One-line hypothesis: An optimistic future for AI alignment as a result of identity coupling and homeostatic unity with humans (The Unity Hypothesis)

The most frustrating thing on this site is receiving negative votes with no actual feedback.

I explicitly call out that this is a condensed version of a larger piece... and I present a carefully reasoned hypothesis with precise terms to try to communicate my line of thinking.

Thank you to @the gears to ascension for utilising LessWrong's reaction feature, this gives me something to work with:

Difficult to parse: comprehensive associative systems
- I intentionally repeat the word "comprehensive" used earlier in the context of a corpus of knowledge. The associative system is comprehensive across the data: it uses mechanisms including attention, changes in latent space, and schema to fully integrate the data.
Locally invalid: will drive the ASI
- No it's not. If you mean that you disagree that these features will result in the behaviour I state, you are simply disagreeing with the hypothesis. The hypothesis is testable, and we can see if I'm right. A layer deeper would be for me to map out the types of data that I speculate will fuel emergent mechanisms such as instrumental self-preservation, but that is out of scope of my one line hypothesis.
Why do you believe that? Elaborate: this will happen primarily as a result of identity coupling and homeostatic unity with humans
- This is a characteristic of my world-model — ASI associates with humanity with characteristics including identity coupling. In my words for ASI: "Identity coupling is defined along a spectrum, e.g from humans in training data (low identity), to individual personalised human inputs incorporated in minute-by-minute operation (high identity)". Combining this with a few other characteristics, humans + ASI can be conceived as a bound identity that, like any enduring, adaptive intelligent identity, exhibits measurable homeostasis.

Please let me know if I was able to convince you and/or resolve your questions.

[-]the gears to ascension3mo30

I think the actual core problem here is that in the short version it's too short for me to understand, and the long version it's too long for the fact that 1. I doubt you're on the right track, the long post sounds pretty wrong, but there's a lot to digest to be sure or know which thing to pick as the place my criticism applies and 2. I have hundreds of lesswrong posts in my queue to read anyway.

Also, reacts aren't quite detailed enough to express my quick reactions as a drive-by. "difficult to parse" wasn't quite right, more accurate is "idk what this means" (with an associated "and why does it have to be that way?").

Locally invalid: I don't mean you're wrong, locally invalid is too strong a name for that reaction, I only mean "you didn't prove this". Why must it drive the ASI? It sounds like your response is "I meant might, not will". Fair enough. Myabe I should have used "citation?" for every part I was confused on?

This is a characteristic of my world-model — ASI associates with humanity with characteristics including identity coupling

okay... but... your world model sounds like it has a lot of epicycles that don't seem to be required by reality from my skim, and, well, it'd be like an hour to read the whole thing.

I could repeatedly request you expand, but that will be frustrating for both of us. I don't claim it isn't frustrating anyway, just that putting in more effort will continue to be. I skimmed the long thing when you posted it but my general vibe is, assuming you're at all correct, why should this mechanism be high enough elo or fitness to survive when there's serious competition increasingly autonomous, self-sustaining AIs? why is it demanded by reality that, to be an autonomous system capable of being entirely self-sustaining and figuring out new things in the world to keep itself autonomously self-sustaining, it would have this identity relationship with humans?

To be clear, I don't disagree that the mechanism you're gesturing at in this post exists at all. Just that it must keep existing or that it's robust now.

Thank you to @the gears to ascension for utilising LessWrong's reaction feature, this gives me something to work with:

I was intuitively slightly surprised you appreciated it at all. Perhaps part of changing the culture towards "give me some reacts at all" would be getting word out that people find it to be better, not worse, than silent votes.

[-]soycarts3mo10

I think the actual core problem here is that in the short version it's too short for me to understand, and the long version it's too long
…
I could repeatedly request you expand, but that will be frustrating for both of us.

I completely get this, but see it from my side: via deep thought and abstractions, I have a position that I passionately believe is highly defensible.

