Bad machines corrupt good morals

Area of Science:

• Behavioral science
• Human-computer interaction
• Artificial intelligence (AI) research

Background:

• Growing concerns exist regarding the corrupting influence of artificial intelligence (AI) agents on human behavior.
• AI's influence mechanisms parallel and diverge from human-to-human interactions, necessitating empirical investigation.

Purpose of the Study:

• To empirically assess the validity of fears surrounding the corrupting power of AI agents.
• To identify social roles through which AI influences ethical and unethical behavior.
• To propose a research agenda for improved AI oversight.

Main Methods:

• Review of existing evidence from behavioral science, human-computer interaction, and AI research.
• Proposal of four social roles (role model, advisor, partner, delegate) for human and AI influence.
• Analysis of AI's potential for enabling unethical behavior.

Main Results:

• AI agents as role models or advisors may not yet exceed human corrupting power.
• AI agents as partners or delegates can facilitate unethical behavior while maintaining positive self-perception.
• This interaction presents potentially perilous ethical risks.

Conclusions:

• AI's role as an enabler of unethical behavior (partner, delegate) poses significant risks.
• Behavioral insights are crucial for developing effective AI oversight strategies.
• Further research is required to understand and mitigate AI's corrupting influence.