Current structure of my sequence on C. metaphysics:
Post #1: introduce
the four ModelsPost #2: discuss simplicity, the justification behind it, and formalize it as well as possible. Solomonoff Induction doesn't quite work here for complicated reasons.
Post #3: apply this to our four models. Eliminate 1 and 2 (but frame the elimination of 2 as a reinterpretation of 2 as 3).
Post #4: introduce
Marr's Levels of Analysis. Use this to split Model 3 into 3.1 (implementation level), 3.2 (algorithmic level), 3.3 (computational level). Eliminate 3.3 based on various arguments, many involving
GLUTs( so now the "introduction" is over, and we can talk about the hard part, i.e., matter vs. bits, Model 3.1 vs 3.2 )Post #5: Introduce the Binding Problem and its basic properties, including its causal effect, why you can't ignore it, and why it's not easy
Post #6: Discuss in agonizing detail all the reasons why two people talking about an image aren't going to solve the Binding Problem for vision
Post #7: Introduce regular and convolutional neural networks for those not familiar, make the point that they really don't do anything different from humans talking about images they've seen, conclude that they won't solve the binding problem either
Post #8: review algorithmic theories of consciousness from the literature -- global workspace theory, higher order theory, integrated information theory -- with a focus on how they address the Binding Problem. (spoiler: they all fail miserably.) Interestingly however, they all agree that feed-forward neural networks don't exhibit C., but integrated information theory would claim that this changes as soon as backward connections are added. Literally a single backward connection would give it minimal C.
Post #9: discuss in agonizing detail why this latest claim, that adding backward connections changes everything, is not plausible. I think this will include
all the other problems with digital consciousness, but they all relate to the binding problem so there is no sharp distinction. Ask the readers to justify why of course more complex digital systems do still solve the Binding Problem
Post #10: go through all possible answers to this question, conclude that they all don't work. finally actually call the premise into question
( now the hide and seek is over and at last the competing hypothesis must be discussed )Post #11: Discuss how a non-algorithmic theory would work -- with C. being about the electromagnetic field -- why we can't rule it out and what its advantages are. Also discuss the U-shaped information curve
Post #12: Go through conscious vs. unconscious processing and how this checks out with AI research
Post #13: Explain how topological segmentation can solve the Binding Problem
This is not complete but planning further won't be useful since I'm only at post #6.