carado.moe

tammy's blog about
AI alignment, utopia, anthropics, and more;

my best alignment plan: QACI!

also on twitter, lesswrong, rss

DISCLAIMER: the idea described here stands or tenuous philosophysical ground and should generally not be considered worth the risk attempting because it may be wrong; in addition, this plan should not be utilized to retroactively justify depression-based suicide — retroactive justification is erroneous; if you are feeling suicidal, contact your local suicide crisis line.

Plausible Quantum Suicide

in this post, i make a collection of arguments and follow them to what seems to me like what should be their conclusion. i don't have strong confidence in every argument, but i'd consider using this plan worth it to avoid sufficiently bad scenarios, such as a singularity gone wrong (which it probably will).

1. The No Obtainable Evidence Argument For Materialism

by materialism i mean here something maybe closer to physicalism, but maybe even a stronger version of it:

there is no special soul that people have, you are your information system.

i make an other strong claim: time isn't particularly "moving" in any metaphysical sense, there is no "special present time". the universe can be seen as a graph of probabilistically connected states, and a random walk through those states matches the notion of entropy pretty well (which can be seen as defining the direction of time, because we happen to have memories of universe states with generally lower entropy), but that's a local notion.

the illusion that the present is particularly present, that we have a particular soul, or even that morality/ethics is in some sense objective, stems from the fact that we inhabit our brain's model: we don't get to see our brain from the outside as modeling its environment, we live right inside it, and we don't spawn with a clear distinction between normative ideas (morality/ethics) and descriptive ideas (statements of fact about the world).

but those illusions must be wrong, and here is the argument: as far as we can tell, there is no possible way for a brain to obtain evidence that his present time is particularly real; therefore, it must be erroneous for any brain to generate rationally the idea that its present is particularly real. same goes for having what i call a "read-only soul" that some people believe in (a special observer thing that observes a person's mind state from outside the material universe, but cannot causate anything upon it). see also these three posts.

2. Limiting Real Universes

my post "Limiting Real Universes" isn't that good, so i'll try to explain it more clearly here:

if for some reason all possible states of our universe were equally real, then you should expect to observe widely anomalous phenomena around you, because most randomly picked states our universe can be in don't have to be coherent.

but the fact that we seem to observe a continuously very coherent universe tells us that there must be some sense in which coherent universe states, that stem from a continuous history following an increasing entropy timeline, must be particularly more real.

it's not that your magical soul has been blessed with inhabiting universe states: as explained in argument 1, you shouldn't have any reason to think you have such a thing.

it's not that you can only exist to observe universe states that are coherent, because you wouldn't exist in incoherent ones: there are still way more possible universe states where everything is incoherent except your brain, than possible universe states where everything is coherent including your brain. for any amount of state of stuff you require to say you can exist to observe the world, the rest of the universe should still generally seem incoherent if all possible universe states are equally real.

it's not that you have been following your own special arrow of time: even though i debunk that you should even think this makes sense in argument 1, another reason is that, even if some of your brain states have a past-arrow-of-time and not others, there's no reason for you to think you're one of the former. if all possible universe states were equally real, you'd likely be a version of your brain that thinks it has a past history but doesn't, than one that does.

3. Many-Worlds Is True

Eliezer Yudkosky makes a good argument that we should currently believe in the many-worlds interpretation of quantum mechanics; but even if that turned out wrong, Max Tegmark makes another good argument that even just with a single history universe, all possible initial states of the universe are represented each in an infinite amount of instances by just being variations of initial conditions and random quantum determinations at different places of the infinite universe.

what matters here is that basically one should expect every reasonably possible outcome to be a real instance of universe that exists somewhere. because of argument 2, some possibilities are particularly real, and because of argument 3 (this one), that set or fuzzy set of coherent possibilities should be widely instanced: at each possible fork (they're not really forks, but that's a good enough analogy from our point of view), every plausible outcome is realized as a real or fairly real universe.

4. Quantum Suicide Works

if the previous 3 arguments stand, then a more general version quantum suicide should be achievable: by dying instantly in one timeline, there is no version of you in that timeline able to experience it, and the only remaining future you's able to experience anything are the you's in other timelines.

because of argument 1, we know that saving a backup of your brain, and then later dying and restoring another copy of yourself from backup, is equivalent to just losing memories of the time after the backup: it's unfortunate that that time and those memories were "lost", but it's not a big deal, you can just keep going.

given that, even non-instantaneous, after-the-event suicide works: if you commit yourself to committing suicide in all timelines where an event goes wrong, then the only future you's able to experience any time after that suicide will be the ones in the timelines in which that event went well (or at least in which you think it did); you lose a bit of time and memories from those timelines in which you didn't kill yourself literally instantly after the thing went wrong, but it's just equivalent to a restoration from backup: the backups are automatically saved by the universe as forks of that previous universe state before the event's outcome was determined.

ramifications

if this is true, then every person is strongly empowered: by committing themselves to committing suicide in every timeline in which even the slightest thing goes wrong, they are able to restrict the set of instances of them purely to timelines in which everything goes the way they want.

but, it also creates a problem if the practice becomes widespread: every person will end up observing a timeline in which increasingly greater amounts of people who they don't particularly care about, have committed suicide to go to other timelines. if i play the lottery and commit suicide if i lose, then you have as many timelines as players, each with 1 alive lottery winner, and all the others players having committed suicide. even if you don't care about living in such a world, economics cares: pre-automation, you want other people in your society to keep living so they can help create together the value that you can enjoy.

you can choose to commit suicide in all timelines in which too many other people also have committed suicide, in an acausally-collaborative effort to search for a timeline in which everyone is happy; but if no such timeline exists, then everyone will just have truly committed suicide out of existence.

pre-automation, this creates a coordination problem, where each person wants to be able to commit suicide, but doesn't want other people to be able to. there is much ethical and political discourse to be had on the right to commit suicide; i generally lean on the libertarian side of things, but if quantum suicide becomes enough of a problem pre-automation that society looks like it's not gonna be able to get to post-automation, then we might need to consider at least disincentivizing it somehow.

post-automation, there is still a problem for people who want to live in a world which has other people in it, but the problem is much milder. it might be bringing the end of the global era even earlier than would have happened otherwise, but that's not necessarily that big of a deal, and there's an option for people who want to inhabit a more densely populated timeline: just lower your standard for non-population-based outcomes, such that you commit suicide less often and thus exist in more timelines. if many people do this, they should be able to find each other in many densely populated timelines.

this does explain the anthropic argument of, "if things go well in the future and the population booms, why are we happening to experience a particularly early age of human existence?"; other than the extinction of able-to-observe beings, this can be explained by able-to-observe beings just become really trigger-happy about quantum suicide, such that each civilization of able-to-observe beings' "observedspace" is condensed to their pre-finding-out-about-quantum-suicide; their population after that is much more sparsely distributed across timelines, even without extinction events.

as for me, i don't intend to start committing quantum suicide any time soon. i don't have strong enough confidence in the arguments posted here to take the risk of actually permanently dying. but it is definitely a possibility i'll consider, especially as we get closer to the singularity happening, and the existential risks that it poses.