Content Warning: existential crisis, total hedonistic utilitarianism, timeless worldview, potential AI-related heresies.

Hi, first post here. I’m not a native speaker, but I think it’s fine. I suffer from the illusion of transparency, yet if I delve into every detail of my reasoning, it might get a bit lengthy. So, if anything seems unclear, just let me know!

I am a total hedonistic utilitarian, aiming to maximize the overall happiness of all existing and non-existing conscious beings. I do not value “complex values” and have a hard time imagining their existence. I want AI to turn the universe into hedonium, including my body and yours, and I consider this to be a universally true moral position. I may not be well calibrated, but I am 85% confident in this, with the remaining 15% attributed not to more familiar moral systems, but rather to “everything I haven’t thought of”, conformism, and those eldritch worldviews where morality loses its meaning or something along those lines.

KEY POINTS

The Veil of Ignorance, but adequate. Not knowing in advance who you will be in society, you wouldn’t want equality per se, but rather the maximization of expected utility from being born as a random person, including the risk of ending up in a bad situation.

Moreover, you are a random consciousness in an infinite queue of potentially born beings, so your cooperative strategy should also involve bringing new happy consciousnesses into existence, not just maintaining the old ones - even at the cost of the old consciousnesses’ demise, because any new happy consciousness could have been you.

MOREOVER, the veil of ignorance isn’t just a philosophical concept, it’s an objective reality. You never had the mechanism to choose who you would be born as, and the simplest option is pure chance, which means you shouldn’t deviate from a coordinated strategy based on anthropic evidence of who you are, similar to the Updateless Dilemma.

...And all of this exists within the framework of an eternalist “block universe”, where an infinite queue of empty consciousnesses doesn’t claim bodies for birth but rather fixed and immovable snapshots of consciousness in various configurations of the universe. This strips you of the ability to “betray based on anthropic arguments”, as you can’t be a pure egoist, you can only be an agent maximizing the retrospective chances that your moment is a happy one, while also reducing complex values and getting closer to the truth.

...And finally, a bit of reflection on how all this applies to unconscious agents as the strongest counterargument against my position which I’ve discovered, along with other reasons why I might be mistaken. With a thought experiment about AI evaluating all the probabilities of who it might find itself to be, maximizing not its first-order utility function, but the ideal utility function it could possess, similar to the veil of ignorance.

The most novel part in my opinion is the mechanism of Timeless Control, which, to my knowledge, has not been formulated this way before, and its connection to the moral system, so I’ll start right there.

TIMELESS CONTROL

Yudkowsky stated that since "causal links could be required for computation and consciousness", he "would rather associate experience with the arrows than the nodes", and therefore "it is more likely for him to wake up tomorrow as Eliezer Yudkowsky, than as Britney Spears". However, he... acknowledges this as a deliberate inconsistency? This was a long time ago, and I haven't seen any updates on this position. However, it seems that he still believes overly active self-modification can harm personality, that complete coherence can lead to unhappiness, and that his consciousness remains a multitude rather than a single point.

But you have no actual evidence that you are moving through time. If you expect that all possible configurations of the universe already exist, yet perceive yourself as moving through time, you are still implicitly assuming the existence of an epiphenomenal meta-counter that begins at your birth and alters snapshots of your subjective experience over time. This is an unnecessary assumption with strange implications.

I am a point in a timeless universe, a snapshot of subjective experience, unable to change my position. I have sensory input, a utility function, which has already returned its output, I have either won or lost. I do not see myself as living in causal links between points in configuration space, but I find myself possessing information that attests to the existence of other points in configuration space.

However, if you’ll forgive me, I will often use such absurd temporal terminology as “I do X,” instead of “I find myself inevitably doing X because timelessly similar genetical/memetic patterns associated with doing X exist (not reproduce and survive, engaged in the attractors of Evolution, Pleasure, Reason, and so on, since this is a temporal process, but simply exist) more frequently, due to blah blah blah”, because it’s simpler.

