In your view why doesn't Attention Schema Theory fill in most of the gaps? I see AST as demystifying the key element for me: consciousness is just an information signal and therefore needs a mechanism, an architecture to handle it. That means we can build conscious or unconscious machines by supplying or preventing the awareness of attention signal.
> There is no method we know of that could definitively tell us what beings/objects are conscious and “how much” - if such a thing is even applicable
Not definitively, but anything with information flow fitting the AST could reasonably be considered conscious. Building systems with AST might tell us more. Crude attempt 3 years ago https://arxiv.org/abs/2305.17375
It's as good a theory as any - but from the perspective of this post, it cannot state what is conscious and what isn't or propose any experiment that would definitively answer that. I think Graziano has been pretty upfront about it.
So where is the information signal and why did it evolve? I think AST proposes answers for both: all social organisms that explicitly model self as well as the attention states of other organisms have this "consciousness" information signal via second-order modeling. Basically, consciousness doesn't come for free, it requires a plausible evolutionary history creating a series of complex mechanisms with critical parts that can be observed and measured, even if only in theory at this point in time.
The experiments being done on AST seem less like grand experiments looking for grand proof and more like testing the various parts and mechanism of a complex system, steadily building a stronger hypothesis; i.e neural imaging on attention and awareness (several like this https://doi.org/10.1073/pnas.2307584120), and now exploring deep-neural-networks AST architectures (https://arxiv.org/abs/2411.00983, https://arxiv.org/abs/2509.16058v1).
So at this point in time, what is conscious via AST? Answer: social organisms that both model self and others, probably. All others lacking this kind of sophisticated architecture are not, probably. Experiments with brain simulation will be very helpful here (Exascale clusters now exist?), and DNN experiments with LLMs are proving to be a hugely useful.
In Graziano's words:
"Take a step back and look at it from a distance: Imagine that zombie aliens come to visit the Earth from another planet, and they learn that some of us share a very adamant belief in something as yet never before physically detected called qualia. With no institutional review board to stand in their way, they rifle around in human brains with sophisticated equipment, and, as Leibniz predicted, they find nothing but neurons and glia and the like. And at the same time, they conclude that input→processing→output is enough to account for, well, pretty much everything else, including the belief in qualia itself. What might those visitors from another world rightfully conclude about this extra something that we call ‘qualia’? AST explains that conviction as a necessary by-product of a useful mechanism: an internal model of selective attention. With a viable explanation like this in hand, and Occam’s razor as a guiding principle, what should we conclude? "
In your view why doesn't Attention Schema Theory fill in most of the gaps? I see AST as demystifying the key element for me: consciousness is just an information signal and therefore needs a mechanism, an architecture to handle it. That means we can build conscious or unconscious machines by supplying or preventing the awareness of attention signal.
> There is no method we know of that could definitively tell us what beings/objects are conscious and “how much” - if such a thing is even applicable
Not definitively, but anything with information flow fitting the AST could reasonably be considered conscious. Building systems with AST might tell us more. Crude attempt 3 years ago https://arxiv.org/abs/2305.17375
It's as good a theory as any - but from the perspective of this post, it cannot state what is conscious and what isn't or propose any experiment that would definitively answer that. I think Graziano has been pretty upfront about it.
> it cannot state what is conscious or what isn't propose any experiment that would definitively answer that
On a very basic level, all organisms that have an information signal reporting consciousness are conscious, I think that is logically sound (i.e. see https://grazianolab.princeton.edu/sites/g/files/toruqf3411/files/graziano_iit_comment.pdf "Consciousness is already solved"). That at least moves consciousness into the realm of math, physics, science.
So where is the information signal and why did it evolve? I think AST proposes answers for both: all social organisms that explicitly model self as well as the attention states of other organisms have this "consciousness" information signal via second-order modeling. Basically, consciousness doesn't come for free, it requires a plausible evolutionary history creating a series of complex mechanisms with critical parts that can be observed and measured, even if only in theory at this point in time.
The experiments being done on AST seem less like grand experiments looking for grand proof and more like testing the various parts and mechanism of a complex system, steadily building a stronger hypothesis; i.e neural imaging on attention and awareness (several like this https://doi.org/10.1073/pnas.2307584120), and now exploring deep-neural-networks AST architectures (https://arxiv.org/abs/2411.00983, https://arxiv.org/abs/2509.16058v1).
So at this point in time, what is conscious via AST? Answer: social organisms that both model self and others, probably. All others lacking this kind of sophisticated architecture are not, probably. Experiments with brain simulation will be very helpful here (Exascale clusters now exist?), and DNN experiments with LLMs are proving to be a hugely useful.
In Graziano's words:
"Take a step back and look at it from a distance: Imagine that zombie aliens come to visit the Earth from another planet, and they learn that some of us share a very adamant belief in something as yet never before physically detected called qualia. With no institutional review board to stand in their way, they rifle around in human brains with sophisticated equipment, and, as Leibniz predicted, they find nothing but neurons and glia and the like. And at the same time, they conclude that input→processing→output is enough to account for, well, pretty much everything else, including the belief in qualia itself. What might those visitors from another world rightfully conclude about this extra something that we call ‘qualia’? AST explains that conviction as a necessary by-product of a useful mechanism: an internal model of selective attention. With a viable explanation like this in hand, and Occam’s razor as a guiding principle, what should we conclude? "
From https://academic.oup.com/nc/article/2022/1/niac001/6523097
Not "definitive proof", agree with you there; but AST seems more to me than a complete lack of understanding of consciousness.