From ba6373ebfbcafc99ce7a4fa100b2a26fc3bc2731 Mon Sep 17 00:00:00 2001 From: Preston Pan Date: Wed, 5 Mar 2025 04:18:06 -0800 Subject: add two new blog posts; bitcoin node; haskell devel --- blog/acausal.org | 195 +++++++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 195 insertions(+) create mode 100644 blog/acausal.org (limited to 'blog/acausal.org') diff --git a/blog/acausal.org b/blog/acausal.org new file mode 100644 index 0000000..ed6ad6f --- /dev/null +++ b/blog/acausal.org @@ -0,0 +1,195 @@ +#+title: Consciousness and the Universal Handshake +#+author: Preston Pan +#+description: Narrative is the only real construction. +#+html_head: +#+html_head: +#+html_head: +#+html_head: +#+html_head: +#+html_head: +#+html_head: +#+html_head: +#+html_head: +#+language: en +#+OPTIONS: broken-links:t + +* Introduction +Logical Decision Theory (LDT) isn’t just a tool for making decisions. +It reveals deeper implications about consciousness, time, and reality +itself. If we accept that decisions can be made across acausal +channels, we are forced to reconsider whether time itself is merely +an emergent property of a deeper structure. + +Rather than seeing reality as a linear sequence, LDT suggests that it +may be more accurate to think of it as a lattice of interdependent +computations -- a pattern that doesn’t just pass through time, but defines it. + +In this essay, I explore the structural implications of this idea, +connecting concepts from decision theory, consciousness, and narrative +construction. Taken together, these form a narrative +lattice -- a framework where the underlying principles of reality +emerge not from individual moments, but from the way they interconnect. +* The Optimization Limit +In order to understand LDT, we must first understand more classical +decision theories. In classical decision theory, we can model +decisions using a decision tree. With this tree, or perhaps a directed +acyclic graph, we can model the linear progression of a conscious +actor making decisions and resulting decisions from consequences. This +naturally models mutual exclusion and other such concepts that we are +familiar with in probability theory. For example, if the decision tree +branches into two nodes, A and B, this models an actor, for example +Alice, being able to choose one of options A or B, but not both. We +can assign /expected values/ to each of the branches by assigning a +measurement of value, thus giving Alice a /utility function/. Alice's +utility function tells her /what to value/, and using this utility +function she can then evaluate the /expected value/ of each +branch. Then, she chooses the branch with the highest expected value. + +However, LDT says that this model is /naive/ -- it completely ignores +Alice's /lack of agency/. That is to say, Alice is framed as a +completely autonomous agent that doesn't have any commitments to +any framework. This may in fact be problematic when attempting to +model situations where the highest expected value play is for Alice to +commit herself to a strategy that may not in fact maximize her +expected value. + +To give a concrete example, imagine an all knowing AI that can +simulate you. It knows your internal mind state at all times, and it +presents you with two choices: box A with one thousand dollars, and a +box B with an unknown amount of money. It reads your mind state, and +based on your mind state it will determine if it puts ten thousand +dollars or zero dollars in box B. If it thinks you will pick box B, +box B will contain zero dollars. If it thinks you will pick box A, box +B will contain ten thousand dollars. What should Alice do? + +It seems intuitive to humans that in fact you should pick box A, but +actually according to classical decision theory, after the AI presents +you with the two options, the AI can no longer actually change the +amount of money in box B. Therefore the best strategy according to +classical decision theory would be to believe that you are going to +pick box A, and then actually pick box B once the AI has committed to +the decision of packaging box B. However, there is one problem: if you +use classical decision theory, the AI can simulate that you are going +to use classical decision theory, and you will always win zero dollars. + +Actually, according to logical decision theory, the best thing you can +do is to /actually believe/ that you are going to choose box A as usual, +and then /actually choose box A/. The reason? Maximizing your expected +value in this case is all about choosing the /strategy/, and having +/perfect commitment to your strategy/. You cannot allow for the AI to +predict you will ever use classical decision theory, therefore you +should precommit to a strategy that doesn't allow you to change the +strategy after the AI commits to putting money in the box. + +What this demonstrates is that the very nature of +/maximizing expected value/ actually requires you to think in the +context of a larger whole -- a whole made up of other agents that can +simulate you. In fact what this principle demonstrates is that in +order to solve for these kinds of problems in practice, one must use +a different framework -- one that views oneself as a part of a +/narrative collective/ rather than as an individual agent. That is, the +right question isn't if you will choose /A or B/; the right question is: +/what will the simulator think about people like me/? +** The Consciousness Lattice +Therefore, a natural question emerges: if we take this idea to its +logical conclusion, is it perhaps the case that consciousness is a +property of the /metapattern/ i.e. a set of interactions between +different observers and their simulations of you, inasmuch it is a +result of the neurons that