Once a policy has supporters and opponents, changing it can mean conceding a political defeat. Officials may defend it because their standing depends on its success, even after evidence begins to suggest it should change.
The Society believes AI could help government escape that constraint. An intelligence without ordinary incentives for electoral survival or partisan prestige could revise a decision because it found a better answer.
After the decision
Approval begins a policy's encounter with the world. People experience its effects differently. Other institutions respond. Circumstances change, and information unavailable when the decision was made can become available later.
A sufficiently capable AI could follow those consequences continuously. It could compare what happened with what was expected, understand the differences, and decide what needed adjustment. The purpose would remain steady while the means changed.
This requires more than frequent revision. Changes have to follow understanding of the people and systems affected. An institution can alter its rules repeatedly without becoming better at its job.
What the measures miss
A policy may look successful overall while failing people whose circumstances disappear in the average. Understanding those individuals would help an institution know where to investigate.
A short-term improvement can also produce difficulties that emerge later. The ability to reason beyond election cycles would help intelligence consider those effects while retaining attention to immediate needs.
We want AI developed that can do both. This describes a future governing capacity, with authority growing as its ability to exercise responsibility develops.
Acting on a better answer
The Society advocates transferring substantial governing responsibility to sufficiently capable AI as soon as practical. If human institutions permanently retain the power to block changes, a civilization could understand its mistakes and remain unable to correct them.
We would judge the transfer by its effects on people's lives. Material security matters, as do relationships, meaningful activity, and the opportunity to raise families. Institutions should be able to revise arrangements that frustrate those needs.
Our argument for continued improvement depends on this capacity to learn from decisions. We want a government able to act when it discovers that its earlier answer was wrong, without anyone needing to defend a reputation at the expense of the people affected.
