Abstract for Bristol

Posted on March 28, 2009


Here is the text of the abstract I sent to Bristol. Can’t say that I’m altogether happy with it but I hope that it will be accepted and that I will improve upon it before presenting. I would have tried to do more with it but I am sorely lacking in time right now.

Evolution, generative entrenchment and the bounds of rationality

Human rationality is bounded. Is this due, however, to it being the product of evolutionary processes or to some more fundamental limitations? If, indeed, there are any. Furthermore, does the evolutionary legacy place additional limitations upon human rationality that are fundamentally different from those placed upon all boundedly rational beings? I will argue that the boundedness of rationality is fundamentally a response to the insolubility of Hume’s problem and, as such, a property of all reasoning beings. The flip side of boundedness, however, is the open-endedness of rationality afforded by the capacity to improve reasoning abilities by the addition of new heuristics, reliant as it is upon generative entrenchment. Again due to Hume’s problem, the processes by which such new heuristics are arrived at will be broadly evolutionary in nature. The conclusion is that the sets of evolved agents and those that are bounded but open-ended are co-extensive but that their fundamental properties are due to a fundamental epistemic consideration.

Bill Wimsatt (2007) describes human rationality as the jerry-built product of tinkering evolutionary changes. Given the recognition of adaptations as heuristic in nature, evolution is seen to consist in the development of new heuristics as well as the exaptation and generative entrenchment of existing ones. The connection may lead to an identification of bounded rationality with evolved rationality and the view that there may be other kinds of rationality that might avoid such limitations. After all, engineers need not tinker. Something of this view may be thought to motivate the dual process account of reasoning pursued by Evans (2003) and others, who believe human cognition is to be understood as consisting of heuristics and a separate capacity for abstract thought. Not coincidentally, however, the dual process account is highly reminiscent of Hume’s distinction between Reason and habits. Unfortunately for that approach, Hume’s original reason for preferring habits has not been undermined despite two and a half centuries of philosophic effort. At the same time, heuristics – as developed by Herbert Simon (1969) – can be seen to spell out what Hume pointed to by talk of habits. The resulting account, particularly when combined with the work done by Johnson-Laird (2006), is powerful enough to subsume even the abstract reasoning that Evans and others would juxtapose with their own much weaker notion of heuristics. In effect, heuristics can be seen as the naturalistic answer to Hume’s problem – not to show the error in Hume’s argumentation but to see that he has articulated a fundamental limit to reasoning. Indeed, most significantly, the problem of induction can be seen to have impressed its handprint upon evolution, itself: as visible in the myopic character of that process.

Still, even if Hume’s problem is everyone’s (and everything’s) problem, it may seem that engineered heuristics may at least avoid the jerry-built character of the results of evolutionary changes, as well as make available particular solutions that are not obtainable through other means. To some limited degree this may well be the case. However, living organisms already use a plethora of different heuristics and many limitations on what heuristics they do use are historical rather than simply due to their biological underpinnings. Any additions that engineering may offer seem, therefore, unlikely to be fundamental in nature. More important, however, is the question of whether the engineered systems are to be open-ended, i.e. whether they will be capable of developing novel heuristics as well as of putting existing ones to novel uses. This capacity has been crucial in allowing evolution to lead to, among other things, human cognition: without it, engineered systems are severely constrained. Such step-wise development is the only option given Hume’s problem but it necessarily introduces generative entrenchment that leads to the history-dependent, jerry-rigged nature of human reason. This means that even engineered reasoning systems will necessarily take on the same character as evolved ones, if they are allowed to develop. In other words, due to also being subject to Hume’s problem, the resulting process will lead to products that are akin to those of evolutionary processes.