Computer Models as Existence Proofs

A common objection to the use of computer models is on the ground of their lack of realism. Thus, for example, it might reasonably be objected to Axelrod's famous work on the Iterated Prisoner's Dilemma that the model of human interaction that it presupposes is a mere caricature, far too crude to deserve serious consideration as any sort of representation of reality.

Even where this is admitted, however, the computer model can remain valuable, because such models often function – for the philosopher, at least – not so much as attempted descriptions of reality, but rather as thought experiments or existence proofs. Suppose, for example, that someone were to claim that evolution by natural selection inevitably favours selfish behaviour, so that altruism cannot possibly evolve. In that case, the demonstration that altruism can indeed be shown to evolve within a simple model serves to refute the claim by counterexample, and the fact that the model falls well short of reality is entirely beside the point. If the claim is then refined to suggest that altruism is extremely unlikely to evolve, then this too can be countered by the demonstration of a wide range of models within which altruism does evolve. And again, the fact that these models may be simplistic and unrealistic is irrelevant, unless the claim can be plausibly and explicitly backed up by relevant considerations that the models fail to capture. Thus even where models are, in the end, unpersuasive, their use can serve a positive purpose by forcing such considerations into the open.

Generating Counterexamples

Philosophers characteristically aim to establish general claims about what is, or is not, possible. And claims of impossibility are standardly tested by attempting to produce counterexamples: existence proofs of a possibility, which thus refute the impossibility claim. Such testing does not typically require that the counterexamples be plausible, and this can make them especially hard to come up with unaided, limited as we are by our reality-conditioned human imagination. Computer-generation of counterexamples can – in appropriate cases – provide a powerful solution to this problem, automatically searching through a combinatorial mass of options that would be unfeasible without such assistance.

Avoiding "Just-So" Stories

Evolutionary or anthropological accounts of observed phenomena (e.g. biological traits or cultural practices) encounter a serious problem of substantiation and verification. By their very nature, they appeal to historical trains of events that cannot be reliably reconstructed by appeal to straightforward causal laws, nor typically proved from historical evidence. Even where detailed evidence is available, many of these accounts are hard to connect directly to that evidence, since they take a narrative form which explains long-term trends as arising not from specific events, but rather from general tendencies that play out over time at the level of population statistics. Opponents of such accounts have often called them just-so stories, implicitly criticising them by "reminding the hearer of the essentially fictional and unprovable nature of such an explanation".

Until the rise of computer modelling, the most successful way of countering this sort of criticism – at least within the field of evolution – was to produce a mathematical model of the phenomenon in question, using the techniques of Population Genetics. Mathematics could then replace narrative as the medium of explanation, demonstrating that conclusions were being drawn from the theory with rigour rather than undisciplined imagination. But this virtue is counterbalanced by some significant weaknesses. Population Genetics focuses on gene frequencies, ignoring epigenetic influences and feedback mechanisms; likewise it tends to assume a simplistic genotype/phenotype relationship which takes no account of embryology and development. Though valuable for injecting rigour at the "broad brush" level of genetic change in populations over time, therefore, it is inherently limited in scope.

Computer modelling provides a far more versatile and powerful technique for replacing narrative accounts of long-term change with rigorously specified models. As pointed out elsewhere, "computerised thought experiments – by their very nature – require explicit statement of all assumptions, and their outcome cannot be unconsciously moulded by wishful thinking". So a computer model of an evolutionary or anthropological story, though of course it might be unrealistic or simplistic, at least has the considerable merits of explicit precision and constrained outcomes. Whatever it is that results from such a model (if reliably implemented) must genuinely be a possible consequence of the assumptions that have been built in. So if we are able to produce a model that does indeed match observed phenomena, then we have the makings of a hypothesis for explaining those phenomena which may (of course) be false, but which at least cannot be accused of narrative vagueness and optimistic gerrymandering. So far from being a "just-so" story, it is an existence proof of a possible account.