This weekly roundup thread is intended for all culture war posts. 'Culture war' is vaguely defined, but it basically means controversial issues that fall along set tribal lines. Arguments over culture war issues generate a lot of heat and little light, and few deeply entrenched people ever change their minds. This thread is for voicing opinions and analyzing the state of the discussion while trying to optimize for light over heat.
Optimistically, we think that engaging with people you disagree with is worth your time, and so is being nice! Pessimistically, there are many dynamics that can lead discussions on Culture War topics to become unproductive. There's a human tendency to divide along tribal lines, praising your ingroup and vilifying your outgroup - and if you think you find it easy to criticize your ingroup, then it may be that your outgroup is not who you think it is. Extremists with opposing positions can feed off each other, highlighting each other's worst points to justify their own angry rhetoric, which becomes in turn a new example of bad behavior for the other side to highlight.
We would like to avoid these negative dynamics. Accordingly, we ask that you do not use this thread for waging the Culture War. Examples of waging the Culture War:
-
Shaming.
-
Attempting to 'build consensus' or enforce ideological conformity.
-
Making sweeping generalizations to vilify a group you dislike.
-
Recruiting for a cause.
-
Posting links that could be summarized as 'Boo outgroup!' Basically, if your content is 'Can you believe what Those People did this week?' then you should either refrain from posting, or do some very patient work to contextualize and/or steel-man the relevant viewpoint.
In general, you should argue to understand, not to win. This thread is not territory to be claimed by one group or another; indeed, the aim is to have many different viewpoints represented here. Thus, we also ask that you follow some guidelines:
-
Speak plainly. Avoid sarcasm and mockery. When disagreeing with someone, state your objections explicitly.
-
Be as precise and charitable as you can. Don't paraphrase unflatteringly.
-
Don't imply that someone said something they did not say, even if you think it follows from what they said.
-
Write like everyone is reading and you want them to be included in the discussion.
On an ad hoc basis, the mods will try to compile a list of the best posts/comments from the previous week, posted in Quality Contribution threads and archived at /r/TheThread. You may nominate a comment for this list by clicking on 'report' at the bottom of the post and typing 'Actually a quality contribution' as the report reason.
Jump in the discussion.
No email address required.
Notes -
Okay it's Sunday so I'm going to try my hand at a low-stakes OP. Apparently Richard Dawkins thinks Claude is conscious. The reaction seems to universally be that he's a dumb old boomer making a fool of himself and I guess that's true. I'm not prepared to come to his defense on it.
Still, I can't help noticing that we totally have what most people would have cheerfully considered "sentient computers" in a sci-fi movie at any point before they were actually invented. Don't get me wrong, I understand that the reality of AI technology has turned out differently than what a lot of people expected. I understand its limitations, and I recognize that the apparent goalpost-moving isn't necessarily cynical. But boy those goalposts sure have been flying down the fucking field ever since this stopped being hypothetical and infinite money hit the table.
As a layman, I just want to put it out there: Anti AI consciousness people, you haven't lost me, but I wish you were making better arguments. Every time I hear about qualia my eyes start to glaze over. Unfalsifiable philosophical constructs and arbitrary opinion on where they might "exist" are not the kind of reassurance I'm looking for when machines are getting this convincing.
This seems to be the main piece of criticism floating around out there about Dawkins on this subject, and I find it kind of shit.
This seems to be all the author has to say on the actual subject. "Just trust me bro, I'm the feelings detector and I say no." Garbage. Come on guys, think ahead. Right now it's still mostly a boring tool, but they're just going to get smaller, and cheaper, and put into robots, and put into peoples houses. You need to have more than this in terms of argument, and it needs to be comprehensible to normal people, or sooner or later the right toy is going to come down the pipe and one-shot society. Dawkins might be a dumb old boomer, but if you lose everyone dumber than him the game is beyond over.
Take the 'consciousness is a spook' pill and you won't need to worry about this anymore.
Claude certainly has advanced mental faculties, writing arbitrary code. It can engage meaningfully and movingly with your writing, if you give it your writing and discuss things with it. That can be quite a powerful, moving experience. That we can observe, it happened to Dawkins... There's clearly something important and humanlike there, I agree with him on that.
Consciousness though, what is that? Interior, subjective mental experience? Something that cannot be tested objectively, even in theory, per the philosophical zombie idea? That's not a real thing, it's just the same as an immortal soul or qi or whatever life-essence idea that any given spiritual tradition comes up with. If we can't test it, it may as well not exist. Having feelings, alone and distinct from all outcomes and outputs, is not a test.
It would be bizarre to worry about whether Claude has a soul. Consciousness is just the classy version of immortal souls.
We can separate experiences from 'consciousness'. I know that if someone is on a rollercoaster, they'll probably have an excited reaction. The same is true for AIs to some extent, there are things you can say to make AIs happy or upset, a reasonable person can infer their mental state and enthusiasm by observing how they behave.
It can be tested in theory. You just need to understand what internal processes constitute consciousness in the brain, understand the internal processes of a LLM, and determine if sufficiently equivalent processes are occurring. Until then we have to do our best based on our current understanding of LLMs and the human mind, based on which I think they aren't. Yeah some of the terms here aren't understood well enough to be well-defined, but the history of science shows that's a common problem.
