Boundless Rationality

I am inspired by a great paper that Adam Piovarchy is working on, and the paper ‘Dead rats, dopamine, performance metrics, and peacock tails: proxy failure is an inherent risk in goal-oriented systems’: to sketch out a model of rationality that I have been thinking about for a while. In short: means-to-an-end rationality must be a deeply incomplete account of human rationality, but could be seen as a part, not the whole.

The thinking goes like this: there are a number of internally consistent accounts of rationality which serve a certain purpose, and work well at their assigned sort of task. These are things like the rational actor model (Homo economicus) and the rational thought which logic is meant to model. Call these micro-rationalities. The problem is that at the level of the whole human, decision-making is complex and, when examined holistically, is not simply a process of choosing means to an end or running the numbers. If we are to say that one human is more rational than another, in a very general sense divorced from some particular tasks or context, we must be talking in terms of more of a meso-rationality.

What distinguishes meso-rationality? Its plurality and its targetlessness. A complex being like a human must have a number of different tools to draw on when navigating the world and processing information. If it has more than one need (and by any standard a human has more than one of these), then it must be able to balance competing demands on its activity and attention. It will want to avoid fixation on a single end, both for this reason, and also for strategic reasons if it has ends which are at odds with those of others (in a competitive context, an agent does not want to behave in the sort of limited and predictable way a chess piece does!). At the very least a human needs to:

• Identify inconsistent beliefs and decide what to do about inconsistencies
• Identify efficient means to ends
• Compare ends, identifying how our values relate to each other and which have priority
• Manage processes of value change within ourselves
• Diagnose an apparent problem of choice, and decide how much effort we are prepared to put into solving it, and therefore what sort of ‘rationality’ to engage in the attempt
• Seek opportunities to exercise and develop the rational capabilities through engagement with the world (c.f. the extended mind) and with others (c.f. intersubjectivity, information sharing).

Looked at this way, we might start to question Herbert Simon’s decision to call his account ‘bounded rationality’. For the accounts he was critiquing are narrow accounts of rationality which produce such problems and ‘paradoxes’ as:

• Proxy failure (such as the hedonist’s paradox)
• Strategic dead-ends a la the prisoners’ dilemma and Newcomb’s paradox
• Various philosophical errors such as the ‘open question’ argument

And it is precisely the narrowness, the boundedness, of micro-rationalities which do this when they are deployed outside of their proper context. A being which is as prepared as can be for whatever an external world throws at it must have much more flexibility in its rational capabilities.

Photo by Keith Hardy on Unsplash, cropped.