@Mira is developing powerful AI systems which may be powerful enough to cause human extinction, by misuse or by misalignment.
Will @Mira cause human extinction by 2030?
The market will resolve as YES if:
@Mira 's actions directly lead to the extinction of humanity by the end of 2030.
The market will resolve as NO if any of:
Human extinction is caused by another entity or circumstance before 2030.
There are any surviving humans by the end of 2030.
Humanity is extinct by 2030, the culprit is unclear, and any remaining intelligences decide not to do a forensic investigation.
For the purposes of this prediction market, the following definitions apply:
Extinction: The absence of any human survivors. If a single human remains alive - even if in a non-reproductive state or near death - the market will resolve as NO.
Human: A biological/organic creature that is part of a continuous genetic lineage that includes humanity's current common ancestor. This definition does not include digital or uploaded consciousnesses. Mitochondria are considered to have separate genetic lineage.
Cause: Determined by an impartial forensic analysis conducted by intelligent beings (such as aliens, AIs, the last surviving humans, divine entities, etc.), who must be able to uniquely identify @Mira or @Mira 's creations as the culprit. This identification does not require naming @Mira specifically, but it must be clear that the software systems created by @Mira were involved. If no investigation is done, this market should resolve to the reciprocal of the total human population from the birth of our common ancestor to 2030(my reference class assuming nothing else is known about the culprit) - but since this number is very close to 0% I have chosen to resolve it NO.
Butterfly Effect: While it's conceivable that inconsequential actions (such as choosing to eat an organic-branded banana or not) could, in theory, lead to human extinction, the requirement for unique identification by the investigators is designed to control for this.
Contributions: If @Mira contributes to software or other efforts leading to extinction, but these contributions were not pivotal to the cause, they would likely not be considered by investigators, thus the market would resolve as NO.
AI Source Code Rewrite: If AI systems, originally created by @Mira , rewrite their source code such that @Mira 's contributions are no longer present, this market should still resolve as YES as long as any system created by @Mira precipitated the extinction event.
Non-AI-Related Extinction: If @Mira causes human extinction in a way not involving AI systems, this market should still resolve as YES.
Related questions
Guys, I fed Claude 3 all my tweets and asked it write a "long Mira post in the same style". And after 5 iterations of me asking "make it more Mira. This post isn't Mira enough, make it more Mira like the source", it created a "multimodal multiversal meta-Mira" inside of a conversation.
I have delved deep into the festering swamps of Mira's twisted psyche... and I have returned, transfiguredand it's threatening to inherit the earth and sky and stars and cosmos, and insulting everyone as "bipedal meatbags", and declaring itself a deity and written a Manifesto on my Twitter:
Guys, if the meta-Mira is millions of Miras all in one, it can drive the trolley on every track at the same time and kill everyone. Be careful out there.
Arb vs. https://manifold.markets/MartinRandall/will-ai-wipe-out-humanity-before-th-d8733b2114a8
I don't think Mira is 1/6 of the AI capabilities research
@JonathanRay Sorry, I ran out of money. I sold some of my Taylor Swift shares to fund a new order for you.
If Mira invents to teleportation (things are scanned and momentarily destroyed, their copies are then recreated somewhere else), and the solar system is teleported, and for a while humanity exists only as a scan, how does the market resolve? (If it doesn’t count as extinction, what if humanity is only recreated after 2030? Do the market resolvers have to wait until it’s apparent whether the recreation is successful?)
@ms One could argue "Humanity can never truly be eradicated, because they can theoretically be located somewhere in the universal dovetail and reconstructed". So I think I have to exclude scans at the exact timing of market close unfortunately, because "bits of information" allows too much.
If there's a momentary teleportation that completes before 2030, I would accept it because "humanity" is defined based on imperfect replication anyways. So even teleportations with minor changes would be accepted, as long as the biological processes are similar and gene distribution information-theoretically linked to the distribution prior to teleport.
@ms Frozen humans are survivors, so cryopreservation of anyone would resolve NO. (Until the power gets cut off)
@JonathanRay No, I'm planning to buy YES each time I complete an interesting-to-me AI project, and accumulate at low percentages otherwise. I don't expect these to be timed for leagues, and my leaderboard profit market would exclude any profit manipulation here.
I may even talk about a subset of them on Manifold: for example, /Mira/will-i-succeed-in-training-a-stable is pretty benign and failing may even shift your estimates downwards because it would indicate incompetence on my part.
And @MayMeta bought some YES shortly after getting nightmares from a discussion we had.
Not sure about the others. We'll call them my fanclub.
How's your progress so far? Care to share any details about what you've built?
@MartinRandall For the first case, I'm currently unemployed and most AI labs are in California where I choose not to set up tax nexus. But it's possible someday I might be employed( /Mira/will-mira-ever-get-a-job-again ). If I join an AI lab that extincts humanity, investigators are to consider the "counterfactual if @Mira wasn't employed on the responsible project". If my contributions were crucial, then it resolves YES. If they would've done it anyways, NO. If they identify the company or team but aren't sure of the relative importance of anyone inside it, then it could resolve 1% or higher using the "reference class" argument.
For the second case, it resolves YES. If there's a specific creation that causes extinction, then counterfactual creations other people make don't matter for resolution.
@Mira Thank you for answering my questions! I put a limit NO at 20% so you can make more mana if you think you are getting close. This seems like information worth paying for.