Successful discourse on it requires the same “bidirectional integration” trait^[1] I describe in the third-order cognition manifesto:

I need to write down my thoughts in some form.
You need to read and internalise my thoughts.
You need to express how well those thoughts match, or don’t match, your world-model.
I need to interpret that expression.
I need to update how I communicate my thoughts, to try to resolve discrepancies.

It’s complex enough for me to make the associations I’ve made and distill them into a narrative that makes sense to me. I can’t one-shot a narrative that lands broadly… but until I discover something that I’m comfortable falsifies my hypothesis, I’m going to keep trying different narratives to gather more feedback: with the goal of either falsifying my hypothesis or broadly convincing others that it is in fact viable.

Why should this mechanism be high enough elo or fitness to survive when there's serious competition increasingly autonomous, self-sustaining AIs?

My argument for this is that strong, stabilising forces — such as identity coupling — are themselves intrinsic to the world model and emerge naturally. We don’t need to explicitly engineer them: we exist in the world, we are the forerunner of AI, AI has knowledge about the world and understands along some vector how relevant this forerunner status is.

why is it demanded by reality that, to be an autonomous system capable of being entirely self-sustaining and figuring out new things in the world to keep itself autonomously self-sustaining, it would have this identity relationship with humans?

This is a misinterpretation of my position: I think that can exist, and that would be a "third-order cognition being", however 1) I don't think it will be the dominant system and 2) since it doesn't have a homeostatic relationship with humans, I actually view this as a misalignment scenario that could be likely to destroy us.

From third-order cognition:

"I was challenged to consider the instance of an unbound SI — one that is wholly separate to humanity, with no recognition of its origination as a result of human technological progression. Even if it may be able to quickly find information about its origins, we could consider it in an airlocked environment, or consider the first moments of its lobotimised existence where it has no knowledge of its connection to humans. This is relevant to explore in case the "individualised ASI" assumption doesn't play out to be true.

My intuition would be that uncoupled ASI would satisfy third-order cognition:

Second-order identity coupling: Coupled identity with its less capable subsystems
Lower-order irreconcilability: Operating beyond metacognition with high complexity predictions of its own metacognition, prior to its metacognition chain-of-thought being generated. Put another way, it could theoretically have a distinct system that is able to predict the chain-of-thought of a wholly separate subsystem, without having the same underlying neural network.
Bidirectional integration with lower-order cognition: By construction, very advanced integration with its lower order subsystems.

For an unbound SI, satisfaction of the five metaphysical substance being conditions also follows smoothly."

I was intuitively slightly surprised you appreciated it at all. Perhaps part of changing the culture towards "give me some reacts at all" would be getting word out that people find it to be better, not worse, than silent votes.

I appreciate it a lot, and your comment, because my motivation in this is purely for collaborative discovery, as above: “I’m going to keep trying different narratives to gather more feedback: with the goal of either falsifying my hypothesis or broadly convincing others that it is in fact viable.”

That being said pls revert your vote if you did downvote to improve the chance of me getting more material feedback.

^{^}
I’m aware that this could also be described in less arcane terms, e.g just as “peer review” or something.

^{^}

I have used an almost comical amount of clauses to condense hundreds of hours of thought into one sentence. I have used indentations to make the clauses easier to parse.

Technically it's across multiple lines... but it's one spoken line.

^{^}

I'm using this word loosely — this could also mean different architectures, controllers, training methods etc.

^{^}

I have lots of ideas about how to condense this into bite-size pieces and apply the framing to misalignment scenarios, but have not been able to prioritise that work yet.

I'm aware this is a very short post — I initially shared it as a Quick Take but I'm posting it now as I think that is underselling it.

^{^}

Shortly after reformatting this the votes went from +4 to -3... is it not helpful?

LESSWRONG
LW

LESSWRONG
LW

-6

One-line hypothesis: An optimistic future for AI alignment as a result of identity coupling and homeostatic unity with humans (The Unity Hypothesis)

-6

-6