Artificial intelligence and radical uncertainty

• Published date: 01 December 2023
• Author: Ann‐Kristin Weiser,Georg Krogh
• Date: 01 December 2023
• DOI: http://doi.org/10.1111/emre.12630

VIEWPOINT

Artificial intelligence and radical uncertainty

Ann-Kristin Weiser | Georg von Krogh

Eidgenössische Technische Hochschule Zurich,

Zurich, Switzerland

Correspondence

Ann-Kristin Weiser, Eidgenössische Technische

Hochschule Zurich, Zurich, Switzerland.

Email: aweiser@ethz.ch

Abstract

Artificial intelligence (AI) offers new possibilities to augment human decision-

making under radical uncertainty. This viewpoint commentary explores how AI

can relax limits of bounded rationality. It offers a framework for analyzing how

AI can support human decision-makers confronted with deep uncertainty by bol-

stering key decision-making sub-processes. Specifically, AI can help set agendas

by scanning environments, formulate problems by providing contextual insights,

identify creative alternatives through combinatorial abilities, select options by

modelling scenarios and enable rapid experimentation cycles. Connecting the role

of AI with the contributions in this special issue, this viewpoint commentary con-

cludes by outlining directions for future research regarding the function of AI and

augmented human intelligence in decision-making under conditions of radical

uncertainty.

KEYWORDS

artificial intelligence, decision-making, uncertainty

INTRODUCTION

In recent years, the exponential growth in the utilization

and capacities of artificial intelligence (AI) has reshaped

the scientific and technological landscape. This surge can

be attributed to innovations in deep learning algorithms,

the proliferation of big data and advancements in compu-

tational hardware. As a result, AI systems are now able

to process information at unprecedented speeds, decipher

intricate patterns from vast datasets and execute tasks

creatively that were once deemed too complex or

nuanced for machines.

Given the increasing information processing capacity

resulting from the recent developments in AI, it is useful

to revisit the concept of bounded rationality in the con-

text of radical uncertainty and how our understanding of

the limits of bounded rationality might be changing with

the application of these novel technologies. Simon (1990,

p. 15) defined bounded rationality as ‘rational choice that

takes into account the cognitive limitations of the

decision-maker—limitations of both knowledge and

computational capacity.’He introduced the theory of

bounded rationality to challenge the three primary

assumptions of the subjective expected utility theory pro-

posed by Savage (1954) that assumed fixed decision alter-

natives, subjectively known probability distributions and

the maximization of the expected value of the utility

function (Simon, 1955,1957,1990,1997).

Instead of assuming fixed decision alternatives, Simon

emphasized the alternative generation processes. More-

over, rather than assuming known probability distribu-

tions, the emphasis shifted to procedures for estimating

the probabilities, or to methods addressing uncertainties

that cannot be assigned probabilities. Furthermore,

Simon observed that decision-makers rarely optimize but

rather satisfice expected utility. In the face of radical

uncertainty, these bounds to rationality become re-

inforced and decision-makers might even fail to recognize

an imperative or opportunity for decision-making, a

decision-making situation.

Here, we posit that AI could affect decision-making

in contemporary organizations, by relaxing some of the

limits of bounded human rationality when making

decisions under conditions of radical uncertainty. First,

AI may assist in generating more comprehensive identi-

fication of decision alternatives based on analogical

pattern recognition and creative combination from vast

amounts of prior data, similarly to what humans do

when developing creative strategies through innovative

combinations of existing elements (Martins et al., 2015;

Rindova & Martins, 2021). Second, modern generative

AI can assign probabilities even to rare, highly unlikely

DOI: 10.1111/emre.12630

European Management Review. 2023;20:711–717. wileyonlinelibrary.com/journal/emre © 2023 European Academy of Management. 711