Can we trust AI to make better decisions than us?
OpenAI’s ChatGPT gets billions of monthly visits from people curious about its transformative influence. But does the world’s most popular AI chatbot think more like us than we are aware? Can it be just as biased? A study led by a team of Canadian researchers set out to answer this question by involving ChatGPT in several tests. Published in the journal ‘Manufacturing & Service Operations Management’(opens in new window), the findings showed that it made mistakes that humans also make in certain situations. The AI chatbot doesn’t just process massive amounts of data, it ‘thinks’ like we do.
Human-like bias
The researchers put ChatGPT through 18 different bias tests. While excelling in logic and maths, it exhibited many of the same biases as humans when making subjective decisions. The biases include overconfidence, risk aversion, and the gambler’s fallacy – the false belief that if an event has occurred several times before in the past, it will occur less often in the future. These biases remained consistent across different business scenarios, but the research team believes they could change as AI keeps improving with new versions. “As AI learns from human data, it may also think like a human – biases and all,” commented lead author Yang Chen, assistant professor at Western University, in a news release(opens in new window). “Our research shows when AI is used to make judgment calls, it sometimes employs the same mental shortcuts as people.”
Falling into the same decision-making traps
The researchers also found that ChatGPT liked to play it safe, overestimate itself, seek confirmation and avoid ambiguity. “When a decision has a clear right answer, AI nails it – it is better at finding the right formula than most people are,” stated Anton Ovchinnikov of Queen’s University. “But when judgment is involved, AI may fall into the same cognitive traps as people.” “AI isn’t a neutral referee,” added Samuel Kirshner of the University of New South Wales Business School in Australia. “If left unchecked, it might not fix decision-making problems – it could actually make them worse.” Because of all the biases, the study authors assert that businesses and policymakers should be cautious about counting on AI for key decision-making. It may not eliminate human error, but automate it instead. “AI should be treated like an employee who makes important decisions – it needs oversight and ethical guidelines,” elaborated Meena Andiappan of McMaster University. “Otherwise, we risk automating flawed thinking instead of improving it.” The researchers recommend systematic audits with programming and reviews so as to help cut down on the biases that today’s chatbots exhibit. “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,” explained 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.”