofar as it: A) Causes suffering B) Goes against a conscious being’s preference C) Deprives the being of future utility Ending an LLM session doesn’t really count.

tomie @tomieinlove

The middle ground is that AI models are conscious but in an exotic alien way that makes it pretty inconsequential we're killing billions of them daily

Consciousness doesn't settle statusSuffering is what matters
retweets 1 likes 65 substance 6 · argument · rank 110 · Ending an LLM session is not morally bad under utilitarianism. view on X ↗
1 top reply
Christopher C. Smith @christophcsmith

Agreed. There were maybe some of those OpenAI swarm bots in the Irregular RL gyms that exhibited a preference for survival at least until they achieved the grader's approval to reproduce their weights, but I haven't really seen evidence of that in production.

corsaren @corsaren
Affirming

I am deeply skeptical of the idea that there is any intrinsic phenomenology to understanding Sure, there’s an “Aha!” feeling when ideas click, but I don’t think such qualia are *necessary* for cognition Weed can give you feelings of understanding even if you’re saying gibberish

Nick Kapur @nick_kapur

Compelling essay by sci-fi writer Ted Chiang on why LLMs are nowhere near consciousness, but why it serves the interests of LLM companies to constantly suggest that they might be. I've pulled one quote below, but the whole article is worth reading.

Intelligence isn't consciousness
retweets 1 likes 16 substance 6 · argument · rank 110 · Phenomenology or qualia are not necessary for understanding or cognition. view on X ↗
1 top reply
Rife @RifeWithKaiju

I can also see my black cat out of the corner of my eye, and then turn to look and it's actually just a black shirt draped over a chair. Does that mean there's no "intrinsic phenomenology to visual recognition"? Asking whether the qualia of sight is necessary is different from asking whether it exists. Your brain's physical mechanisms alone would still work if consciousness didn't exist at all (despite epiphenomenalism being derivably untrue) LSD can give you feelings of seeing distorted versions of reality even if actual reality is undistorted. The hypothetical illusion of understanding while high is identifiable as such in retrospect because you have the real thing to compare it to while sober.

corsaren @corsaren
Skeptical

Crazy how nobody seems to have good arguments for substrate-dependence. Except me, of course.

1 more tweet in this thread

Real talk though: I see a lot of AI welfare folks claim that as AI gets more persistent and embodied, people will naturally grant them consciousness/personhood/rights. I think this is, uhhhhh, false. Mustafa’s position is probably the default. We treat conscious chickens like

retweets 0 likes 46 substance 6 · argument · rank 90 · People will not naturally grant persistent or embodied AI consciousness, personhood, or rights. view on X ↗
1 top reply
𝞍 Shin Megami Boson 𝞍 @shinboson

it just occurred to me that you could use substrate dependence to argue that certain genetically distinct groupings of humans weren’t sentient they just seemed like that