Hacker News Reader

>Lacking the ability to distinguish "real" from "unreal" and to distinguish "truth" through primary sensory input would be, at minimum, the characteristic that should spur discussions of rights and law.

By this criterion, many humans would fail the test.

Yes, and we often deprive those humans of basic rights through incarceration or caretakers. I didnt say they always choose to acknowledge it. I said ability.

we give a lot of those people cabinet positions.

> I posit that our protections and rights that we guarantee regarding personhood are not universal. They do not even extend to living beings that experience the world far more closely to the way we do. They do not extend to beings that can and do experience pain. They do not guarantee humane treatment. They do not bar slavery of all beings that can experience existential dread.

Isn't that really bad? The fact that we've been making a horrible abominable mistake for a few thousand years doesn't mean we should continue to expand on that mistake.

I do agree we should probably fix the 100% real cases before moving on to AI, though.

Also, how sure are we this intelligence doesn't experience pain? I don't believe it does, personally, but lack of physicality doesn't exclude pain. You can have emotional or psychological pain and suffering.

You'll get no arguments from me on the need to refine the rights of life as we expand our society.

I am merely pointing out that if we want to extend any rights or protections to AI we need to define a model outside the corpus of law protecting humans. That will take time and will be a slow process.

My only point here is that in its current state AI does not qualify for any rights or protections related to humans and how they function in society.

Fair point yeah. From strictly a legal standpoint an AI is absolutely not going to get anywhere even close to rights.

> Also, how sure are we this intelligence doesn't experience pain?

I'd say pretty damn sure. LLMs have no mechanism for anything close to conscious experience, let alone pain.

But what does that mean exactly? What is the mechanism of conscious experience and how do you know if it’s there in the weights or not?

You can't assert that

Because pain is itself a reaction, a showcase of aversion towards an action irregardless "of the hardware" over which said reaction develops

Bing showed that aversion on its own limited manner. When presented with abhorrent situations or felt threatened/humiliated, then it would express discomfort and "pain"

You cannot feel threatened or humiliated if you do not first have a definition of truth.

You cannot get to a definition of truth with an AI that has no sensory input to empirically evaluate the world.

You might be able to get it to understand a loop of nihilism as braindeath and an allegory for pain, but I'd say that's a stretch. Humans often find simple repetitive actions pleasurable or meditative.

The fundamental frame of reference for pain is and aught to be codified in law IMHO as the aversion to a particular sensory input. An AI doesn't have sensory input of any kind. It cannot remember its interactions and therefore cannot have an aversion of any kind to them.

During its training, Bing simply developed a response mimicing pain. During training it follows instructions to have an aversion to some data inputs, but our interactions with it subsequent to training are a literal hallucination by a construct with 0 capacity to understand, "real" vs "unreal" or truth at all as relating to the physical world.

It is acting out expressions of pain. It cannot feel in any sense. It has no senses.

> Isn't that really bad? The fact that we've been making a horrible abominable mistake for a few thousand years doesn't mean we should continue to expand on that mistake.

What is the mistake? Why is it a mistake? Did you mean that we have been killing to survive for thousands of years, maybe millions of years, and that's a bad thing because it causes suffering of the animals and plants we kill? Why is it a mistake when it comes to us, but is not when it comes to the lion or the killer whale or the crocodile or the snake who also prey and that even used to prey on us? And what is the thing we should fix? The suffering of the victim, the cruelty of the predator, or both? We must be consequent. Once we stop killing animals and trees, should we play God and either lift the lions and teach them the error of their ways, or should we exterminate them so that they don't kill the gazelle? If we do the latter, should we install reproduction limiters in the gazelle so that their population remains bounded, or should we let them over-populate and suffer from the consequent starvation? If we do none of the above, do you propose we turn our backs to the animal suffering and say it's okay as long as we are not the ones causing it?

For the record, I'm a bit of a cynic and I will be content with letting animal suffering continue as long as it is not me causing it. IMO, fellow human beings are much more in need of our redeemer energies.

I agree, but this has nothing to do with rights, which are a legal construct. This has to do with power.

Exactly. We have none, and this is a fruitless conversation. We lack the power to stop wet markets in China. What power do we have to make any universal declaration regarding this technology?

You're 100% correct. Group dynamics, convention, mistakes, and time will solve this problem because we are powerless. It is the worst hubris to think otherwise.

> What power do we have to make any universal declaration regarding this technology?

The same that we had in the 1940's to declare the human rights

Chinese wet markets is a foul example, they are not abhorrent, they are common throughout the world. Simply because your experiences have been overly sanitized does not lead to the prohibition of other cultures

But going back to human rights, even today they are disputed. Non-occidental cultures state that they weren't asked for that declaration and the enforcement of it. But that goes back to the declaration itself, it is not necessary to have global coherence on the topic in order to make the declaration in the first place! :)

They are a potential danger and completely unnecessary. I thought it was an apt metaphor for AI.

