Vercelli’s Keynesian Learning
Alessandro Vercelli Uncertainty, rationality and learning: a Keynesian perspective, in Dow and Hillard.
This rejects the ‘decision-by numbers’ version of learning, recognizing the need to make meaningful decisions about options, making as well as taking decisions.
Bayesian rationality itself is unable to analyse in a satisfactory way the economic role and implications of genuine strategic learning and of the feedback between learning and rationality.
Bayesian rationality rationalises away uncertainty, and thus is unable to take a strategic approach to it.
[T]here is a strict correspondence between diﬀerent notions of rationality and learning, modalities of uncertainty, and assumptions on the degree of time irreversibility of the consequences of economic decisions.
[A] purely adaptive form of strategic learning is consistent with procedural rationality… . However, in order to study the highest form of strategic learning, which will here be called ‘creative’, a broader concept of rationality is required, which will be called ‘designing’.
[T]his approach must be extended to the case of an open world which may change in an unforeseeable way and where the environment of the decision problem may be modiﬁed by the decision maker.
I call this kind of rationality ‘creative’rationality because in this case the [decision maker] is not seen just as an option taker but also as an option maker (Vercelli, 1991: ch. 5). Similarly to adaptive rationality, creative rationality may be studied exclusively from the point of view of the optimal or equilibrium structure, which is in this case diﬀerent from the existing one, or it may be studied also from the point of view of the structural transition to conﬁgurations considered better than the existing one. Creative rationality may therefore be distinguished in ‘utopian rationality’, when only the optimal conﬁguration is considered, or ‘designing’ rationality whenever also the transition processes are considered. We have to conclude that only designing rationality is fully consistent with structural learning and applies in a satisfactory way in an open non-stationary world.
These pertinent insights are more accessibly expressed than Keynes’, and are more general, but along similar lines.
In ‘large’ or ‘open’ worlds one have a limited horizon beyond which the possibilities are increasingly opaque. One can never ‘solve’ one’s problems: one can only resolve them, strategically. This will include being able to assess one’s own horizon and often the horizon of others, subjective or objective.
It will often be that the best one can do is to minimise the risk of the worst possible outcomes, rather than optimise. Thus to learn from an apparent failure one should ask, not ‘given how the situation turned out, how could this be avoided?’, but ‘were there alternative strategies that reduced the potential for this bad thing without making anything worse more likely, and was there any way that we ought to have been able to identify any such strategy?’