AI - The Benevolent Dictator?

The Acceptance of Automated Decision-Making and its consequences

Duration: 1 January 2019 until now
Project member(s): Dr G.J. Schaap (Gabi) Prof. T. Bosse (Tibor) Dr P.G.J. Hendriks Vettehen (Paul)
Project type: Research
Organisation: Communication and Media, Behavioural Science Institute, Faculty of Social Sciences

Abstract

Automated decision-making (ADM) is rapidly becoming ubiquitous. Aided by big data and advanced machine learning, AI and algorithms are applied to tasks that were previously exclusively human prerogatives. ADM ranges from professional to private domains, such as financial investment, medical diagnoses, dating, and personal fitness.

What’s more, when it comes to decision-making, algorithms are getting as good as or better than we are. This raises an important question: Do we want machines instead of humans to make consequential decisions in our personal lives? And what will the human consequences be if we do?

The answer is far from clear. While past research has suggested a general aversion to letting intelligent machines make decisions for us, recent studies suggest that when push comes to shove, we are pretty happy to cede decision-making authority as long as it leads to better outcomes.

This project delves deeper into the processes behind this: how we weigh the various costs and benefits of ceding decisional authority to AI. How do we trade off the costs of giving up our autonomous decisions to the expected benefits of more accurate, faster, or easier decision-making done by AI? And, importantly, if we increasingly choose to let AI make decisions for us, will this fundamentally change us as humans, such as our sense of autonomy or our ability to think and solve problems ourselves?

Find out more about the Communication Media research programme:

Communication and Media

Research Questions

In its first phase, the project clarifies the literature on so-called Algorithmic Aversion: whether a generalized aversion to the use of ADM exists and what psychological mechanisms underlie the acceptance or rejection of ADM.

The Main Research Questions here are:

Do users prefer human decision-makers (including themselves) to AI decision-makers?
How do people trade off ceding personal decision agency and the benefits of ADM (accuracy, efficiency, ease, etc.)
Is a preference or aversion to using ADM dependent on the domain it is applied to (e.g. finance versus dating, or medicine)?
Is acceptance or rejection of ADM dependent on a person’s basic psychological need (e.g. autonomy, competence, or comfort)

We hypothesize that whereas in an abstract sense, people have perhaps a slight aversion towards ADM (based on their basic psychological needs), in actual interactions with ADM, a costs-benefit calculation provides the underlying mechanism used to determine whether or not to use ADM or make the decision themselves. The first studies published show that optimization of benefits plays an important role in accepting ADM; more studies are underway.

The Benevolent Dictator

In a second phase, called ‘The Benevolent Dictator’, we aim to find out where our limits are, and what the consequences are of letting AI make many important and everyday decisions. Research Questions include:

What optimization benefits (success optimization, accuracy, efficiency, ease, loss avoidance, social pressure, etc.) determine the acceptance of ADM?
What human costs may determine the rejection of ADM (e.g. agency, competence, sense of self-actualization, human ‘uniqueness’, etc.)?
Do we have a tipping point at which the benefits of ADM will be so high that we will give up on the factors mentioned under 2?
What are the psychological consequences of interacting with (superior) AI decision-making, for instance, for our sense of agency and competence, and motivation?

Research methods

The project mostly uses experimental research to investigate these questions. To go beyond abstract beliefs and attitudes towards this technology, we create concrete decision paradigms in which participants are confronted with ADM scenarios or interact with ‘AI’ to study their decision-making behavior. In the second phase, immersive game-like environments will be created to study participant behavior in incremental benefits (and dictatorial) scenarios. Other research methods (survey and qualitative interviews) will be employed further down the line.

Results

Schaap, G., Bosse, T., Hendriks Vettehen, P. G. (2023). The ABC of algorithmic aversion: Not agent, but benefits and control determine the acceptance of automated decision-making. AI & Society

BNR Podcast: Willen we meer of minder algoritmes? (in Dutch)

Contact information

Contact

Dr G.J. Schaap (Gabi)