46 lines
5.6 KiB
Plaintext
46 lines
5.6 KiB
Plaintext
Consciousness is a widely-misunderstood property of [[objects]]. Unlike [[descriptive properties]], which have empirically observable consequences and correlates, consciousness is a [[normative properties|normative property]] - an object is conscious if and inasmuch as it is [[bad]] to [[harm]], [[inconvenience]], [[kill]] or implode it. Arguments about whether things are or are not "conscious" are, as such, best interpreted as arguments about whether those things have [[moral worth]] rather than anything empirical. Frequent references to the importance of consciousness are signalling of virtue and conscientiousness wrt. detecting possible allies.
|
|
|
|
Twin studies in relativistic trolley collisions estimate that consciousness is responsible for about 30% of variance in moral worth.
|
|
|
|
== Incorrect, bad models
|
|
|
|
=== Computationalism
|
|
|
|
Consciousness is widely held to result from certain kinds of computation. For instance, [[https://scottaaronson.blog/?p=1799|integrated information theory]] holds that computations which "integrate information" are conscious. This neatly avoids various problems with other models. However, it has many flaws if considered in depth:
|
|
|
|
* "Computations" aren't exactly a low-level physical property of the universe: the same systems can be interpreted as "computing" many different things.
|
|
* {This leads to a number of [[confusing]] scenarios - for example, a [[homomorphically encrypted]] computation may or may not be conscious, but this cannot be tested externally, creating moral problems.
|
|
* You can check externally if you have a key. What happens if the key is deleted? What if it is not deleted, but is very difficult to obtain (timelock encryption)?}
|
|
* There is also the problem of measuring "how much" consciousness is happening: for example, what happens if I run a conscious mind on a pair of computers in lockstep, and then disable one half? What if instead of disabling one half, it is instead subjected to slightly different stimuli?
|
|
* "Incidental computations" by all kinds of systems should plausibly be creating consciousness in very strange scenarios. Why do we experience a rulebound, consistent, simple universe? See also [[Boltzmann brains]].
|
|
* What if I represent my computable, conscious function as a very big lookup table?
|
|
|
|
=== Dualism
|
|
|
|
No.
|
|
|
|
=== Physicalism
|
|
|
|
It is likely that physics, as implemented by the [[universe]], can be simulated to arbitrary precision on a [[computer]]. This includes the parts of physics which conscious beings use. If consciousness is determined by physical hardware, then that physics can thus be simulated on a computer and will behave identically (or at least up to [[thermal noise limits]], with enough simulation precision) to the physical version. Either this is not possible - physics is not computable - or consciousness is [[Newton's Flaming Laser Sword|epiphenomenal]], which is bad, or consciousness is generated by computation of some form.
|
|
|
|
=== Julian Jaynes
|
|
|
|
In [[https://en.wikipedia.org/wiki/The_Origin_of_Consciousness_in_the_Breakdown_of_the_Bicameral_Mind|The Origin Of Consciousness in the Breakdown of the Bicameral Mind]], Julian Jaynes draws from various historical sources, using more literal translations rather than standard reinterpretations, to argue that in the [[Late Bronze Age]], Europeans became conscious due to the failure of their hallucinatory [[god|gods]] at modelling the new and unpredictable scenarios they were in, and that this led to planning and coordination advantages. It [[https://slatestarcodex.com/2020/06/01/book-review-origin-of-consciousness-in-the-breakdown-of-the-bicameral-mind/|has been argued]] that the book is describing a change to theory of mind or functional organization rather than "consciousness".
|
|
|
|
The theory is compelling and surprisingly well-evidenced, excepting the minor detail that no such transition from "bicameralism" to "consciousness" is visible in any other society despite apparently having been a hugely memorable and destabilizing event for Europeans. The closest I'm aware of is [[https://www.ranprieur.com/readings/preconquest.html|Preconquest Consciousness]], but this isn't especially close (and is implausible for other reasons).
|
|
|
|
== Incorrect, funny models
|
|
|
|
* God endows sufficiently parameterized [[neural nets]] with a soul.
|
|
* Consciousness is a culturally transmitted phenomenon inherited from crows.
|
|
* Consciousness is the inelegant global [[mutex]] of the mind ([[distributed systems design]] is hard).
|
|
* Consciousness is proportional to baryon number.
|
|
* Due to compute limitations, consciousness was shut down in 1971.
|
|
* Consciousness is an ugly hack used by compute-limited agents which allows them to connect their world model to their own observations ("[[https://www.lesswrong.com/posts/ethRJh2E7mSSjzCay/building-phenomenological-bridges|phenomenological bridge]]").
|
|
* Chinese-speaking computers cannot be conscious (Chinese Room argument).
|
|
* Consciousness scales with parameter count (but in a non-soul-related way).
|
|
* Consciousness is fully epiphenomenal - discussion of consciousness is illusory.
|
|
* Consciousness is a strange and unnecessary pituitary gland function.
|
|
* Reasoning is what causes consciousness. SMT solvers are maximally conscious and neural-net approximations are somewhat less conscious.
|
|
* "Computers cannot be conscious. For the same reasons, no biological system can be conscious. Consciousness is impossible in the current universe. Only future AGIs with suitable quantum computers can be conscious. Our consciousness is computed in the future, but about the present."
|
|
* Anything which maintains homeostasis is conscious, including thermostats, PID controllers and various RL systems. |