A simplified measure of interference. The impact of interference requires a little more consideration. In the next chapter, we will look at the full mathematical form of the above argument. If followed to the letter, this leads to a horrendously complex calculation for the agent to perform for every memory. But as we have noted, our agent, although she may be ideally rational, is not cognitively infinite: she has, much like a human, extensive cognitive limitations. Now, the determination of how often an item X is likely to be recalled when it is unhelpful requires a rather unpleasant three-step deduction, a knight’s move away from the concept of X itself. Most problematically, it requires the agent to identify the situations in which remembering X may cause interference – by no means an easy task. As usual, there are a minority of cases in which it may not be too hard to xx- xxxx: consider someone with an upcoming driving test. The learner is told some amusing fact about her driving examiner, which – although perhaps fascinating – she certainly does not want to bring to mind during her test, as it will (a) cause a lapse in her concentration, and (b) cause her visible hilarity which may impair the examiner’s assessment of her. In such a case, the impact of remembering the fact will be very large – indeed it is likely to dominate the terms in eq. (3.1) on page 51, giving a rational prescription for a very low encoding strength. In more run-of-the-mill cases, however, the utility impact of interference may be much harder to assess.15 In this case (bearing in mind our aim is to find a set of heuristics that the rational but cognitively-limited agent should follow to find her encoding strengths), we would like a more tractable proxy for its value. In particular, we would like a proxy that does not require the three-step computation that should properly be required, wherein the agent identifies the cues connected to X, then identifies other structures already in memory connected to those same cues, and then considers the situations in which those other structures are relevant. In this instance, a starting point is to take as a proxy the first step of the above derivation. If the agent can at least identify how ‘unique’ X is in terms of its likely cues, she can use this as a very rough proxy for the importance of its interference contexts (given the computational difficulty of going on to consider all these situ- ations, a two-step away). If we accept this – and note it is again only a guide for stan...
