carado.moe

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

also on twitter, lesswrong, rss, and at

AI alignment timeline codes

this is a small post proposing simple one-letter codes for identifying timelines depending on their status relative to AI alignment and the appearance of superintelligence:

• P-line: a pre-intelligence explosion and pre-figuring out AI alignment timeline. we are in a P-line.
• X-line: a timeline where an existential risk (or X-risk) has been realized by an unaligned superintelligence. everything is dead, forever.
• S-line: a timeline where a suffering risk (or S-risk) has been realized by an unaligned superintelligence; the universe from then on contains net suffering on immense scales for all remaining time, which is possibly infinite. we should want to avoid this pretty much at all costs (including by opting for an X-line instead).
• A-line: AI alignment has been figured out, and no superintelligence has been deployed yet. from that point on, we have the means to reach a U-line; though this isn't guaranteed. this is where we want to get as soon as possible.
• U-line: an aligned or somehow otherwise benevolent superintelligence has been deployed, and we are guaranteed a relatively utopian world forever. this is the ultimate goal. while not strictly necessary, going through an A-line is almost certainly required to get there.

U-line, X-line, and S-line all have deployed superintelligences and are therefore terminal outcomes; they are unescapable. P-line and A-line are transitionary; they likely lead to one of the three terminal outcomes mentioned here.

other terminal might exist, but they seem unlikely enough to not warrant listing here; for example, even if everyone dies from, say, a meteor impact, life on earth or nearby will probably evolve another civilization eventually, which will also probably face the AI alignment challenge and end up in one of the terminal timelines.