New research has found that popular AI chatbots are often susceptible to several of the same decision-making mistakes as humans.
In recent research involving OpenAI’s ChatGPT, the popular chatbot exhibited mistakes humans are prone to make in certain circumstances, including overconfidence displays or falling for the “gambler’s fallacy,” a phenomenon where one interprets infrequent occurrences to be more likely to happen in the future.
However, in other situations, AI models aren’t as likely to make the same errors as humans, which include people’s tendencies to ignore important facts or stick with bad choices once they already feel invested.
The findings, detailed in a new study, show that ChatGPT doesn’t just process data—it “thinks” like humans, including relying on mental shortcuts and other behaviors that can result in blind spots. These biases were found to remain consistent across various business situations, but researchers believe they could change as AI improves with each new version.
“As AI learns from human data, it may also think like a human – biases and all,” says Yang Chen, lead author and assistant professor at Western University. “Our research shows when AI is used to make judgment calls, it sometimes employs the same mental shortcuts as people.”
Revealing the Human-Like Biases of AI Chatbots
According to Chen and the team, the results of putting ChatGPT through 18 different bias tests revealed that AI can fall into the same decision-making traps as humans. ChatGPT displayed biases such as overconfidence, ambiguity aversion, and the “conjunction fallacy” (a cognitive bias that causes one to mistakenly believe that two events happening together is more likely to occur than just one event happening alone) in nearly half of the tests.
While the AI excels at math and logical, probability-based problems, it struggles with human-based judgment calls rooted in subjective reasoning. Even with the newer GPT-4 model, which is said to be more analytically accurate, biases remain and, at times, appear to be even stronger now when dealing with some human judgment-based tasks.
Additionally, the study found that ChatGPT likes to play it safe by avoiding risk, overestimating, seeking confirmation of its assumptions, and avoiding vagueness by favoring detailed information.
“When a decision has a clear right answer, AI nails it – it is better at finding the right formula than most people are,” says Anton Ovchinnikov of Queen’s University. “But when judgment is involved, AI may fall into the same cognitive traps as people.”
The researchers say their findings indicate that we should be cautious about relying on AI for important decision-making, as it can exhibit biases very similar to those of humans.
“AI should be treated like an employee who makes important decisions – it needs oversight and ethical guidelines,” says Meena Andiappan of McMaster University. “Otherwise, we risk automating flawed thinking instead of improving it.”
“AI isn’t a neutral referee,” says Samuel Kirshner of UNSW Business School. “If left unchecked, it might not fix decision-making problems – it could actually make them worse.”
The research recommends regular audits with programming and reviews aimed to help reduce the biases chatbots currently display.
“The evolution from GPT-3.5 to 4.0 suggests the latest models are becoming more human in some areas, yet less human but more accurate in others,” says Tracy Jenkin of Queen’s University. “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.”
However, this leaves the question of developers’ views on the role of ethical guidelines and oversight in AI development, especially in the context of maintaining accountability.
“Ethical guidelines and oversight mechanisms are the backbone of responsible AI governance. Those audits must be embedded within a risk-based, equity-centered framework,” Daryl Lim, the H. Laddie Montague Jr. Chair in Law at Penn State Dickinson Law, told The Debrief in an email.
“As my article explores, a key turning point was the 2024 Framework Convention on Artificial Intelligence, which introduced binding commitments for transparency, accountability, and human rights protections in AI governance.”
“Ethical oversight isn’t just about technical robustness—it’s about legitimacy and trust. I argued in my article that AI is not just code; it’s a norm-generating system. It shapes behavior and decisions, particularly in contexts like courts or health systems, where power asymmetries are steep,” he said.
Fundamentally, Lim says that ethical guidlines should be used to intigrate legal concepts like due process, equal protection, and the overall promotion of transparency, “as well as include independent regulatory bodies with teeth and public disclosure of model behavior in high-impact domains.”
The study, “A Manager and an AI Walk into a Bar: Does ChatGPT Make Biased Decisions Like We Do?” was first published in INFORMS.
Chrissy Newton is a PR professional and founder of VOCAB Communications. She currently appears on The Discovery Channel and Max and hosts the Rebelliously Curious podcast, which can be found on The Debrief’s YouTube Channel on all audio podcast streaming platforms. Follow her on X: @ChrissyNewton and at chrissynewton.com.