Don’t trust heuristics, unless you can (1) re-derive them, (2) know their limits explicitly, or (3) are willing to accept the risks for the moment, but will reevaluate them later.
I’m not sure that I read your point (3) correctly. One feature of heuristics is that you need to trust them to do a better job than you can do without them. As you refine the understanding of where the heuristics are appropriate, the expected effectiveness of heuristics increases, but all the way heuristics need to pay rent. The useful side of heuristic needs to win over the bias part, which is more of an issue for heuristics than for declarative beliefs.
(3) is the branch for when urgency prevents the use of cognitive or data collection resources needed to adequately trust the heuristic under normal circumstances, but that same urgency requires a decisions. Loosely speaking it is the emergency clause. So the heuristic for that branch is to use the best heuristic available under what resources you can muster, and schedule a reevaluation at a later time, to recover from the habit forming nature of what you might have just done.
Of course, many heuristics are far to complex to personally derive or even to fully and explicitly describe their limits (at least in a single evaluation), so instead we need to keep calling (3) to manage them even outside of a proper emergency. What this means is that heuristics “paying rent” is a sub-heuristic of the heuristic I propose here.
Of course the overall limit of this heuristic remains (it applies to the “rent charging” dynamic, as well as the other applications). To manage this limitation requires a higher level check (likely itself heuristic) to enable the aspiring rationalist to operate with greater caution in domains where self-trust is less reliable.
I’m not sure that I read your point (3) correctly. One feature of heuristics is that you need to trust them to do a better job than you can do without them. As you refine the understanding of where the heuristics are appropriate, the expected effectiveness of heuristics increases, but all the way heuristics need to pay rent. The useful side of heuristic needs to win over the bias part, which is more of an issue for heuristics than for declarative beliefs.
Vladimir,
(3) is the branch for when urgency prevents the use of cognitive or data collection resources needed to adequately trust the heuristic under normal circumstances, but that same urgency requires a decisions. Loosely speaking it is the emergency clause. So the heuristic for that branch is to use the best heuristic available under what resources you can muster, and schedule a reevaluation at a later time, to recover from the habit forming nature of what you might have just done.
Of course, many heuristics are far to complex to personally derive or even to fully and explicitly describe their limits (at least in a single evaluation), so instead we need to keep calling (3) to manage them even outside of a proper emergency. What this means is that heuristics “paying rent” is a sub-heuristic of the heuristic I propose here.
Of course the overall limit of this heuristic remains (it applies to the “rent charging” dynamic, as well as the other applications). To manage this limitation requires a higher level check (likely itself heuristic) to enable the aspiring rationalist to operate with greater caution in domains where self-trust is less reliable.