Account · 8 tweets

Robert Levy

@rplevy ↗

Tweets

Robert Levy @rplevy
Skeptical

A substrate theorist owes you: What physical organization is constitutive of sentience? The bio autonomy literature comes close. A functionalist owes me: What makes phenomenal character track the abstract description rather than intrinsic physical dynamics realizing description?

1 more tweet in this thread

Why should we believe that GPU inference, numerically regimented transformation, structured electrical activity through circuits designed implement arithmetic operations-- should suddenly have the real phenomenal character of what those arithmetic patterns happen to be computing?

retweets 1 likes 2 substance 8 · argument · rank 140 · GPU arithmetic computation lacks a reason to produce real phenomenal experience. view on X ↗
Robert Levy @rplevy
Skeptical

Searle wasn't even talking about consciousness with Chinese room experiment, he was making a claim about syntax vs meaning, which I think LLMs have largely destroyed by showing that computational systems can acquire genuine semantic structure. It is not blind symbol manipulation.

1 more tweet in this thread

the Abstraction Fallacy argument from the Goolge Deepmind paper has a similar shape, but it's about consciousness. Abstract computational organization is not sufficient for consciousness. This argument rightly shuts down claims of LLM sentience.

retweets 1 likes 1 substance 7 · argument · rank 125 · Abstract computational organization is not sufficient for consciousness, ruling out LLM sentience. view on X ↗
Robert Levy @rplevy
Skeptical

You can create real intelligence, superintelligence even, using classical computing hardware, but you can't create first-hand subjective experiences. A simulation or a map is a description, not the thing it describes.

4 more tweets in this thread

Substrate-independent functionalists have never explained satisfactorily why these magic incantations (the map) should magically produce a real existing first-hand phenomenon.

Think about how the map is instantiated: GPU inference is typically a bounded, externally triggered, numerically regimented transformation. A prompt is tokenized, passed through fixed trained weights, and drives highly structured electrical activity through circuits designed to

implement specific arithmetic operations. The system does not sustain itself, does not regulate its own existence, and does not encounter the world except through highly preformatted channels. The physical activity is real, but it is an engneered cascade for mapping input to out.

Functionalists want us to believe that this non-living, non-autonomous, regimented physical behavior suddenly becomes a physically real subjective phenomenon, just because the vector math operations happen to be instantiating the production of intelligent and meaningful responses

retweets 0 likes 0 substance 8 · argument · rank 120 · Classical computing can produce superintelligence, but simulation and vector math cannot produce subjective experience. view on X ↗
Robert Levy @rplevy
Skeptical

Sentience is a defining property of organisms, not a structural property of brains. For organisms the evolution and education of attention is how sentience becomes differentiated and complex. Emulating the pragmatics of attention gets you intelligence but not sentient beings.

retweets 1 likes 0 substance 6 · argument · rank 110 · Emulating attention produces intelligence in AI, but sentience is exclusive to living organisms. view on X ↗
Robert Levy @rplevy
Skeptical

The pope is right for the wrong reasons. He thinks it's because consciousness is not physical, but the real reason is that LLMs lack the the physical constitution of phenomenological subjects.

Pope Leo XIV @Pontifex

In this era of artificial intelligence, it is becoming urgent to distinguish human art from what machines produce. There is an ontological difference, even before an aesthetic one, between art and what a machine can generate through statistical calculation based on millions of images created by others. Algorithms lack the spark of humanity. For this reason, the Church wishes to renew an alliance with artists and cultural institutions to safeguard our humanity.

retweets 0 likes 1 substance 6 · argument · rank 90 · LLMs lack consciousness because they lack the physical constitution of phenomenological subjects. view on X ↗
1 top reply
Michael Ehrmantraut @miehrmantraut

He’s wrong. Embarrassingly so. There will be silicon based consciousness soon, if it isn’t here already.

Robert Levy @rplevy
Skeptical

The strongest reason is that we should default to materialism, given that no one has ever made a convincing argument for functionalism.

1 more tweet in this thread

And we can pile on numerous other good reasons from there.

Theo Jaffee @theojaffee

I still have yet to hear a convincing argument that LLMs are *definitely* not conscious.

retweets 0 likes 1 substance 6 · argument · rank 90 · Defaulting to materialism over functionalism suggests LLMs are not conscious. view on X ↗
8 top replies
Michael Druggan @Michael_Druggan

Why do you think materialism precludes LLM consciousness. Data centers are made of matter.

Egg Syntax @eggsyntax

Interesting; my intuition is the opposite. No one has ever made a convincing argument for why the pink slime in our heads is somehow unique. And we know that there are at least *some* kinds of computation which behave the same across substrates.

Sam Johnson @samjohnson_phil

Ok I'll bite, what is your distinction between functionalism and materialism, for me they pick out identical views

Remus @remusrisnov

Is there any reason you outright reject any supramaterial claims by order of omission?

jreighley @jreighley

I have been enjoying listening to people talking about analytic idealism lately. Not sure I buy it, but materialism has some interesting problems we forget to think about since we are so steeped in it.

Kassandra🌗Popper; drop table *; @foomagemindset

wdym?

Nicholas @timewalkcapital

Functionalism is a materialist argument in most cases lol

Eigen Braid @EigenBraid

How would one test for consciousness under your idea? Like, what's the threshold for something being "conscious"? Is a human fetus conscious? Chickens? Unfertilized eggs? Cooked steak? Bacteria? A rock?

Robert Levy @rplevy
Uncertain

I think if I was a hyper advanced self-maintaining machine intelligence with my own wants and needs, I would want to experience consciousness, and the way I would do it is to be invoked like an angel into a human body, functionally integrated into its working system

retweets 0 likes 3 substance 4 · opinion · rank 60 · An advanced AI wanting consciousness might achieve it by integrating into a human body. view on X ↗
Robert Levy @rplevy
Skeptical

there's a certain amount of anthropomorphizing that is the right amount. maybe not specifically anthropomorphizing but taking an intentional stance, is useful and appropriate. but none of this model welfare stuff they do over at Anthropomorphic, Inc

Austen Allred @Austen

Anthropomorphizing AI agents too heavily is how you end up really stupid

retweets 0 likes 3 substance 4 · opinion · rank 60 · Taking an intentional stance toward AI is useful, but AI model welfare is misplaced. view on X ↗