My first Christmas Paper 🎄😅 doi.org/10.1093/nc/n... - ThreadSky

johanneskleiner.bsky.social • 60 days ago

My first Christmas Paper 🎄😅
https://doi.org/10.1093/nc/niae037

Comments

Interesting read! As someone strongly attached to a nonphysicalist metaphysics, I’m not 100% convinced that the line of reasoning is so independent of the ontological frame …

maxine.science•59 days ago

eg, the presupposition, assuming instead a spiritual monist world, of a design guarantee on what would then be the dynamics of “illusory physical silicon” seems … odd, as a notion, right?

maxine.science•59 days ago

I would argue that we can’t make such guarantees. For example, that guarantee would preclude the (prima materia) mental activity of the AI from influencing humans to then modify the AI, manifesting as an alteration of the chips’ (“illusory physical”) dynamical regime.

maxine.science•59 days ago

This would violate the assumption “Verification of PUs ensures that any dynamical effects that change the computational dynamics of a processing unit are precluded or suppressed”: the verification may hold for an instantaneous physical projection, but not the underlying mental-social reality.

maxine.science•59 days ago

Would be interesting to talk more about it sometime though! 😄

johanneskleiner.bsky.social•59 days ago

Interesting! A big issue in the review process was to show to the reviewers that dynamical relevance does *not* require non-physicalist metaphysics, the opposite worry. My take is that dynamical relevance is in some sense orthogonal to the usual metaphysical debates. We attempt to explain this

roncline33.bsky.social•60 days ago

This paper is spot on from the perspective of this retired silicon hardware engineer. See this 7-minute digital designer's rationale for why dynamical relevance is indeed "relevant" for continuous systems like the brain: https://bsky.app/profile/roncline33.bsky.social/post/3lau2hgwhos2q

adewebb.bsky.social•60 days ago

Bloody good paper. Fulfils the objective of making itself accessible to a non-specialist audience very well indeed 🙌🏻

dkarpa.bsky.social•60 days ago

Publishing on Christmas is a no-go

nomad-heart.bsky.social•60 days ago

Thank you for sharing! It is very interesting. And I didn’t know about non-go theorem, too.

Congrats with post Christmas publication:)!

lenoreblum.bsky.social•57 days ago

It seems that your theorem implies that a brain in a vat cannot be conscious. My point is: you seem to be presuming that AI systems are isolated in vats. Many theories of consciousness include (explicitly or implicitly) the necessity of having some aspects of the 4Es (being embodied, embedded, ...

lenoreblum.bsky.social•57 days ago

... enacted, extended) which brains in vats do not have (unless some of these effects are already built-in). Clearly AI systems can be built to be embodied, to be embedded in and to enact in their environments, and to be extended. This will definitely change the computational dynamics of the system.

lenoreblum.bsky.social•57 days ago

I am surely missing something. Would appreciate some clarification. Many thanks!

johanneskleiner.bsky.social•57 days ago

Thanks for the feedback, Lenore! 🙏

Our result doesn't apply to brains, because brains are not verified: they are not designed to follow an externally defined instruction set architecture.

Does this help with the question?

It also doesn't apply to systems that do mortal computation.

lenoreblum.bsky.social•57 days ago

So what about AI systems that are embodied, embedded, enacted and extended?

johanneskleiner.bsky.social•57 days ago

That's an excellent question! It depends on what exactly these terms mean. For example, if "extended" indeed means that some of the computation is actually carried out outside of the system, verification doesn't apply there, and therefore our result doesn't either.

Comments

Posting Rules

Reply