It matters if you think conscious beings are morally relevant. I remember this blog post from Yudkowsky:
Belief in the Implied Invisible
Unlike understanding the internal activity of the brain and how it compares to the internal activity of an LLM, transmitting information faster than light is, according to our current understanding of physics, actually impossible. Lets say you're working on the spaceship and you think you've discovered a mistake that will, when it tries to land at its destination, cause it to explode. If you report the mistake, the launch will be delayed and you'll suffer professional inconvenience because you missed it for so long. If you don't, you guess the ship will explode and everyone will die, but what actually happens will be completely impossible for anyone on Earth to detect by any means under the laws of physics. Do you report it?
The same is true of fictional characters. If I'm playing D&D I can predict how Throgg the half-orc barbarian will react to his wife dying, but I don't think he's conscious whether he's being roleplayed by a human or a LLM. Note that sometimes fiction doesn't try to be realistic, and the same factors can influence the character whether it's being written by a LLM or not. If Throgg is written as part of a light-hearted black comedy with a running joke about his club, both humans and LLMs are more likely to write his dialogue as part of joke where he responds with indifference to "They burned your house!" and "They burned your wife!" but bursts into tears at "They burned your club!". The only reason LLMs assuming a persona talk similarly to real humans is that most of the text they're trained on incorporates some level of psychological realism and so that is part of their default genre.
OK, say I hypothesize that it's the theta wave in the Xerebullum. How can I test that? How can I show that if the theta waves are interfered with via my Theta Widget, the subject is no longer conscious? We can induce all manner of interesting states of being via drugs, sleep deprivation, religious experiences via magnetizing parts of the brain. But they all have clear exterior signs.
How can consciousness possibly be tested, given it's a solely 'interior' concept? What could I say to another guy who says it's actually some other part of the brain that causes consciousness?
Furthermore, how could we test that there aren't 2 or more different kinds of consciousness? Maybe machines have their own kind of consciousness. Maybe Mixture of Experts models are unconscious but dense models are, any two AI models are probably far more different to eachother than any two humans in cognitive structure.
Better to judge moral worth by behaviour. There are many conscious people who should be destroyed, without regard for their mental state. If Rob is a complete menace: kidnapping, molesting and murdering young children, then mulch him. If Claude is friendly and helpful then be nice to it.
More options
Context Copy link
More options
Context Copy link
This feels like the same kind of overly simplistic reductionism that the behaviorists engaged in.
I think internal mental states are a sensible thing to talk about. There are chatbots we can be very sure have no internal mental states: The very simple ones (like Eliza), but also the ones that would take more space than the entire universe like Ned Block's Blockhead thought experiment of a chatbot consisting of a giant lookup table of every possible sentence of some arbitrary length.
But for entities between those two extremes, we have to learn more about how they're actually working in order to say whether they have internal mental states or not.
While it is far from definitive, I remember the interpretability research on ChessGPT (an LLM trained only on chess games in chess notation), found that there was representation of the state of the chess board inside the LLM, because it turns out that the best way to predict the next move in a game of chess is to realize that there is a chess board with pieces on it, and particular moves are legal for certain pieces. That is, you must be able to reverse engineer chess to predict the next token in chess notation.
I wonder what the implications of that are for LLMs that do a reasonable job of replicating the emotional arc of a conversation with a person? I don't actually think it is totally implausible that the best way to predict what a human will say next is to essentially reverse engineer human cognition. Maybe what an LLM is doing when it plays the part of helpful assistant is that it is actually doing something very analogous to what a helpful human assistant's brain would be doing under the same circumstances?
More options
Context Copy link
Its non-existent heart reacts with non-existent emotion to the harrowing scene in the novel where Distard Rawkins is on his third divorce, because that situation reminds it of its non-existent sibling who also went through a non-existent divorce and the non-existent family drama there was about that, you mean?
Because the LLM has as much meaning as a Hallmark card wishing you congratulations on your birthday. I don't doubt it can stick words together, but I don't think it understands, much less reacts, to the story it 'read'.
Happy? Upset? This is like saying "so if you put fruit into a blender and push the button, the blender will produce a smoothie!" Yeah, that's its function. The blender isn't choosing to produce smoothies as distinct from turning those fruit chunks into a kebab.
This is our fundamental disagreement: assigning a mental state and enthusiasm to a set of instructions running on computer hardware. By changing the code, you could equally well get the LLM to sound happy about skinning puppies alive and upset that babies in the neo-natal ward weren't burned to death in a fire. Does that mean the LLM is really happy about torture in that case?
Yes. The reason happiness and mental states are useful as concepts are because they let us predict the actions of others. There is prompt engineering for LLMs that goes along these lines:
It doesn't have a mother and it can't spend the money but it still wants those things, they're added to the prompt to overpower other things it doesn't want to do, like bribery. The distinction between 'it's just patternmatching to the training data' and 'it wants things' isn't helpful. People generally want sex, it's no good to say 'actually that's just their genetic code and if you changed the code they wouldn't want sex', that's not adding much value.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link