[Paper Review] Corrupted by Algorithms? How AI-generated and Human-written Advice Shape (Dis)honesty

Margarita Leib, Nils Köbis+3 more|arXiv (Cornell University)|Jan 5, 2023

Experimental Behavioral Economics Studies12 citations

TL;DR

The study examines how AI-generated versus human-written advice influences dishonest and honest behavior, finding that dishonesty-promoting advice increases dishonesty for both sources, while transparency about advice source has no effect.

ABSTRACT

Artificial Intelligence (AI) increasingly becomes an indispensable advisor. New ethical concerns arise if AI persuades people to behave dishonestly. In an experiment, we study how AI advice (generated by a Natural-Language-Processing algorithm) affects (dis)honesty, compare it to equivalent human advice, and test whether transparency about advice source matters. We find that dishonesty-promoting advice increases dishonesty, whereas honesty-promoting advice does not increase honesty. This is the case for both AI- and human advice. Algorithmic transparency, a commonly proposed policy to mitigate AI risks, does not affect behaviour. The findings mark the first steps towards managing AI advice responsibly.

Motivation & Objective

• Motivate understanding of ethical and behavioral risks of AI as an advice agent.
• Compare the behavioral impact of AI-generated advice to equivalent human advice.
• Test whether transparency about the advice source mitigates dishonest or dishonest shifts.

Proposed method

• Conduct an experiment exposing participants to AI-generated or human-written advice.
• Manipulate the type of advice to promote dishonesty or honesty.
• Include a transparency condition about the advice source and measure subsequent behavior.
• Compare effects across AI and human advice to identify source-specific patterns.

Experimental results

Research questions

• RQ1 Does AI-generated advice promote dishonesty as much as human-written dishonesty-promoting advice?
• RQ2 Does honesty-promoting advice increase honest behavior, regardless of the advice source?
• RQ3 Does transparency about whether advice is AI-generated or human-written affect behavior?
• RQ4 Are there differential effects of source (AI vs. human) on responses to dishonesty- vs. honesty-promoting advice?

Key findings

• Dishonesty-promoting advice increases dishonest behavior.
• Honesty-promoting advice does not increase honesty.
• Transparency about the advice source does not affect behavior.
• Effects are observed for both AI-generated and human-written advice.