On Philosophy

September 3, 2007

Memory As Evidence

Filed under: Epistemology — Peter @ 12:00 am

Almost all reasoning ultimately rests on memory, if only short-term memory. But memory is notoriously unreliable. This might seem to raise a kind of epistemological paradox. If memory is known to be unreliable then how can we know that memory is in fact unreliable? Isn’t our judgment that memory is unreliable based itself on memory to some extent? Obviously one way out of this dilemma is to note that memory is not completely unreliable, and that we can predict to some extent when it will be more or less reliable. But this doesn’t rule out the skeptical possibility that memory is in fact much less reliable than we think, and that its inaccuracies are falsely leading us to believe in its general accuracy.

But before we address the skeptical possibility let us return to the idea that the problematic nature of memory can be dealt with simply by quantifying its unreliability and taking that fact into account. One way of doing this would be to assign a confidence value to each memory, representing the probability of that memory being accurate, with various factors raising or lowering that confidence value, such as the age of the memory. These confidence values then influence the confidence of any conclusions drawn from them, with conclusions that have a confidence value over some threshold being candidates for knowledge. And, if we wish to make things slightly more complicated, we can also inspect whether the confidence value is increasing or decreasing, and make our determinations based at least partly on that (I personally find that to be a more accurate measure, since it tracks whether the proposition is trending towards confirmation or disconfirmation).

For the sake of charity let us assume that this system has been worked out in full detail, and that our confidence in it is high. One consequence is that the farther away from events we get the less we know about them, which makes sense. Consider then a man who was present at the assassination of Kennedy, and who from his memories of that day is able to conclude that no one was on the grassy knoll. The system of confidences implies that at some point in the future this same man will be unable to know that no one was on the grassy knoll, even if his memories of the event have never changed. And, moreover, if he derives this conclusion from his memories every day there will be one day on which he can know that no one was on the grassy knoll, and yet on the next he can’t, because the confidence in that conclusion has fallen below the necessary threshold. Of course that absurdity may very well be a product of using fixed thresholds to determine when some conclusion is knowledge and when it isn’t. The idea of relying on the change in confidence to determine when something is knowledge is intended as one way around such problems, but here it is no help at all. Since memories are always becoming less certain relying on the change in confidence would imply that we can’t know anything on the basis of memory alone (unless it has recently been supported by new memories).

But perhaps it is unfair to pick on a confidences system for those reasons, because all confidence systems tend to suffer from such problems (meaning that the problems lies not in overcoming the fallibility of memory by estimating the likelihood it is correct, but with how we draw conclusions about what we do and don’t know on the basis of them). However, this isn’t the only problem that such a confidence based approach suffers from. It also fails to capture how people actually reason on the basis of memory; people certainly do not critically evaluate the likelihood that their memories are in error unless something has prompted them to consider that possibility. A confidence system says that we are in error when we treat a conclusion as equally justified when it is based on a memory of events from a few minutes ago as when it is based on a memory of events that happened yesterday, or even a few hours ago. Of course it is not impossible that our usual patterns of reasoning are wrong, certainly there are many examples of that. But certainly our naïve approach to memories seems to work, at least most of the time. Given that it is natural to inquire as to how we actually reason on the basis of memory, and whether that approach or something like it has any merit.

And besides failing to capture how memories actually factor into reasoning the confidence system fails to address skeptical worries (such as the possibility that our memories change completely from second to second, but always form an internally consistent set). Fortunately there is a way of solving both problems at once, by treating memories as evidence. As I detailed previously we are justified on reasoning on the basis of evidence, propositions that can possibly be shown to be false, and which are independent of judgments. Most memories fulfill both criteria, recalling a memory brings with it a recall of experience, which is not a judgment, and memories can be shown to be false by conflicting with other memories or experience. Thus reasoning on the basis of memories is perfectly acceptable, because it is likely to lead us to true beliefs. Of course I say that most, and not all, memories can be treated as evidence, because there are some that are not defeatable (cannot be shown to be false). For example, if you remember your long dead uncle preferring blueberry to raspberry jam that memory may not be defeatable if there isn’t anyone around who knew your uncle better than you and he hasn’t left anything behind to indicate his preferences on this matter. Such memories are not problematic, even if we don’t realize that they are undefeatable and treat them as evidence. This is because, to invoke another epistemic principle, if a claim is undefeatable it can’t concern anything that matters (in the sense of what we want knowledge to do). Either evidence leads to conclusions that might be discovered to be false, in which case those conclusions might matter to us, or it is undefeatable, in which case nothing can be found that will either confirm or deny it or any of its implications, in which case it doesn’t matter.

Blog at WordPress.com.