HMN 2025: How Linking pay to efficiency boosts AI use in decision-making

Artificial intelligence has improved by leaps and bounds over the previous couple of a long time and has modified the way in which many individuals, together with company managers, conduct enterprise.

But the usage of algorithms in managerial decision-making is not common, and there are just a few components that spur better use of AI: how the supervisor will get paid, and the way synthetic intelligence is framed, based on a brand new study co-led by a Cornell researcher.

Contrary to extremely cited analysis from greater than 30 years in the past, an incentivized pay construction will result in better reliance on AI in decision-making than flat, fastened compensation. And if the AI is described as combining each knowledge and human knowledgeable data, persons are extra possible to make use of it than if it is framed as strictly algorithmic recommendation.

Martin Wiernsperger, assistant professor of accounting on the Samuel Curtis Johnson Graduate School of Management, within the Cornell SC Johnson College of Business, is a co-author of “Incentives, Framing, and Reliance on Algorithmic Advice: An Experimental Study,” published in Management Science. His co-authors are from the Vienna University of Economics and Business.

Two research from the Eighties and ’90s contended that incentivized decision-makers—these being paid both based mostly on efficiency, or in a “event” setting wherein the highest performer in a bunch will get the reward—will rely much less on algorithmic recommendation and as a substitute really feel compelled to “earn” the reward by way of their very own effort. In different phrases, the inducement backfires attributable to a phenomenon generally known as “algorithm aversion.”

“They have been fairly well-known research on the paradoxical results of incentives and decision-making, and I believed it is most likely value re-examining this query now, when now we have very various kinds of choice aids and algorithms,” Wiernsperger mentioned. “We needed to see if this backfiring nonetheless holds.”

He and co-authors Philipp Grünwald and Georg Lintner have been all doctoral college students in Vienna in 2021 after they attended a seminar and have been assigned to work on a challenge collectively. They designed an experiment to check how reward construction and the framing of AI would influence decision-making in a particular job: estimating a per-night fee for an Airbnb house. The different co-authors, professors Ben Greiner and Thomas Lindner, met the scholars on the seminar and advised they flip their challenge right into a analysis paper.

For their study, the researchers recruited round 1,500 individuals from three massive public universities in Austria. Subjects have been randomly assigned to considered one of 9 experiment circumstances: fastened pay; efficiency pay; or event pay, then inside every of these circumstances: no AI recommendation; AI recommendation; or human-AI recommendation.

The framing of the bogus intelligence as both strictly by way of an algorithm or an algorithm with human knowledgeable involvement was a key consider whether or not the decision-makers trusted AI below sure circumstances.

Participants have been proven 10 Airbnb listings for flats in Vienna—extracted from a dataset of roughly 12,000 house listings, on which the algorithm was educated—and given all related itemizing data apart from the worth. They have been tasked with estimating a per-night fee for all 10. In the no-AI-advice group, individuals needed to estimate a worth simply based mostly on the itemizing data they acquired.

But each AI teams carried out two estimations: the primary with simply the itemizing data, the second with the itemizing data and algorithmic recommendation. That decided “weight of recommendation”—how a lot affect a chunk of recommendation (on this case, algorithmic assist) has on an individual’s decision-making.

The researchers discovered that people who have been compensated based mostly on both efficiency or event incentives relied considerably extra closely on AI than those that obtain a hard and fast cost.

Also, those that used the algorithmic recommendation—no matter the way it was framed—carried out the duty of estimating the per-night fee considerably higher than those that didn’t get the AI help.

“In common,” Wiernsperger mentioned, “paying managers or decision-makers based mostly on their efficiency has constructive results, and never unfavorable results, with regards to the usage of AI.”

These outcomes have implications for firms attempting to introduce AI into decision-making, the researchers mentioned. If you need folks to make use of AI instruments, the way you inspire them and the way you speak in regards to the know-how each matter.