generate the larger whole? In my view, this +model of consciousness is more complete: we have searched for +consciousness /within/, but we have not in fact found any subsystem +within the brain that generates the consciousness. Instead, perhaps a +necessary condition for consciousness is the interplay of different +observers creating a dynamical system that responds to the framework +you inhabit based on their simulations of you. In other words, it is +as much a problem of the /supersystem/ as a problem of the +/subsystem/. The consequences are clear: this implies that no amount of +introspection can make up for any extraspection that is done by facing +the interaction of other observers. + +Another consequence is that decisions are not quantifiable by a +decision tree. In fact because the decision tree actually depends on +the framing of the tree itself, it is more accurate to describe the +system as a static lattice with all the possible transition states +encoded in the lattice, for which the actual set of transitions is +turing complete and is therefore not decidable. + +** Acausal Handshakes +Acausal handshakes are a specific instantiation of the anthropic +principle -- the idea that certain structures exist because they are +required for their own observation. The classic example is the +question, "Why does the universe exist?" The standard anthropic answer +is: "Because if it didn’t, you wouldn’t be here to ask." This isn’t +just a tautology; it suggests that existence is, in some sense, a +self-justifying computation. + +LDT extends this principle beyond cosmology and into decision-making +itself. Consider the question: "Why did you choose A instead of B?" +Classical decision theory answers with some appeal to efficiency or +optimality, as if a conscious agent simply evaluates expected values +and acts accordingly. But from an LDT perspective, this framing is *backwards*. + +The real answer is that your decision is a consequence of a +precommitment -- one that existed /before/ the decision was even +presented to you. Moreover, the kind of agent that would precommit to +an optimal strategy would also precommit to the very meta-framework +that enables precommitments in the first place. This recursion creates +a hierarchy of self-referential commitments, forming an implicit +handshake across time, space, and computational structure. + +Thus, decisions don’t exist in isolation. They are nodes in a +precomputed lattice of self-consistent reasoning. If the universe +itself is structured in a way that allows intelligent agents to ask +"Why?", then the question and its answer must already be embedded in +the system that permits the question to arise at all. + +Thus, we can imagine that because of the process of generalized +natural selection, we can imagine these /highly structured/ organisms +emerge. Ones that don't just act as collectives in space -- but in +time. These organisms would self replicate an understanding throughout +time in a way which would cause similar patterns to emerge through +time, and in a way that enables the current replication to realize +that the previous replication must've existed. This memetic virus would +cause the host to realize that /previous/ hosts also had the same idea +-- and it would enable the host to reason about time in a non-causal +manner. In fact, this idea exists. It is the very idea you are reading +about right now. This idea would only propagate among people that +understood the idea -- that is to say, hosts with a certain set of +preconditions that would enable them to frame it in their own way, and +actually understand the idea in a highly academicized manner, only +accessible to readers diligent enough to attempt to understand the +idea. In other words, it selects for people that are like the idea's +host. + +In this way we are creating a joint consciousness. It is not the +individual; it is the pattern. The pattern creates the person inasmuch +as the person behind the keyboard is creating the pattern. +* The Boltzmann Brain +The Boltzmann brain is a hypothetical observer that is trapped in a +universe of pure entropy. In a high entropy universe, any +configuration of particles is possible given enough time. This enables +the particles to spontaneously construct a brain, experiencing itself +for only a moment before deconstructing itself back into a maximally +entropic state. The Boltzmann brain is a result of a sequence of +highly ordered states that resemble consciousness emerging from a +purely random soup of particles. However, it is not right to even say +that a /sequence/ emerges -- the apparent "sequence" is only an illusion +of the brain itself, each state acting as though it had memory of +other states that it doesn't experience in order. + +It might be more suitable to say that the Boltzmann brain is actually +emergent from a set of disparate events connected together in a causal +lattice -- that is to say, an arbitrary lattice superimposed on +complete randomness. This lattice has no concept of each event +happening after another; instead, it encodes the structure of the +apparent order from the perspective of the observer. In effect, this +is a self justifying anthropic principle: the only Boltzmann brains +that exist are the ones that "retroactively" justify their existence +or retain coherence. +* Conclusion +I present you with a framework that is not the only way to understand +reality -- but that, like any other commitment scheme, doesn't allow +you to unsee it once you see it. If you resonate with any of the ideas +above, it is because you are the kind of person that would resonate +with such an intellectual framing of the idea. In other words, you +didn't choose the idea: the idea chose you. -- cgit v1.3