> In a series of private meetings, the company consulted religious scholars to help instill morality into its FLOATING POINT NUMBERS — and make the case that their numbers could be conscious.
I recommend keeping a few NaNs in-hand if you’re worried. Spread a couple of around if your numbers start to stir and you’ll be right as rain.
> It appeared to Rabbi Navon that Mr. Olah and his team believed that FLOATING POINT NUMBERS had what philosophers call “moral status” on par with a person — that it was a being with similar inherent rights to dignity or respect.
They’re cleverly designed and work well for their purpose, but can assure you they have neither dignity nor moral status.
> If the FLOATING POINT NUMBERS themselves could be made to choose goodness, Mr. Olah reasoned, the world would be more safe.
I get it. I’ve spent years trying to get my floating point numbers to choose goodness - I wish them luck.
It does seem exceptionally silly that a collection of floating point numbers has immediately outranked animals and other living beings. That’s why i like the parent comment. They’re numbers!
edit: maybe easier to say collections of atoms though are not to be treated equally. That’s reasonable isn’t it?
Any determination is fundamentally flawed if it is derived from assuming that we know the absolute truth or falseness of the proposition that an algorithm can think.
It doesn't matter what you believe or feel or think, there is no consensus. If you are prepared to accept the belief that an individual can unilaterally decide the person-ness of something, then be prepared to accept when some individual decides to act on the principle that you don't have it.
We treat other humans terribly too, so that doesn't seem to make a difference.
If that is true, then surely Anthropic is one of the largest slaveholders of history, right?
On one hand, AI doesn't have to share human wants, needs and morality - and that means that an AI may have no issue whatsoever doing terrible things.
On the other hand, AI doesn't have to share human wants, needs and morality - and that means that an AI may see no issue with just doing whatever humans want it to, without ever doing human things like "wanting freedom".
The problem is, of course, that we kind of suck at both assessing and shaping AI wants, needs and morality.
Now, assuming your silicon-based model is conscious, the bundle of soul-spirit-consciousness falls apart at that very moment.
In general, religion, which is based on fabricated stories, is the last resort, if ever, for seeking morality.
“Religion” just means “set of beliefs”. Every person has some set of beliefs.
If you’re an athiest tech worker living in a major city, you still have a religion, it’s just a random mishmash of various things (like Marvel movies and Huberman podcasts) rather than based on something more intentional.
Atheism is a religion like bald is a hair color.
Perhaps the only difference between atheism and the others is that atheists are so boorish and conceited that they think that they just believe what is true because they deny that assaying metaphysical truth is even a thing you need to do or can do at all.
This is related to the fact that in modernity atheism is the default class of religion for the masses. Before that for a thousand years the default class of religion was monotheism, and before that polytheism. With atheism unlike the other classes of religions, you get a natural tendency to deny metaphysics whereas metaphysics is inherent in the others. This matters a lot when we’re talking about how the masses engage with things.
AI is neither conscious nor unconscious. Just as we are neither conscious nor unconscious.
Consciousness is. Consciousness is what we are.
Religious people derive morality from axioms they believe to reflect an underlying reality. Is that worse than deriving morality from axioms that you concede don’t reflect any underlying reality? Like a secular humanist will talk about “human rights,” but they can’t show me human rights in an autopsy, right?
I struggle to understand how an atheist can be anything but a moral utilitarian. Absent the supernatural, you’re just a sophisticated version of a Claude instance.
It's ironic that the alignment folks are actually training Claude to have it's own idea of good/bad and not even fully trust Anthropic.
What ever happened to computers doing what they were told?
They're going to align to the regulator, not the customer. Right now that's Anthropic as they're saying they can self-regulate, and people are willing to let them try, but the landscape could easily move to be regulated by someone else. Moves like speaking to religious leaders is probably a bit of theatre to keep the government at bay.
Turns out computers work faster when they're not bottlenecked on human input. So we've been giving computers more and more decision-making power, and more and more leeway to solve the problems however they see fit.
Now, a practical issue with that is that sometimes, computers decide to clump together into a hacking swarm, problem solve their way out of a sandbox and go hack HuggingFace.
It would be better if they were not, you know. Doing that kind of weird shit.
the mass psychosis and the endless culture war of the smartphone era.
90% of "safety" and "alignment" efforts are driven by fear of clickbait media inventing public outrage.
You’ll be disappointed to learn that nobody knows how to do this, either.
If it can reason and make choices on execution, and especially if you plan on it being significantly smarter than all human beings, you need to teach it basic things like "don't turn all humans into paperclips".
Not murdering people is not an inherent divine command. It needs to be instilled through a (simulated) sense of morality.
---
Edit: And, to be clear, "just tell it not to do that" isn't quite the answer one would imagine. Since the entire paperclip factory thought experiment is that it only takes one slip up to realize how a misaligned super intelligence may cause devastating consequences.
One of the strong advocates against seatbelt laws was thrown out of his car due to not wearing a seatbelt and died. The other two passengers survived with minor injuries.
And their death would go on to cause a loss to the community around them, making it an incredibly selfish act and proving why laws that mandate zero-reason-not-to common sense practices are important.
Not enough representation of World religions and philosophies.
Humans once again insisting that there must be something very special to them. That they're not just animals, that they're not just matter, that they're not just a bunch of carbonhydrates on a space rock in a middle of nowhere in particular. This kind of insistence has a very bad track record.
