The quest for non‐Bayesian decision‐making heuristics: Towards a logic of imagination

• Published date: 01 December 2023
• Author: Armand Hatchuel
• Date: 01 December 2023
• DOI: http://doi.org/10.1111/emre.12625

EDITORIAL

The quest for non-Bayesian decision-making heuristics: Towards a

logic of imagination

INTRODUCTION

In this special issue, papers explore different trends of

non-Bayesian decision heuristics. They all address, to

some extent, the challenges of managing in the

unknown, that is, in contexts where surprises, discover-

ies, and innovations are frequent events. After a brief

reminding and synthesizing of four main conceptual

assumptions of Bayesian decision making (Savage,

1951,1972), we will develop a quick overview of non-

Bayesian heuristics that we classify according to the

Bayesian and Savagean assumptions they selectively

reject. This approach hopefully captures the manage-

ment issues and world views that each trend tries to

address.

The development of Bayesian decision making in the

1950s has been a major achievement both for theory and

for practice (Raiffa & Schlaifer, 1961; Savage, 1951,

1972; Wald, 1950a,1950b) in deep interaction with man-

agement science (Erickson et al., 2013; Giocoli, 2013). In

theory, utility functions for probabilistic lotteries allowed

to compare decision alternatives facing uncertainties; in

addition, the introduction of—so called subjective—prior

probabilities of uncertain events made possible to com-

pute, through Bayes’s conditional probabilities theorem,

the value of a new information about these events.

Bayesian theory is still highly influential in several fields

(statistics, medical science, finance, cognitive psychology,

AI, …) and in industrial practice when investment poli-

cies face costly uncertainties like in pharma or oil and gas

explorations.

However, Bayesian probability and logic have been

also a highly controversial topic in statistical and man-

agement science (see, in particular, Hey, 1990,

Lachman, 1990, Shackle, 1972,1983). It was often

argued (see, for instance, two recent special issues in

European Management Review, Elmquist et al., 2019,

and Academy of Management Review, Alvarez &

Porac, 2020) that the assumptions underlying the

Savagean model were not adapted to major aspects of

organizational life that are now commonplace for com-

panies, states and consumers, namely:

i. Contexts where fast and unpredictable changes occur

and where unknown situations emerge and not only

known—yet uncertain—events,

ii. The routinized mobilization of entrepreneursh ip,

research, design, and innovation that forces to recog-

nize that unknown events are also proactively sought

and provoked by organizations.

Actually, criticism of the Bayesian model also indi-

cates directions of investigation and potential progress.

As often in science, research tends to keep existing

models as much as possible, while attempting to reject

the most unrealistic assumptions (Ehrig & Foss, 2022;

Grandori & Cholakova, 2013) and research has explored

new families of heuristics that extends Bayesian decision

framework (Ehrig & Schmidt, 2022; Feduzi et al., 2020).

Such endeavors are expected to, hopefully, offer tractable

alternatives to the standard approach.

THE MAIN CONCEPTUAL

ASSUMPTIONS OF BAYESIAN DECISION

MAKING

Bayesian decision theory (Savage, 1951,1972) was built

by combining conceptual and technical axioms coming

from rational choice theory and from Bayesian probabil-

ity theory. These axioms can be synthesized through four

conceptual major assumptions that we remind broadly

below, avoiding unnecessary mathematical details.

Assumption 1: uncertainty in a closed and known

world. The Bayesian decision model assumes (i) a

fixed list of decision alternatives D that are opera-

tional and actionable; (ii) a fixed set of events E that

are all “uncertain”: meaning that the decision maker

ignores which one will happen; yet, it must be men-

tioned that the events that are elements of E are per-

fectly known, only their occurrence is uncertain. No

lack of precision or incompleteness in the definitions

of D and E are assumed (Hey, 1990; Loch

et al., 2008; Shackle, 1972).

Assumption 2: measurability of decision utility with

uncertain events. Coming from classic rational choice

this assumption means that it is possible to measure

(and compare) the utility (or costs, or consequences)

of a decision alternative in D when any event in E

occurs. Thanks to this measure, the decision maker

can evaluate the variations of the utility of each

DOI: 10.1111/emre.12625

632 © 2023 European Academy of Management. European Management Review. 2023;20:632–637.wileyonlinelibrary.com/journal/emre