Guth ea.’s Coping
Economics Special Issue on Radical Uncertainty and Its Implications for Economics
[Exposure] to theoretical disciplines does not prepare well for coping with practical uncertainties. …
Psychologically it seems that the ascent of formal decision theoretic modelling in practical disciplines has nurtured the spurious belief that formal rigor can transform uncertainty into risk.
2. Modeling uncertainty and ignorance
2.1 The historical and present relevance of Pascal
The mathematical formulation lured many into a false belief that it was only [probabilistic] risk they were dealing with. That they had a clear incentive to believe did not support a critically rational attitude either. Moreover, withholding judgment in a disinterested way was impossible since unavoidably inaction … is a form of action … .
2.2 The standard model of rational decision making
The focus on the causal structure is a crucial aspect of all rational-choice modelling simply because representations of causal relations underlie all decision-making on purposefully rational intervention into the course of the world. There are:
(i) plans, s, functions – comprised in a set S
(ii) states, z, of the world – comprised in a set Z
(iii) results, r, that emerge – comprised in a set R
(iv) technologies, f, described by functions f : S × Z→R.
The standard model assumes that the set Z of states of the world and a probability distribution p on Z are known.
3. Uncertainty as risk?
[To] the extent that real individuals must make a decision without … empirical evidence the fiction of a complete model loses its practical usefulness and becomes a risk factor itself.
4 Practices of coping with uncertainty in interactive decision-making
4.1 Other minds as a source of uncertainty
According to the standard view, so-called mixed behavioural strategies are probability distributions over choices of an actor that describe not the randomness of her choices … but rather the ignorance of other actors.
That other individuals despite the complexity of their mental models seem predictable to us is related to the presence of institutions. … Yet in view of the fact that human actors can always decide on the basis of their own idiosyncratic models of the future it seems that this will not work in cases that are “new” in the sense that no (institutional) rules for these cases yet exist.
For participants of interaction the strategy sets, the states of knowledge and the likelihood of actions they cannot control will more often than not be such that the situation should be modeled as one of uncertainty rather than risk. In particular a Bayesian risk model approach with “given subjective probabilistic beliefs cum utility representations”, though adequate for ideal game theory, will be inadequate for describing what is actually going on among real boundedly rational decision- and choice- makers.
4.2 Scenario-based practices of coping with complex decision-problems
… A boundedly rational actor can focus merely on a few [implementable strategies] along with exemplary scenarios [among those implementable by others] to form potential case solutions .. .
Relying on the fiction of unlimited rational capacities is useless from the internal point of view of a boundedly rational decision-maker who actually needs to make choices. She could not handle the complete information even if she had access to it. She selects scenarios [representing others’ possible strategies] to form scenario-specific aspirations … of goal achievement (she deems fulfillment important in view of her broader aims, ends or values). …
If an option satisfies her aspirations for all scenarios this can be a stopping point [else] the actor … may adapt her aspirations … .
4.3 An outline of a decision procedure for coping with uncertainties
An iterative process is described, from mental modelling through scenario generation, aspiration formation to satisficing serach. It ends with a ‘final choice’ when there is a satisfactory solution to the problem pose by the mental model, scenarios and aspiration.
To prepare the ground for learning from scenarios and interventions into the course of the world a boundedly rational actor needs to have some hypotheses concerning the evaluation of steps (moves, actions) she is making. In short, she has to select which variables are to be used in her mental modeling as short-term indicators of long-term success. We will not go over the details of this rather crucial process of finding indicators for the fulfillment of the underlying larger aspirations. We can emphasize merely that the quality of end point results will crucially depend on making the right choices on the level of “intermediate” variables.
4.4 Evaluating the decision procedure for coping with uncertainties
Which action from the internal point of view of the decision-maker should be chosen cannot be justified by utility and probability expectations that (after the fact) merely represent the choices made. … In this realm rationality must be procedural. It is about the rules that generate choices that is about their procedural and not about their so-called substantive rationality.
If our basic implicit empirical hypothesis is correct all successful kinds of decision support that might be charted out should be “prescriptive extrapolations of good practice” as characterized here. The prescriptions derived will always incorporate the distinction between what is and what is not controlled by the decision maker, beliefs, aspirations, and action variables. The focus on causality structures (see Pearl (2000)) based on hypotheses linking plans/actions, scenarios and goal achievement should be essential across the board.
…. Foundational uncertainty cannot be eliminated by a priori arguments. Yet we can act as good empiricists and try to form and test hypotheses on good practice. … We believe that much more piece meal research is necessary to gather evidence on how to cope better with uncertainties. … Starting from probabilistic expectations and utility functions representing preferences conceals the uncertainty that we face in realistically complex one-off decision problems. This being said it needs to be emphasized that whenever possible decision-making should be evidence based and the formulation of the problem as rigorous as possible. If mathematical language is used correctly it should make us alert of what we do not know rather than conceal it from our views as many models of full rationality tend to do.
I take from this that in any complex situation we will have only limited awareness, and within that only a limited model. We will need to develop indicators of incipient problems. This cannot be ‘rational’ in the narrow sense of relying on utilities or probabilities. Instead it must be based on some procedure that has proved reliable. It must thus be consistent with proper mathematics and logic.
Common practice is to update a model from evidence.
- If the updating makes some assumptions or has some implications, such as that certain statistics will remain in certain bounds, then these should be checked from time to time.
- If some statistics have been in a certain range for long time, any departure should be noted and considered.
- If certain variables that are directly influence-able are supposed not to directly affect certain others, then this should be checked from time to time by influencing those variables.
- There should be a trade-off between using variables to have a ‘real’ beneficial effect, and using them to check the model or learn about variables.