Resolves N/A if there is no huge alignment effort, or if there is not an okay outcome from AI. Otherwise, resolves at the level at which AI progress stalled out.

An okay outcome:

An outcome is "okay" if it gets at least 20% of the maximum attainable cosmopolitan value that could've been attained by a positive Singularity (a la full Coherent Extrapolated Volition done correctly), and existing humans don't suffer death or any other awful fates.

A huge alignment effort is an unusual amount of effort being put specifically into alignment, as a separate project blocking capabilities, above business-as-usual, corresponding to a willingness to give up on two years of capabilities lead time in favour of alignment; "making alignment your top priority and working really hard to over-engineer your system for safety".

If MIRI's agent foundations had worked perfectly, that would count; if there was a global ASI Project that took as long as it needed, that would count; if Anthropic RLHF'd Opus to support CEV, that would not count (the alignment is institutionally an optional extra on a capabilities project, aiming for an acceptable threshold of safety, rather than overengineering it).

"Superintelligent" means that it is able to do things at least as well as humanity as a whole. "Superhuman" means that it works as a replacement for any cognitive human task. "Below- or par- human" means there are some cognitive tasks that it can't do that humans can do.

"Stall out" means a capabilities slowdown at that level of intelligence that is at least long enough that the resulting slowdown is critical to a okay future. If there is no such stalling, then this resolves at the highest level of capabilities that is actually achieved - e.g. if we go smoothly to an ASI god building dyson spheres, this resolves at "superintelligent".

Some scenarios:

• Deep learning hits a wall. There is another AI winter long enough for a new paradigm to be developed; this one is more alignable. Resolves at the wall that deep learning hit.

• A new, more alignable paradigm is developed, which stalls out at or slightly above human levels. We work together with these near-superhuman AI and have a significant hand in the successor. Resolves at the level the AI under the new paradigm stalled out.

• There is a global ban on ASI research. After decades, a new, more alignable paradigm is developed, which scales up to superintelligence. Resolves at "superintelligent" (capabilities did not stall out as a matter of computer science, humans decided to stop researching).