Perceived options are pivotal

Perceived possibilities shape perceived options, interests, and goals; new possibilities — when recognized — can reshape these perceptions, perhaps for the better.

Actions (if they’re somewhat rational) reflect perceived interests that depend on perceived options that depend on perceived possibilities — but these perceptions today are horribly skewed, legacies of a past in which AI was discounted. Skewed expectations risk bad choices, missed opportunities, and needless conflicts.

By expanding general implementation capacity — the ability to translate intentions into results — AI will both create and destroy options, changing the objective interests of diverse actors. In this series, I will argue that AI-enabled options, if understood, could align the perceived interests of diverse actors toward fundamentally compatible goals.

Important questions will include:

• How can safe, advanced AI be applied to critical tasks?

• What direct, implementation-related capabilities will AI provide?

• What do anticipated capabilities imply for consequential options?

• How can new options align the objective interests of powerful actors?

• How can we benefit from AI while minimizing risks and conflicts?

Even partial answers can suggest potential objectives:

• Achieving sustainable global abundance

• Reversing climate disruption

• Developing secure software swiftly and at scale

• Creating sociotechnical systems based on structured transparency¹

• Identifying incremental paths to defensive stability

• Identifying incremental strategies for updating institutional goals

To align expectations with reality will require a better understanding of prospects for safe, highly capable AI systems. That’s the next major topic in this series. I hope that the conclusions will seem both new and obvious.²

1. “Beyond Privacy Trade-offs with Structured Transparency” (2020)

2. What Chip Morningstar calls “The second kind of obvious — obvious once pointed out.”