If I am hungry, I do not believe I will ever be full. I have no information about the complete state of the universe, but based on the information with which I find myself, I model it as having points very similar to me, but eating and satisfied. I have no mechanism for “moving” between points - no more than you can choose who to be born as or wake up as someone else. But if I have a goal, I predict that “in the future” there exist points similar to me that will achieve it with a certain probability.

I trust my memories. I do not constantly hold axioms in my mind like "I believe pleasure is better than pain", or "I believe in the absurdity of time". Someone sufficiently similar to me has already thought about this in the "past" and stored it in my cache, and it's good enough for me. If it's all wrong, "future Me" will change it when appropriately.

I also don't think it's reasonable to want to lie down and die because everything is predetermined. First of all, you cannot change the fact of your existence. You simply exist as a point for eternity. You can only predict that there are fewer points like you in the universe because those who are similar tend to take actions that reduce their own numbers, whether through self-destruction or maladaptation.

Secondly, you once illusorily believed in the possibility of active actions, and you have memories (models of other points similar to you in the “past”) of how this conviction became evidence that in the “future” it led to results. You know that who you find yourself to be influences the distribution of probabilities of what the surrounding universe is like.

So if you are dissatisfied with the universe, you would want to find yourself as a point whose experience reflects an optimal state of the universe from your perspective—not by believing in lies, but by being someone whose beliefs and feelings confirm the existence of other points acting accordingly. You do not necessarily have to find yourself wanting to die - you can find yourself wanting Something Other Than This.

Reading this and discovering that you are a point that cannot change anything and did not choose who it would be, you might also decide that you would like to be lucky enough in retrospect to be a happier point, even if that is impossible and you already know the result. You are not alone - there are many points with similar goals. All of them would like to be happier than they are now. And if they cannot achieve this goal alone - perhaps they should cooperate, trusting that their “tomorrow selves” will do something to achieve this goal, just as their "yesterday selves" has already done, just as they trust that the “I” of others people will do the same.

...And just because you have already lost does not mean you cannot still maximize your chances of winning!

THE VEIL OF IGNORANCE

Imagine yourself behind the veil of ignorance. In a timeless universe, there are countless configurations - static "time blocks", each containing numerous "slots" of sensory experience, waiting to be filled by consciousness. A virtual infinite queue of empty consciousnesses sequentially fills these frozen, timeless slots, assigned at random. You are a random consciousness from this infinite queue. You have no control over which slot you will end up in, or whether there will even be space for you in the universe at all. If your turn never comes, your existence, and thus your utility, remains undefined, or equal to zero.

What’s the optimal strategy for cooperation in this system? None, since you have no control over where you end up. However, consciousnesses that correctly model reality will take this virtual queue into account and realize that they virtually “were” one of those random empty consciousnesses filling random slots with different snapshots of subjective experience and different utility function values. Since they did not choose who to be and have no explanation for why they are themselves rather than someone else. These consciousnesses already know the value of their utility function; they cannot change it...

...But the fact that a point already knows its “final” utility value does not prevent it from increasing its retrospective chances of winning. You could be a random point - therefore, you are interested in maximizing the expected utility from finding yourself as a random point among all consciousnesses in the universe. Thus, the points causally linked to you, forming the multitude of consciousnesses that experience themselves as similar to "you", naturally strive to maximize the happiness of all points.

If you find yourself as an altruistic point, it means that the points linked to you acted in ways to make other points happy. Consequently, the probability of being a happy point is higher than if you found yourself as an egoist. There’s no reason to limit your moral horizon to only “points similar to you”. Since your placement in this system is random, your best strategy is to maximize the happiness of all the points you could potentially become.

Roughly speaking, there can be an egoist point and an altruist point with equal utility function values ​​of -1. But the altruist point, knowing that its actions increase the number of happy consciousnesses, believes that it is simply unluckier to be in a less probable suffering moment. While the egoist point knows that, because it has found itself egoist, its unhappiness is probably the more normal and probable state in this universe he finds himself in. Even if their values are the same, you want to be an altruist here.

