Preproceedings 2006 - Austrian Ludwig Wittgenstein Society

3. Potentials and limitations of decision
theory
The incertitude of consequences is elaborated extensively
within decision theory and risk analysis. So let us restrict
for the moment to this kind of ‘uncertainty’. Here further
differentiations seem necessary and we distinguish three
different kinds of incertitude of consequences which differ
by their degree of incertitude (e.g. Leitner 2004; Knight
1948, 19f.): Risk is defined as a setting in which all
possible outcomes of the decision are known and can be
assigned some frequency which offers some confident
estimate of the occurrence probability of the corresponding
outcome. Uncertainty is defined by a setting in which again
the whole set of outcomes is known but not for all
outcomes can one assign the corresponding frequencies.
Situations where one lacks knowledge not only on the
probabilities, but on (part of) the outcomes too, are called
decisions under ignorance.
Actual situations quite often entail elements of risk,
uncertainty, and ignorance: There is a more or less
continuous gradation of incertitudes of consequences
running from the ideal case (choices under certainty) to
complete ignorance. Classifying a certain decision
situation as a situation under risk, uncertainty or
ignorance, is somewhat arbitrary. Here the second kind of
incertitude of demarcation comes into play: The wider the
chosen decision horizon, the more likely we face a
decision under ignorance; narrowing down the decision
horizon, we will encounter a risky situation. Although
incertitude of consequences and demarcation are thus
interrelated, the two are conceptually different. The same
is true for the third type, the incertitude of reliability. This
motivates the above classification of the incertitude of
consequences by its degree where we rather follow the
nomenclature used in ethics of technology than in decision
theory. The terminology used in decision theory (e.g.
Luce/Raiffa 1957, 13) might blur the differences between
probability estimates via frequency approaches and
probability estimates via subjective probabilities. But as will
be detailed in the following, it is the way one handles the
incertitude of reliability that determines which probability
estimate should be chosen. The incertitude of
consequences is treated in a different way.
For the moment we assume the decision horizon to
be fixed and focus on decisions under risk. Probabilistic
decision models which do not restrict to the mean, but take
into account also higher order moments of the probability
distribution, indeed provide a suitable way of how to deal
with such decisions. We refer to the literature with regard
to the debate on which kind of probabilistic model is
adequate. Here it shall be only stressed that this question
can be settled completely within decision theory —
presupposing that the consequences have been evaluated
‘properly’ in a preliminary step. How to actually do this, is
again determined by ethical reasoning and not by decision
theory. Though this constitutes a severe problem, it is not
specific for ethical considerations under ‘uncertainty’ (e.g.
Hillerbrand 2005). Hence it will not be discussed here.
In order to make any probabilistic decision model
fruitful for decisions under uncertainty, one has to assign
‘subjective’ or ‘personal’ probabilities to those outcomes for
which frequencies are not known. More sophisticated
approaches reintroduce in a second step uncertainty into a
closure ansatz: So-called ‘second level subjective
probabilities’ or ‘higher order beliefs’ estimate the reliability
of the assessed occurrence probability of the various
outcomes. Elementary, i.e. non-probabilistic decision
124
Uncertainty as a challenge for ethical reasoning - Rafaela Hillerbrand
models such as Maximin, do not have to make the
additional assumption of a subjective probability. At a first
glance they thus might appear to be most suitable for
decisions under uncertainty.
Nonetheless probabilities seem to be of importance
for a moral assessment. For example, the total harm of a
decision might depend on the number of people harmed;
this is often codetermined by the occurrence probability of
the harm. Furthermore, an overall preference of
frequencies to subjective probabilities does not seem
adequate in all cases. For example considering the
shortcomings of present climate model, it is not at all clear
whether or not the (few) existing frequency estimates of
the occurrence probability of a raise in sea level due to
anthropogenic greenhouse gases are indeed more reliable
than subjective probability estimates of some specialist.
Here it becomes crucial to estimate the reliability of the
probability estimates. Although it is not the primary task of
moral philosophy to reason about the reliability of scientific
forecasts, it seems up to ethics to provide the decision
maker with a kind of ‘threshold’ that indicates when a
reduction approach is appropriate, i.e. that indicates when
to accept subjective probability estimates and treat
decisions under uncertainty or even ignorance in complete
analogy to decisions under risk. Here it seems that —
although we argued for a general guideline, a probabilistic
decision model, when facing decisions under risk — there
is no general, context-independent answer to that
question.
The problems encountered in setting this threshold
parallel (to some extent) the problems encountered when
classifying a given decision situation as a decision under
risk, uncertainty, or ignorance. Here the second incertitude
of demarcation comes into play. A formal way of how to
deal with it was proposed by Savage 1954. But the
relevant question for tackling this incertitude of
demarcation, namely, to state criteria on how to form, in
Savages terms, the ‘small worlds’ from ‘the grand world’, is
not answered satisfactorily by Savage or in any other
decision theoretical approach (Savage 1954, 16; Spohn
1978, 63; Schmidt 1995, 57). Without having to address a
specific example, difficulties seem to arise whenever one
wants to state context-independent criteria to determine
the decision horizon.
4. Outlook: Judgment as a supplementation
of rules
The analysis of the forgoing section seems to reveal that
with respect to certain aspects of ‘uncertainty’ a solution
detached from contingent features of the decision situation
is out of reach. With respect to the colloquial use of the
term ‘uncertainty’ M. Luntley notes something similar
(Luntley 2003, 326):
The ethically competent needs general rules, but these
are not what primarily lie behind ethical competence in
decision making. Wise judgement is not constituted by
grasp of general rules, but by the attentional skills for
finding salience in the particularities of situations. The
important element of decision making [...] is the element
that turns on the possession and operation of these
attentional skills.
We suggest to identify the conceptual skill Luntley
mentions with Aristotle's dianoetic virtue of phronesis (e.g.
Ethica Nicomachea, VI 8-19) as it is used for example by
O. Höffe (1993). Then judgement labels a certain ability
and willingness to identify and to implement the ways and