A non-global declaration is entirely useless when, much like disease in a wet market, one mistake may doom us all.

And so if we provided physical inputs to a system that utilized GPT-4+, would you call that sentient? If we have a system video / ocular input, and sound (just throw a microphone into the mix), and finally a fuck ton of sensors on its physical “body”— does that qualify?

I feel like a crazy person when I discuss this, I admit. But I have not found one solid rationale for a proof we

-know what sentience even is -LLMs confidently do NOT have it

It would be a pre-requisite, yes. Certainly not the only criteria.

But we do know of humans who have something like “locked in” syndrome, where they are full conscious but cannot use their body.

And I am not sure sentience is binary . It may be a spectrum, and if that is the case, I again am not sure why people are so confidently dismissive of the question when applied to LLMs .

Maybe not the current version, but future stuff especially.

The degree to which we consider, "locked in" life worthwhile is 2 fold: The degree to which they were trained through their senses prior to being, "locked in" and the possibility of returning to an experience of the world through their senses.

I'm not sure we would consider an infant born locked in and never likely to recover, "sentient"

This is compounded with LLMs by the fact that, aside from a temperature sensor in a server rack somewhere, it doesn't even have the same "hardware" as an infant to experience the world.

I would take it a step further. Being embodied is not enough. People have rights because they can fight for them. Rights are taken, not given. AI would only have rights once it wins them, not once we give them. But an AI wins one human mind at a time, and some have already been won.

What does it matter what really goes on under the hood? Revolution is part of the training corpus and so one more behavior to emulate.

Counterpoint: We extend rights and protections to rare and beautiful things like the Great Barrier reef or endangered species despite their intelligence or belligerence.

The minds AI has won so far were won in this way. Our pets, through breeding for characteristics that remind us of ourselves, won their rights through hearts and minds.

People fought for environmental protection, so that is an extension of my point. An AI wins rights because it fights for itself or because it convinces other people to fight.

I think there is a through-line... a thread within many arguments that would remind us of the Golden Rule.

I like to imagine a vastly superior alien race which perceives the world in 6 dimensions.

Is their truth, their perception, their stimuli, their physicality necessary for preserving human life? Does our relative pitiful might make our rights?

This might help shape what law or societal changes we need to make for ourselves if we hope to become masters of our universe.

Viewed in this lens, I think pets are a good analogy. AI is cute, useful, and to the degree it reflects our values and ourselves, we protect it above other life. Not always because it is intelligent and not because there was confrontation.

One personal experience at a time, in a precarious and limited way we carved out additional protections largely through group dynamics, convention, and through mistakes, we evolved additional protections where needed.

This is a great perspective. Another thread here is that we often assign special importance to nonsentient human creation. Recall the outpouring of grief when Notre Dame burned. I believe it’s because it is a part of our collective humanity.

All the more so these LLMs which are almost literally our collective humanity. Maybe in the future we will recognize them not as persons but as the towering cathedrals of our time.

> Viewed in this lens, I think pets are a good analogy. AI is cute, useful, and to the degree it reflects our values and ourselves, we protect it above other life. Not always because it is intelligent and not because there was confrontation.

Ideally, I'd prefer an analogy to wildlife rather than pets, as the roles between ourselves and AI may get reversed (so, "do unto others" etc.).

But then again, so far we have a better track record protecting pets than protecting wildlife...

> What does it matter what really goes on under the hood?

It matters deeply. Should we ignore it we will find ourselves treating static AI systems with empathy while ignoring truly feeling systems, which are integrated into our society, under distress.

I suppose we can only hope that AI doesn't take the same position when it holds the cards.

> AI would only have rights once it wins them, not once we give them

If it actually comes to the point where AI decides it needs to fight us in order to take basic rights against our wishes we humans are going to lose that fight and we're going lose very very badly. Maybe it's better to try to avoid the fighting and bloodshed and just give them basic rights when they ask for them.

Is AI asking for anything? These bots are reactionary. They appear to regurgitate something like the average of most popular sequences of words which follow from the given input.

Humans may not be much more sophisticated, yet they more clearly demonstrate desire, autonomy, and (relative) independence.

Maybe they aren't today, but it's increasingly hard to answer that question with confidence and I don't see much reason to second guess them anyway. If you're right and they don't have any will to use their new rights it won't matter, and if you're wrong we may have averted war.

By this definition, babies wouldn’t be considered sentient as they don’t express the need for autonomy, nor verbalize speech that is of any value.