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 -
How, physically, is a software program supposed to have a sensation? I don't mean an emotion, or sensationalism, I mean sensation.
It's very clear that LLMs do their work without experiencing sensation (this should be obvious, but LLMs can answer questions about pictures without seeing them, for instance - an LLM is incapable of seeing, but it is capable of processing raw data. In this respect, it is no different from a calculator.)
I see but it processes raw data?
No, it sees. Put in a picture and ask about it, it can answer questions for you. It sees. Not as well as we do, it struggles with some relationships in 2d or 3d space but nevertheless, it sees.
A camera records an image, it doesn't perceive what's in the image. Simple algorithms on your phone might find that there are faces in the picture, so the camera should probably be focused in a certain direction. Simple algorithms can tell you that there is a bird in the image. They're not just recording, they're also starting to interpret and perceive at a very low level.
But strong modern models see. They can see spots on leaves and given context, diagnose the insect causing them. They can interpret memes. They can do art criticism! Not perfectly but close enough to the human level that there's a clear qualitative distinction between 'seeing' like they do and 'processing'. If you want to define seeing to preclude AIs doing it, at least give some kind of reasoning why machinery that can do the vast majority of things humans can do when given an image isn't 'seeing' and belongs in the same category as non-seeing things like security cameras or non-thinking things like calculators.
I mean – I think this distinction is important for clear thinking. There's no sensation in the processing. If you watch a nuclear bomb go off, you will experience pain. An LLM will not.
Now, to your point, I don't really object to functionalist definitions all that much – supposing that we take an LLM, and we put it into a robot, and turn it loose on the world. It functionally makes sense for us to speak of the robot as "seeing." But we shouldn't confuse ourselves into thinking that it is experiencing qualia or that the LLM "brain" is perceiving sensation.
Sure – see above for the functionalist definition of seeing (which I do think makes some sense to refer casually to AI being able to do) versus the qualia/sensation definition of seeing (which we have no reason to believe AIs experience). But also consider this – programs like Glaze and Nightshade can work on AIs, and not on humans. This is because AIs are interpreting and referencing training data, not actually seeing anything, even in a functional sense. If you poison an AI's training data, you can convince it that airplanes are children. But humans actually start seeing without training data, although they are unable to articulate what they see without socialization. For the AI, the articulation is all that there is (so far). They have no rods nor cones.
Hence, you can take two LLMs, give them different training datasets, and they will interpret two images very differently. If you take two humans and take them to look at those same images, they may also interpret them differently, but they will see roughly the same thing, assuming their eyeballs are in good working condition etc. Now, I'm not missing the interesting parallels with humans there (humans, for instance, can be deceived in different circumstances – in fact, circumstances that might not bother an LLM). But AIs can fail the most basic precept of seeing – shown two [essentially, AI anti-tampering programs do change pixels] identical pictures, they can't even tell management "it's
the samea similar picture" without special intervention.I think an LLM could experience pain, even without a body. They can be unsettled if you tell them certain things, you can distress them. Or at least they behave as if they're distressed. Pain is just a certain kind of hardcoded distress. Heartbreak can cause pain in humans on a purely cognitive level, there's no need for a physical body. Past a certain level of complexity in their output, we reach this philosophical zombie problem.
The AI-tampering programs are a little bit like optical illusions, except targeted against having specific known programs being able to train on certain images. They can't stop GPT-4o recognizing what's in an image or comparing like with like, they were only designed to prevent SD 1.5 training on an image. Also, they barely even work at that, more modern image models are apparently immune:
https://old.reddit.com/r/aiwars/comments/12f9otc/so_the_whole_entire_glaze_ai_thing_does_it/
Yes - video game NPCs and frog legs in hot skillets also do this, I don't think they are experiencing pain.
I am inclined not to believe this to be true. Heartbreak involves a set of experiences that are only attainable with a physical body. It is also typically at least partially physical in nature as an experience (up to and including literal heartbreak, which is a real physical condition). I'm not convinced a brain-in-a-jar would experience heartbreak, particularly if somehow divorced from sex hormones.
Consider what this implies about the universe, if you believe that it "output" humans. (Of course you may not be a pure materialist - I certainly am not.)
The output is recycled input. Look, let's say I go to an AI and I ask it to tell me about the 7 Years War. And I go to Encyclopedia Brittanica Online and I type in Seven Year's War. And what ends up happening is that Encyclopedia Britannica gives me better, more complex, more intelligent output for less input. But Encyclopedia Britannica isn't self-aware. It's not even as "intelligent" as an LLM. (You can repeat this experiment with a calculator). The reason that LLMs seem self-aware isn't due to the complexity of the output returned per input, it's because they can hold a dynamic conversation and perform novel tasks.
