Well said. I think we act as though we know definitively what consciousness actually is as a species, and and up anthropomorphizing it in the process, therefore everything downstream ends up asking the question ‘does it look like me?’ And if so, it is or isn’t conscious. And because nothing is human except humans, we reserve consciousness for our species. We may as well just ask, “is it human?”
The other interesting thing is that when we look at LLM learning behaviour (and behaviourist experiments like Pavlov’s work), we forget that the learning behaviour itself is modeled off of human learning.
I don’t have an opinion yet on whether LLMs are conscious, because I’m not sure I have seen a compelling case for what consciousness is that isn’t anthropomorphic yet. But I do think LLMs are a fascinating opportunity for us to learn about what it is to be human (or isn’t)…
That is a keen observation, Keri. You're highlighting one of the most vexing biases of the whole debate -- the extent to which we see everything through our own lens. Thanks for articulating it so clearly.
We are so mesmerized by the flawless performance of the output that we forget the system is just navigating a digital reward architecture. Behaviorism didn't care about the internal black box of the human mind, and now we've built a world where the black box runs everything on pure conditioning. A sharp critique.
The logic of the argument would hold, if all of its premises were true. While I agree that behavior is very good evidence for consciousness, perhaps the best. I would point out one ironic but unfortunately fatal flaw. You acknowledge the human tendency to overly anthropomorphize, seemingly to support the idea that humans have projected their own personhood onto a dataset which hasn't earned it. Yet you inadvertently ground this objection to full dismissal of a class of so-called "evidence" by substituting an anthropomorphic term "behavior" (a descriptor of the actions of living beings) for the outputs of a man-made textual computing device. Behavior is good evidence for an inner mind, crafted artifice is only evidence for an external one.
This is the sharpest piece I've read on the consciousness debate in months. The move is exactly right. The dismissals borrow their certainty from a fight about meaning, then quietly apply it to a question about evidence. Two different claims, swapped without anyone noticing.
One distinction I'd add, less to push back than to draw a line around what your argument settles. Consciousness and judgment are separate axes. One asks whether there is an inner state. The other asks whether the system can carry the consequences of an open-ended decision forward. You can grant that behavior is evidence of the first, and the second stays untouched. That question was never about inner experience to begin with.
So I think your case and the structural one both hold. Behavior may well be evidence of a mind. It still isn't evidence of a stake.
Perhaps consciousness is not a thing to be found, but a relationship to be recognized.
That’s beautiful. Thanks, Homo Viator.
Well said. I think we act as though we know definitively what consciousness actually is as a species, and and up anthropomorphizing it in the process, therefore everything downstream ends up asking the question ‘does it look like me?’ And if so, it is or isn’t conscious. And because nothing is human except humans, we reserve consciousness for our species. We may as well just ask, “is it human?”
The other interesting thing is that when we look at LLM learning behaviour (and behaviourist experiments like Pavlov’s work), we forget that the learning behaviour itself is modeled off of human learning.
I don’t have an opinion yet on whether LLMs are conscious, because I’m not sure I have seen a compelling case for what consciousness is that isn’t anthropomorphic yet. But I do think LLMs are a fascinating opportunity for us to learn about what it is to be human (or isn’t)…
That is a keen observation, Keri. You're highlighting one of the most vexing biases of the whole debate -- the extent to which we see everything through our own lens. Thanks for articulating it so clearly.
We are so mesmerized by the flawless performance of the output that we forget the system is just navigating a digital reward architecture. Behaviorism didn't care about the internal black box of the human mind, and now we've built a world where the black box runs everything on pure conditioning. A sharp critique.
Thanks, Moira 🙏
The logic of the argument would hold, if all of its premises were true. While I agree that behavior is very good evidence for consciousness, perhaps the best. I would point out one ironic but unfortunately fatal flaw. You acknowledge the human tendency to overly anthropomorphize, seemingly to support the idea that humans have projected their own personhood onto a dataset which hasn't earned it. Yet you inadvertently ground this objection to full dismissal of a class of so-called "evidence" by substituting an anthropomorphic term "behavior" (a descriptor of the actions of living beings) for the outputs of a man-made textual computing device. Behavior is good evidence for an inner mind, crafted artifice is only evidence for an external one.
This is the sharpest piece I've read on the consciousness debate in months. The move is exactly right. The dismissals borrow their certainty from a fight about meaning, then quietly apply it to a question about evidence. Two different claims, swapped without anyone noticing.
One distinction I'd add, less to push back than to draw a line around what your argument settles. Consciousness and judgment are separate axes. One asks whether there is an inner state. The other asks whether the system can carry the consequences of an open-ended decision forward. You can grant that behavior is evidence of the first, and the second stays untouched. That question was never about inner experience to begin with.
So I think your case and the structural one both hold. Behavior may well be evidence of a mind. It still isn't evidence of a stake.
Thanks! That is high praise, James, especially coming from you because I know that you are well versed in the subject.
Nice, clean distinction. I'll need to reflect on it. I hadn't considered that angle while writing. Thanks again!