This is a less formal overview of the key ideas in the working paper “Bracketing Cluelessness”. There is also a standalone linkpost for the paper, which I’ve split off from this post, since this one contains personal commentary (not necessarily endorsed by my co-authors).

Summary

We may be clueless about how our actions affect aggregate welfare: we may not be able to say any intervention is better, worse, or exactly as good as another, in terms of its expected total consequences. There is an intuition that the appropriate response is to base our decisions on relatively robust and predictable effects, and that this favors near-term interventions like global health and animal welfare. But there is not yet a principled basis for this move that also respects impartial, risk-neutral consequentialism.
In a new paper, colleagues and I try to fill this gap with a new theory of decision-making: bracketing. Bracketing says to base our decisions on those consequences we are – in a precise sense – not clueless about, “bracketing out” the others. The idea is that the effects we’re clueless about should not override the obligations given to us by the benefits we aren’t clueless about, like the immediate benefits of malaria nets to the global poor. Thus bracketing could provide action-guidance in the face of cluelessness and in particular support neartermism.
- To be clear, bracketing isn’t a heuristic to help boundedly rational agents maximize total expected utility. It is a distinct normative theory from expected total utility maximization.
After an informal explanation, I give the formal definition of bracketing, and then discuss some of the major challenges it faces:
- There may be multiple ways of grouping “the consequences we’re not clueless about”, and these may give conflicting recommendations, in which case bracketing (at least the version we present) won’t be any more action-guiding than standard consequentialism.
- To formalize the “effects” of actions, so that we can talk about which ones we are and aren’t clueless about, bracketing requires us to specify the “locations of value”. These might be persons, spacetime regions, or something else. It turns out that neither persons nor spacetime regions is a great option (though unfortunately a proper discussion of why is out of scope for the post and paper).
- Bracketing will sometimes rank actions in a

...

Bracketing cluelessness