path: root/blog/acausal.org

#+title: Consciousness and the Universal Handshake
#+author: Preston Pan
#+date: [2024-01-01]
#+subtitle: By {{{author}}}, 2024
#+description: Narrative is the only real construction.
#+html_head: <link rel="stylesheet" type="text/css" href="../style.css" />
#+html_head: <link rel="apple-touch-icon" sizes="180x180" href="/apple-touch-icon.png">
#+html_head: <link rel="icon" type="image/png" sizes="32x32" href="/favicon-32x32.png">
#+html_head: <link rel="icon" type="image/png" sizes="16x16" href="/favicon-16x16.png">
#+html_head: <link rel="manifest" href="/site.webmanifest">
#+html_head: <link rel="mask-icon" href="/safari-pinned-tab.svg" color="#5bbad5">
#+html_head: <meta name="msapplication-TileColor" content="#da532c">
#+html_head: <meta name="theme-color" content="#ffffff">
#+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.