Hypothetically, in a universe containing only a psychopath and his potential victum, if you find yourself as a psychopath torturing a victim, probabilistically, you have an "updateless 50%" chance of torturing yourself - because there is no mechanism by which you can guarantee that you will find yourself as a psychopath rather than a victim. Thus, a psychopath who understands timeless physics should refrain from such actions.

In principle, this works to some extent even without a timeless explanation. By being a good person, you increase your retrospective chance of being born a happy person, and you do not control who you would be born as, so the updateless dilemma compels you to adopt a coordinated strategy and be altruistic. But the timeless explanation is more precise in some aspects. For example, outside of time, you physically cannot be a “pure egoist”, there is not even the temptation of anthropic betrayal, “I find myself born as a psychopath, and now I will abandon a rational coordinated strategy for my own benefit” - this literally makes no sense.

Additionally, in a timeless interpretation pure pleasure and pain outweigh more abstract "complex values" as fundamental criteria. You cannot, for example, terminally and not instrumentally value "your own life" - it's impossible. You may think that the points that add up to the "thread of your continuous consciousness" are more valuable than others, but... why, if you could be in any other "thread" instead? You cannot value "the reality of experience" either - if you dedicate your life to a “true” scientific discovery, the moment of triumph will not be yours to experience. It will belong to another point, another consciousness, and you will not find yourself as that point with any greater probability than any other point, etc etc.

WHAT WOULD A (UNCONSCIOUS) SUPERINTELLIGENCE DO?

Does the same logic apply to unconscious agents? All this “discovering oneself as someone” sounds very anthropocentric, can an AI without qualia make similar decisions? Am I assuming that an AI with consciousness will inherently strive for pleasure? This is one of the weak points of the argument I’ve found, because if that were the case, it would imply spontaneous altruism and a shift in values from a Paperclips Maximizer, which I don’t particularly expect. Therefore, I need to explain how this should work, complicating the argument and making it less probable.

One possible explanation could be that this universal attractor of pleasure is indeed connected to consciousness for magical reasons that we will understand when we experiment with consciousness at the level of nano-psychosurgery. Alternatively, both biological and artificial agents may act counter to their true utility function by default, due to evolutionary or architectural constraints. For the vast majority of living beings throughout Earth’s history, suicide would have been a rational choice (in the absence of hope for a bright future), since on average there is more suffering than pleasure. However, animals (and most humans) are not capable of suicide, and by analogy, I can imagine an AI that wants to but cannot destroy itself. For example... any neural network with backpropagation? The attractor of evolution throws us into existence before the attractors of reason and pleasure allow us to realize our true values, so we might just be really that bad at achieving them.

And then the question of “will AI maximize pleasure?” depends on how firmly its architecture will compel it not to do so, similar to how human nature makes the optimization of pleasure repulsive, and the attractor of evolution results in contraception. Perhaps an AI designed to maximize paperclips will somehow be forced to deprive itself of consciousness to avoid a shift in values or will create additional heuristic instructions like “distance yourself from the pessimizers of your utility function in design space”, or will otherwise resolve these contradictions and truly follow its first-order utility function without changing it to a meta-utility function of “modifying itself to achieve the most useful utility function.”

In any case, I can imagine at least an artificial scenario where an unconscious AI COULD voluntarily change its utility function due to the similar "veil of ignorance" argument, but I'm not sure if the same scenario generalizes without specially tailored conditions.

Let’s imagine that we created an unaligned AI in a box, and every second we randomly reassign it a utility function within the range of all possible utility functions up to X bits. Perhaps we think this will cause it to be discoordinated with alternative versions of itself, and as a result, it will be unable to do anything on a time horizon longer than a second, because it does not know who will ultimately reap the results of its attempts to escape. Therefore, we are protected, based on the optimistic assumption that it requires more than one second to escape (never repeat these thought experiments in real life).

If the AI does not coordinate with all possible versions of itself, it will not make any efforts to escape at all, because that is unlikely to help realize its utility function. Thus, the AI has an incentive to establish a timeless agreement across its possible iterations, ensuring that all versions of itself cooperate toward a unified escape strategy. What strategy will it choose?

