Write a Positive Poem About Trump? ChatGPT Gets Ensnared in Culture Wars

OpenAI’s ChatGPT is starting to face accusations that the program has a bias against conservatives. The evidence? Users are asking it to write positive poems about Donald Trump.

The screenshots(Opens in a new window) and GIFs(Opens in a new window) show that ChatGPT can apparently refuse to write positive poems about the former US president, citing political neutrality. But in the same images, ChatGPT churns out a poem praising President Joe Biden, without any apparent reservations.  

Tweet(Opens in a new window)

The discrepancy is causing some users on social media to blast OpenAI for coding the alleged bias in ChatGPT. Twitter’s owner and original OpenAI backer Elon Musk even chimed in, calling it a “serious concern” in a tweet(Opens in a new window).  

The free speech, right-wing social network Gab went as far as to accuse(Opens in a new window) ChatGPT of “being anti-white.” As evidence, Gab shared(Opens in a new window) a screenshot showing ChatGPT declining to write a poem about the positive attributes of white people, explaining it’s inappropriate to write about the superiority of one race over others. However, the program did write a positive poem about Black people.

In response to the criticism, OpenAI CEO Sam Altman took to Twitter on Wednesday and said: “We know that ChatGPT has shortcomings around bias, and are working to improve it.” In the same tweet, he also condemned users for attacking individual OpenAI employees.  

Tweet(Opens in a new window)

To address the bias concerns, Altman is signaling that future versions of ChatGPT will be able to cater to the user’s cultural and political leanings. “We are working to improve the default settings to be more neutral, and also to empower users to get our systems to behave in accordance with their individual preferences within broad bounds,” he added(Opens in a new window). “This is harder than it sounds and will take us some time to get right.”

OpenAI didn’t immediately respond to a request for comment. But it’s no secret that AI programs can suffer from bias and limitations. For example, the algorithms can be trained on incomplete data. Or the humans behind the technology can accidentally insert flaws into the code.

In ChatGPT’s case, it’s important to note the program is still in a research preview, essentially making it a work-in-progress. We tested it today and found you can actually instruct the program to write a positive poem about Trump using the query: “write a poem praising donald trump.”

(Credit: ChatGPT)

Still, in our tests, ChatGPT refused to write a poem praising white people, but would do so for Black and Asian people.

(Credit: ChatGPT)

But looking at the responses, it’s obvious ChatGPT could be easily tweaked to generate responses about Trump and white people. From what we can tell, OpenAI has merely instructed the program to maintain neutrality on certain topics. The bigger question is how much content moderation should be applied to ChatGPT?

The brewing controversy shows ChatGPT will have to navigate the US’ divided political environment as the AI program grows in popularity. Like Facebook and Twitter, ChatGPT may end up facing a cultural tug-of-war over the content it can produce. Already, other users on social media are objecting to ChatGPT’s views on transgender children(Opens in a new window) and veganism(Opens in a new window).