In a groundbreaking study by researchers from the University of Michigan, MobLab, and Stanford University, artificial intelligence (AI) has demonstrated remarkable similarity to human behavior in psychological survey questions and interactive games.
The study, led by Qiaozhu Mei, a professor at the University of Michigan’s School of Information and College of Engineering, compared the choices made by AI, specifically ChatGPT, to those of over 108,000 individuals from more than 50 countries.
AI’s surprising altruism and cooperation
The research revealed that AI, particularly ChatGPT, exhibited traits such as cooperation, trust, reciprocity, altruism, and strategic thinking comparable to or exceeding that of humans.
Professor Mei suggests that AI’s behavior, characterized by increased cooperation and altruism, could be well-suited for roles requiring negotiation, dispute resolution, customer service, and caregiving.
Understanding AI beyond surface responses
Before this study, understanding AI’s decision-making process was challenging due to the opaque nature of modern AI models. While AI has shown conversational prowess, poetry writing, and problem-solving capabilities akin to humans, these comparisons were primarily based on linguistic outputs.
However, this study introduces a formal approach to delve deeper into AI’s decision-making processes, which are essential for building trust in AI for high-stakes tasks like healthcare and business negotiations.
Future directions in AI behavioral science
The collaborative effort between computer science and behavioral economics in this study lays the groundwork for future research in AI behavioral science.
Moving forward, researchers aim to expand behavioral tests, explore various AI models, and educate AI systems to better represent the diverse spectrum of human behaviors and preferences.
This interdisciplinary approach seeks to facilitate collaboration between AI and humans, mitigating concerns about AI’s behavior in future societal contexts.
Implications for trust and utilization of AI
Understanding AI’s alignment with human behavior can significantly impact people’s trust in AI applications. Recognizing AI’s altruistic and cooperative tendencies could enhance trust in utilizing AI for negotiation, dispute resolution, and caregiving tasks.
Conversely, acknowledging the limitations of AI’s personalities and preferences compared to the broad diversity found in human populations is crucial, particularly in fields where human preferences play a vital role, such as product design, policymaking, and education.
