A recent study unveils that OpenAI’s ChatGPT, despite its advanced capabilities, mirrors human decision-making biases in nearly half the tests conducted, raising questions about AI’s role in critical business decisions.
As artificial intelligence continues to play an increasingly significant role in decision-making across various sectors, a new study reveals that AI may not always be making better or more rational choices than humans. Researchers found that OpenAI’s ChatGPT, a leading AI model, exhibits the same decision-making biases as humans in certain situations.
Published in the INFORMS journal Manufacturing & Service Operations Management, the study reveals that ChatGPT doesn’t just crunch numbers — it “thinks” much like humans, including inheriting some of their decision-making flaws.
The study subjected ChatGPT to 18 different bias tests. The researchers discovered that ChatGPT showcased biases such as overconfidence, ambiguity aversion and the conjunction fallacy, commonly known as the “Linda problem,” in nearly half of the tests conducted.
“As AI learns from human data, it may also think like a human — biases and all. Our research shows when AI is used to make judgment calls, it sometimes employs the same mental shortcuts as people,” lead author Yang Chen, an assistant professor at Western University, said in a news release.
Understanding AI’s Cognitive Shortcomings
Further findings indicate that while the newer GPT-4 model surpasses its predecessor GPT-3.5 in analytical accuracy, it sometimes demonstrates stronger biases during judgment-based tasks.
This revelation raises significant concerns considering AI’s growing role in critical decision-making areas like job hiring and loan approvals.
“When a decision has a clear right answer, AI nails it — it is better at finding the right formula than most people are,” added co-author Anton Ovchinnikov of Queen’s University. “But when judgment is involved, AI may fall into the same cognitive traps as people.”
Implications for Business and Policy
The discovery that AI models like ChatGPT are susceptible to human-like biases calls for enhanced scrutiny over their decision-making processes.
“AI isn’t a neutral referee,” added co-author Samuel Kirshner of University of New South Wales Business School. “If left unchecked, it might not fix decision-making problems — it could actually make them worse.”
The study underscores the necessity for businesses and policymakers to conduct regular audits of AI-driven decisions.
“AI should be treated like an employee who makes important decisions — it needs oversight and ethical guidelines,” added co-author Meena Andiappan of McMaster University. “Otherwise, we risk automating flawed thinking instead of improving it.”
The Path Forward
To mitigate these biases, the study’s authors recommend continuous refinement of AI systems and regular evaluations of their performance.
“Managers must evaluate how different models perform on their decision-making use cases and regularly re-evaluate to avoid surprises. Some use cases will need significant model refinement,” added co-author Tracy Jenkin of Queen’s University.
With AI’s influence only expected to grow, ensuring it aids in making better decisions rather than merely replicating human biases will be critical. The journey from GPT-3.5 to GPT-4.0 demonstrates a combination of increasing human-likeness in some areas and improved analytical capabilities in others.
Source: INFORMS