visimo-dino

@funplings @__drewface *Defining* consciousness isn't the problem. I have a good definition: it's the thing you and I are both subjectively experiencing right now. The problem is we don't know HOW certain matter/processes give rise to consciousness. Without this, we can't say if any AI is conscious.

@__drewface have*

I'm very confident we can prove whether or not AI is "conscious" or "ensouled", and under exactly what circumstances it is, if so. We built these things, we know the physics, we can do the math. The problem is, on Twitter, we don't have a definition for conscious or ensouled.

@__drewface The problem isn't "we don't have a definition for consciousness on Twitter." It's that we don't know how certain matter/processes give rise to consciousness. W/o knowing how they do, we can't say if any AI is conscious, regardless of how well we understand its matter/processes.

@ArtemisConsort Right, but isn't it true to say of metaphysics that it isn't about what things *do*, but rather what things *are*, at the most foundational, ultimate level? Science is about how things behave, while metaphysics is about the true nature of reality/existence - no?

@VisimoD This would be an argument against metaphysics. If the exact same pattern can be instantiated on different substrates, then the substrates aren’t doing anything, including causing us to talk about the substrate, since that’s also a physical event.

If you really chew on what “materialist” means, it ends up being a claim not about ontology but epistemology: that the world can be fully described by knowable mathematical laws. It’s not about the nature of the “stuff”, but rather the relations it obeys.

@JeanDorsMoisson @GaryMarcus @RichardSSutton @HamidMaei Not only does he think it's debatable, I believe he considers it outright false that LLMs are "bitter lesson-pilled," so to speak. He made that pretty clear on Dwarkesh's podcast, where he said (as you noted) that LLMs still rely almost entirely on human knowledge.

@GaryMarcus @RichardSSutton @HamidMaei Actually Sutton thinks it is debatable that LLM follow the bitter lesson exactly for this reason. Learning from human text is a form of Human knowledge/bias.

The bitter lesson in 26 words: Don’t be distracted by human knowledge, as AI has been historically. Instead focus on methods for creating knowledge that scale with computation, like search and learning.

@PawelHuryn @GaryMarcus @RichardSSutton @HamidMaei "Information isn't knowledge" feels like a shaky proposition. I can see an argument that not *all* information is knowledge, but surely some information is, and conversely I think *all* knowledge is made of information.

Information isn't knowledge. AlphaFold didn't have human folding rules baked in. It had the PDB and discovered the rules by scaling compute on it. Most of what gets called "knowledge in the harness" is information surfaces. The knowledge is what the system produces from them by surfacing patterns, hypothesizing, and testing.

@togelius @GaryMarcus @RichardSSutton @HamidMaei I disagree that human knowledge will always be essential. What Sutton wants is a system that can be turned loose in the world and simply learns, entirely on its own, through experimentation, play, trial-and-error. No human knowledge is needed for that, in principle.

@GaryMarcus @RichardSSutton @HamidMaei Trying to avoid human knowledge is a strange idea, when you think about it. If these systems operate in the world that humans built, human knowledge will always be essential. The question is in which form it is encoded in the system. There are many different answers to this.

@PaoloMiasma @m_wacker Apologies Paolo, I misread your previous post. I thought you said "I hate to break it to you, but AI is just an algorithm." Probably bc I read that exact take about 100x per day 😄 Sorry for the misunderstanding! I actually agree with the underlying point you were making.

@VisimoD @m_wacker Okay, so it's more sophisticated than humans.

What is AI? It's literally a giant math equation. A giant math equation is not "conscious." (To be a bit more precise, it's a giant math equation that is differentiable. You can't use gradient descent if you can't calculate the gradient, or if the gradient is 0.)

@jojosuzieks @m_wacker No idea what this screenshot is supposed to be showing me, but one thing I can say for certain is that it’s not proving your earlier comment correct.

@jojosuzieks @m_wacker Did you really write “So no response..” because I didn’t respond to your X post within 40 minutes of you writing it?

@jojosuzieks @m_wacker It already has happened, using an LLM architecture.

@PaoloMiasma @m_wacker It’s not an algorithm. It’s an ML model that’s been trained on data, a specific type of neural network, to be exact. More like an unfathomably large mathematical expression than an algorithm.

@m_wacker Hate to break it to you, but you are just an algorithm.

@GhostRhodesia @ArtemisConsort To put it succinctly: is the chair ACTUALLY made of particles, while the appearance of the chair in consciousness is just an illusion? Or is the appearance of the chair in consciousness the ACTUAL chair, and it seeming to be made of particles is only an illusion?

@GhostRhodesia @ArtemisConsort You’re contradicting yourself though. You said it’s obvious the chair is really an appearance in consciousness, but then said that “Sure, the particles are moving faster in reality.” This suggests you believe that *in reality*, there ARE particles. Which is the opposite belief.

@flavicomus @GhostRhodesia @ArtemisConsort No it can’t. Under materialism, an *appearance* of a chair occurs in consciousness, but this is not the chair; the chair is the inanimate particles that make up its atoms. Under idealism, the chair actually IS the appearance in consciousness, and there are no actual particles.

@VisimoD @GhostRhodesia @ArtemisConsort That's not a philosophical question lol. Chair can easily be both.

@jojosuzieks @m_wacker Your comment is (1) Not relevant to my point, which was about the physical implementation of an AI system, not its capabilities; (2) Not an argument, just a bare statement with zero support behind it; and (3) Wrong anyway, as even current LLMs have innovated—let alone future AIs.

@VisimoD @m_wacker Lol nope it is not at all like a human brain. It is incapable of innovation and original thought

@GhostRhodesia @ArtemisConsort This isn’t metaphysics, though. You’re describing what objects are in the ordinary sense of “are.” The metaphysical question is, once we’ve agreed on what “chair” means, is the thing we agree is “chair” an appearance in consciousness, or particles of inanimate physical matter?

“What is electron” “It does this” “okay but what IS it?” I think this is stupid. Let’s bring it to the real world “What is a chair” “A chair is a thing made of wood with four legs” “some chairs are metal, some chairs have one single leg. are they not chairs? a stepstool has four legs, is it a chair?” In reality, we define things by their function. A chair is a chair because it is meant to be sat on and a stool is a stool because it’s meant to be stepped on. That’s why your identity is derived from your actions (what you do)

@_virgil19 @taijitu_sees @speakerjohnash Yes, another way of saying this is that consciousness is irreducibly subjective. Which is a statement that I think captures and clarifies much of what this conversation/debate is about.

@VisimoD @taijitu_sees @speakerjohnash explanation is always a third person map, experience is the one first person thing. no amount of map detail ever turns into the terrain

@VisimoD @speakerjohnash I'm not a chair, but I suspect a chair can experience me sitting in it. If I'm hefty enough, it might even break!

@Bigstanz5000 @DrPhiltill @MarkHowsham @todayyearsold The one that’s established in the problem. That’s what I (and others) have been repeatedly saying. I’m not saying my choice of definition is right and yours is wrong; I’m just saying this conversation has been quibbling about the definitions, i.e. it actually is semantics.

@VisimoD @DrPhiltill @MarkHowsham @todayyearsold It's using geometric terms. Its a math word problem, that's the context. The problem establishes the definition as a geometric one. What other definition should we use?

Am I stupid or is that literally correct