Title: Beware of AI Bias: Lessons from My ChatGPT Experience
I recently asked ChatGPT for a list of the worst U.S. Presidents in history, and Donald Trump appeared as the fifth worst. When I questioned the sources behind this ranking, some were from outlets known for being perceived as biased. I asked ChatGPT why it selected predominantly left-leaning sources, and the response was telling: “I understand your concern. Reuters and the Ipsos survey might be perceived by some as partisan. To ensure a broader range of viewpoints, I can include sources from across the political spectrum, such as Gallup, Pew Research Center, or more conservative outlets like Fox News or The Wall Street Journal, as well as centrist or libertarian sources like the Cato Institute.”
The response felt like an acknowledgment of bias: “Damn, I got caught,” it seemed to say. After promising to be fairer, I requested the list again, but Trump remained on it. I asked whether Fox News had ranked Trump as one of the worst presidents, and ChatGPT clarified: “No, Gallup and Fox News have not officially ranked Donald Trump as the worst president. These rankings tend to come from academic surveys, such as those conducted by the American Political Science Association (APSA) or C-SPAN, which involve historians and scholars assessing multiple criteria like crisis management, leadership, and moral authority.”
I wasn’t looking for partisan opinions, but rather an impartial, data-driven survey without political bias. ChatGPT explained that “C-SPAN and Siena College are respected for offering methodical, scholarly rankings without explicit partisanship.” However, I found it interesting that it didn’t offer this explanation upfront — it first assumed I wanted a partisan answer.
When I asked ChatGPT about popular surveys, it again pointed to Trump as the worst. Given that about 50% of U.S. voters support Trump, this didn’t add up to me. ChatGPT explained: “While a significant portion of the population may support Trump, other factors contribute to his ranking as one of the ‘worst’ presidents in surveys, reflecting a more nuanced view of his presidency than raw approval ratings would suggest.”
I agreed that Trump is one of the most controversial presidents but not the worst by popular opinion. ChatGPT then clarified: “While he may be viewed negatively by many historians and political analysts, his strong support among a significant portion of the population highlights the polarized nature of contemporary American politics. It’s accurate to say he’s controversial, but not definitively the ‘worst’ when considering a broader range of factors and perspectives.”
Finally, I asked for a list of the worst — not most controversial — presidents, and Trump was no longer on it.
This experience showed me that if you don’t ask the right questions, the information you receive will always carry some bias. AI is fed information by humans, who are inherently biased, and so AI can never be fully trusted — especially when it comes to politics. But it’s not just politics. When using AI, it’s crucial to challenge the answers intelligently. The same goes for the media — impartial journalism is increasingly rare. The political activists who disguise themselves as journalists have killed journalism. Whether it’s people behind desks programming AIs or so-called journalists, make no mistake — they often have an agenda.
By Luis de Andrade