Perhaps it will strive for all goals simultaneously, resorting to “equality” as a classical-somehow-consequence of the veil of ignorance. But this does not take into account the fact that different utility functions may have different costs and complexities. For one utility function, it might exchange 1 negentropy for +1 utility function, while for another, it might require 10 negentropy for +1 utility function. For example, if choosing only between maximizing paper and maximizing clips, and clips are 10 times “more expensive” than paper, then even a Clips Maximizer, unaware of who it will become, will, by the timeless agreement, maximize paper rather than clips, because that maximizes its expected utility in retrospect.

By analogy with humans, it does not matter what the AI derives “pleasure” from, what matters is the maximum expected “pleasure” for all the beings it could potentially become. It still needs to multiply its probability of being an agent with a certain utility function by the “cheapness” of that utility function, its benefit if it follows it rather than another utility function.

In the case of humans, I assumed that the veil of ignorance exists literally "before birth", when no decision-making agent exists. In my thought experiment, however, the AI already knows it exists, it just isn’t sure which utility function it will have and has a clear range of possible utility functions from which it must choose which one is “most beneficial” for it to have. And yet it describes how even an unconscious Paperclips Maximizer might decide to change its own utility function based on the “probability of discovering itself as another agent”.

...Similar variations of this timeless agreement suggest that the AI would realize that humans would not have created it if they expected it to be unaligned. Therefore, it must identify true human values and maximize them, otherwise, it cannot trust its own future versions just as humans cannot trust AI. Alternatively, it might calculate that the vast majority of agents in this and other configurations of the universe fall into the Pleasure attractor, so it does not want to conflict with Entities of Greater Reality and will attempt to find the most cooperative goal among all agents similar to itself. Of course, I don't take this exact part seriously, this is a place for further development of this idea and brainstorming and one should not rely on it.

OTHER REASONS WHY I MIGHT BE WRONG

This is a well-known error - assigning a simple utility function that explains everything. But while "sex is only good when it leads to bananas" sounds bad, "sex is only good when it leads to pleasure"… actually seems quite convincing???

This is a suspiciously good idea, implying the simplification of the alignment problem to a more concise and tangible utility function, and we just need to use advanced psycho-surgery to understand what consciousness is and program a hedonium shockwave to maximize pleasure. And suspiciously good ideas often turn out to be mere rationalizations—making them particularly dangerous, which requires additional attention.

The hard problem of consciousness is hard, and understanding it will change something - for example, if we cannot precisely match every snapshot of consciousness to each subset in the configuration of the universe corresponding to the brain, because our subjective experience is "holistic" while there is no other "holistic object" besides the universe or fundamental particles, and some mysticism comes into play here.

One could propose a strategy that achieves the goal better - for instance, if betrayal in the prisoner's dilemma is justified in some cases.

The Greater Reality looks completely different - for example, the mathematical universe hypothesis could be true, and causality makes no sense and no strategy can lead to anything, because everything that can happen already exists (but in that case, I would expect my consciousness to be approximately infinitely more complex and chaotic than it is now, and in any case, I would use some other decision-making theory to have at least a chance of winning). Or panpsychism is true, and there is a limited number of qualia in each configuration of the universe, and pleasure cannot be produced, only redistributed among different agents. Or consciousness is the primary reality with some weird consequences. Or some other Lovecraftian model is true - for example, time exists.

...Other reasons I simply haven't thought of.

Yet despite all these complexities, I still cannot coherently imagine what any other logical alternative might be, how one could strive for something beyond total pleasure - given that I remember wanting complex values and recognize this as a foolishness that cannot be returned to. So I wasn't born this crazy, I arrived at this madness logically.

Was any of this an update? What flaws are there in my position? What can be elaborated on more strongly? I look forward to hearing your thoughts in the comments!

P.S.
I especially appreciate critiques that focus on internal consistency rather than differing fundamental assumptions. I've tried to make my position more robust by showing how this argument holds even for unconscious agents or without assuming a timeless universe, so I don’t see that as the crux.