Following up on some thoughtful suggestions in the comments to my previous post, I ran a couple follow-up studies with modified stimuli.
One main question was whether people understood an “unreliable memory” to mean (1) most of the agent’s apparent memories contain false information, as opposed to (2) most of the agent apparent memories contain true information, even though the agent fails to retain information most of the time. If people understand “unreliable memory” in sense 2, then it could complicate the interpretation of one of the findings that undermine the proto-reliabilist hypothesis. Will we observe similar results if it’s made clearer that sense 1 is at issue?
To emphasize sense 1, I tested the following pair of cases (the reliability manipulation is bracketed):
Alvin is [unreliable/reliable] at remembering driving directions. Usually when it seems to him that he should make a particular turn, he’s [incorrect/correct]. Today Alvin is visiting a friend in an unfamiliar town. Alvin needs to pick up a prescription while he is there, so his friend gives him directions to the pharmacy. On the way, Alvin needs to turn right at Main Street. Alvin gets to Main Street and turns right, which is the correct turn.
People rated whether Alvin “knew” or “only thought” that “he needed to turn right at Main Street.” The rate of knowledge attribution was very high in both conditions: 83% in the unreliable condition and 97% in the reliable condition. This numerical difference did not reach statistical significance at the conventional .05 level (N = 60, p = .098, v = .07). I then conducted a second follow-up with the same cases but a more sensitive 7-point scaled knowledge attribution. Mean knowledge attribution was again very high in both conditions: 6.13 in the unreliable condition and 5.71 in the reliable condition and the difference was not significant (N = 61, p = .290).
I believe that these results should make us more confident that knowledge ordinarily understood does not require reliability in sense 1. Of course, these results cannot show that there is no conceptual connection between reliability and knowledge, but the relationship does not appear to be (in the ballpark of) a necessary condition.