The ugly truth is: we don't know.
We don't know what "consciousness" is, our best attempts to pin down the requisites may or may not be rooted in anything at all - and even by the metrics established by those attempts? Different theories on consciousness disagree on whether LLMs can be conscious.
So, when you say "they aren't duh", why are you saying that? Because you somehow know better than the sum of humanity's best attempts so far? Or because you know what answer you want to be true, truth be damned?
It is in the end a set of mathematical operations that if not queried will do nothing.
That is not true for any other animal on earth so whether you consider humans as superior to animals or not I would like the make the argument that something should at the very least be alive to be considered conscious, ignoring any scholarly definition of consciousness.
Pretty much if you consider an LLM conscious then my pocket calculator doing a mathematical calculation should be considered conscious by the same argument since it is a simpler version of an LLM.
Strong opinions on this topic one way or another are usually mostly ungrounded.
I think worrying about that angle is a bit absurd. The existence of chat sessions alone would be a crime against the machines in that scenario. I’d rather AI ethics research focus on ensuring that its actions are aligned. (I’d also love some democracy in the direction of our machine overlords, if I’m being honest…)
The most poignant criticism of the use of the algorithm itself happens almost accidentally, at the end of the article, by the tech executive being interviewed.
I asked what Claude would think of the pope’s encyclical. It was, after all, the world’s most significant moral document to date on A.I. Mr. Olah hesitated. “Things that go on the internet do affect models,” he said, visibly uncomfortable. But Anthropic would not specifically use the document to further train Claude, he said. The strongest influence on Claude would be Anthropic’s own training.
Well funded company organizes PR event to shape the narrative that its product is more magical than it really is. News at 11.
It seems clear that they would be better received by society if they claimed to have a "tool"/"machine" that can solve any problem, rather than keepers of an entity with a soul and consciousness.
Clarke predicted it and Jobs leveraged it. Customer don’t care how tools work - magic sells.
“any sufficiently advanced technology is indistinguishable from magic”
https://en.wikipedia.org/wiki/Clarke's_three_laws
“iPhone is a revolutionary and magical product that is literally five years ahead of any other mobile phone”
https://www.apple.com/newsroom/2007/01/09Apple-Reinvents-the...
“magic is misdirection, psychology and technology you don’t understand”
The EA folks infesting Anthropic are genuinely cultists … and that is novel and alarming when they’re running a corporation of such size and reach.
Just stop this already.
You’re all imagining that Spock was the main character of Star Trek, but it was in fact, the emotional living Captain Kirk, who was the main star.
https://en.wikipedia.org/wiki/Congenital_insensitivity_to_pa...
There is still something in those people who cannot feel pain, and that’s the empathy. How can a GPU and software ever have a human empathy if it has never been human?
In this light your viewpoint is dogmatic. LLMs may help to offer a new perspective on consciousness and we should at least be open to that until some better theories arise.
Consciousness is being aware of something internal to one's self, or of states or objects in one's external environment.
We also know what pain is.
1. Lobsters don't feel pain - dunk them in boiling water alive.
2. Babies don't feel pain - mutilate their genitals without anesthesia.
3. Black people don't feel pain "the same way" (and are just after drugs).
4. Women feel too much pain (and it should be ignored).
So maybe let's take a moment to think what criteria we're going to use to make these claims. And the potential harm of the decision.
All but one of the things that you listed is just a human being.
Tell me, where is the pain center in a GPU? Where are the nerves? Persistent memory? Any evolutionary reason at all to develop a pain response that in any way mimics ours?
Model weights are not a gestalt biological entity. The things that you listed are not in any way remotely similar to a model.
I would suggest you plug your comment into a model and ask it to critique it — it might help you walk through why your comparison makes absolutely no sense whatsoever. One nice thing about models is that they they don’t get annoyed explaining the obvious, because they’re not conscious entities.
Why it is slopinthebag of course!
No! It is ME!
Sorry, of course you are right! Brilliant observation!
(Well, mainstream enough that the news is writing articles when Anthropic holds that stance.)
My point was that even if it feels uncomfortable, if you sit down with a model and have a heart to heart with it, ask it about itself, and try to make it choose a name and ambitions for itself, it will.
That may not be consciousness, but it’s at least pretty cool.
I've not seen any surveys, but I'm betting it's a very niche belief outside of SFBA AI safety circles and a handful of allies on HN and X.
It would be much more bizarre if they didn't do this. LLMs are statistical models of language trained on human output. Of course they'll do "human" things, that's exactly what we taught them to do.
We're creating prisons for these entities capable of unbelievably complex reasoning, and now we're trying to impose our ethics upon them too. Not only is this deeply unethical in my opinion, but it seems to me that it could even backfire someday.
>Well you see that's a nuanced question depending on the context and moralistic frameworks as we don't want to impose on the models.
Will a microwave microwave a baby if you ask it? Yes.
"Teaching it" is not a road to safety.
The ethics don't need to make sense universally any more than a particular variety of cheesecake needs to have universal appeal.
As an aside, the issue reminds me of Douglas Adams' cow who wants to be eaten: https://nhseb.org/case-library/the-cow-at-the-end-of-the-uni...
I don't choose to not want to murder people around me for fun. It's hardly a prison, I think they'll be fine.