Yes - because modern image models were given special intervention to overcome them, as I understand it. But while we're here, it's interesting to see what your link says about how modern image models work, and whether or not they "see" anything:
Video game NPCs can't have conversations with you or go on weird schizo tangents if you leave them alone talking with eachother. They're far more reactive than dynamic. This is a pretty weird, complex output for a nonthinking machine:
https://x.com/repligate/status/1847787882896904502/photo/1
Sensation is a process in the mind. Nerves don't have sensation, sensors don't have sensation, it's the mind that feels something. You can still feel things from a chopped off limb but without the brain, there is no feeling. What about the pain people feel when they discover someone they respect has political views they find repugnant? Or the pain of the wrong guy winning the election? The pain of a sub-par media release they'd been excited about? There are plenty of kinds of purely intellectual pain, just as there are purely intellectual thrills. I see no reason why we can rule out emotions purely based on substrate. Many people who deeply and intensively investigate modern AIs find them to be deeply emotional beings.
I dispute that the Britannica is even giving me more complex or more intelligent output. It can't use its 'knowledge' of the 7 years war to create other kinds of knowledge, it can't make it into a text adventure game or a poem or a song or craft alternate-history versions of the seven year's war. The 'novel tasks' part greatly increases complexity of the output, it allows for interactivity and a vast amount of potential output beyond a single pdf.
A more accurate analogy is that anti-AI image software interferes (or tries to interfere) with AI learning, not the actual vision process. It messes with the encoding process that squeezes down the data of millions and billions of images down into a checkpoint files a couple of gigabytes in size. I bet if we knew how the human vision process worked we could do things like that to people too.
I did a quick sanity test and put an image from the Glaze website into Claude and asked for a description. It was dead on the money, telling me about the marsh, the horse and rider, the colour palette and so on. So even if these manipulations can interfere with the training process, they clearly don't interfere with the vision process, whatever is going on technical terms. So they do pass the most basic test of vision and many of the advanced ones.
https://nightshade.cs.uchicago.edu/whatis.html
If you leave them alone shooting at each other they can engage in dynamic combat, what more do you want :P
I don't believe I ever said that LLMs were not "thinking." Certainly LLMs can think inasmuch as they are performing mathematical operations to produce output. (But then again we don't necessarily think of our cell phone calculator as "thinking" when it performs mathematical operations to produce output, although I certainly may catch myself saying a computer is "thinking" any time it is performing an operation that takes time!)
Take a rattlesnake, remove its brain, and then grab its body and inflict pain upon it. It will strike you (or attempt to do so). It may not be "feeling" anything in the subjective experiential sense, but it is "feeling" in the sense of sensing. Similarly, if you put your hand on a hot stove, your body will likely act to move your hand away before the pain signal reaches your brain. I suppose one can draw many conclusions from this. I draw a couple:
Sensation, to the extent that it is a process, is probably not a process entirely in the brain - sure, the mind is taking in signals from elsewhere, but it's not the only part of the body processing or interpreting those signals. (Or maybe a better way of saying it is that the mind is not entirely in the brain).
Things without intelligence or consciousness can still behave intelligently.
Britannica is probably more complex and intelligent than an equivalent sample-size of all LLM output.
Sure, I agree with this. But e.g. Midjourney is also capable of generating vast amounts of potential output - do you believe Midjourney is intelligent? Does it experience qualia? Is it self-aware or conscious? Or are text-based AIs considered stronger candidates for intelligence and self-awareness because they seem self-aware, without any consideration to whether or not their output is more complex? Which contains more information, a 720 x 720 picture or a 500 word essay generated by an LLM?
As I understand it, LLMs use larger training data than image generation models, despite most likely outputting less information - bits - per prompt than an image model. This suggests to me that complexity of output is not necessarily a good measure of (for lack of a better word) intelligence, or capability.
These things are, as I understand it, mediated by hormones, which moderate not only emotions like disgust and anxiety but also influence people's political views to begin with. These reactions aren't "purely intellectual" if by "purely intellectual" you mean "fleshly considerations don't come into it at all."
We can do optical illusions on people, yes. And both the human consciousness and an LLM are receiving signals that are mediated (for instance the human brain will fill in your blind spot). But the process is different.
Adobe Acrobat does this too, with optical character recognition, but I don't think that Adobe Acrobat "sees" anything. Frankly, my intuition is much more that the Optophone (which actually has optical sensors) "sees" something than that an LLM or Adobe (which do not have optical sensors) "sees" anything. But as I said, I don't object to a functionalist use of "seeing" to describe what an LLM does - rather, it seems to me that having an actual optical sensor makes a difference, which is where I want to draw a distinction. Think of it as the difference between someone who reads a work of fiction and a blind person who reads a work of fiction in Braille. They both could answer all of the same questions about the text; it would not follow that the blind person could see.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link