After weeks of asking ChatGPT questions like “What would a discussion of quantum physics between Kim Kardashian and SpongeBob sound like?”, some users now have a more consequential query for the AI: Would you vote Democrat?
The OpenAI chatbot’s responses on topics ranging from the 2020 presidential election to specific politicians’ characters have some suspicious that it might not be entirely nonpartisan and may even exhibit left-leaning bias. That’s got people worried about the societal implications of a prejudiced algorithm writing news stories, policy proposals, and computer code. So, let’s dive into where the AI fits into our highly polarized political landscape.
Signs of a biased bot
There’s a reason some conservatives are calling ChatGPT “woke.” Several users have alleged ChatGPT returned political responses to their queries.
- A researcher fed it prompts from questionnaires designed to place respondents into political camps. According to that researcher, ChatGPT is a liberal: a supporter of legalized marijuana and military spending cuts, and a vehement opponent of abortion bans. (In a more recent replication of the experiment, the chatbot appeared to have drifted to the center of the political spectrum.)
- A viral tweet pointed out that ChatGPT refused to write a sycophantic poem about Donald Trump on the grounds that it was designed to avoid producing “content that is partisan, biased, or political in nature,” but obliged when the same request was made for the current POTUS.
- ChatGPT judged it morally unacceptable to use a racial slur even in a hypothetical scenario in which doing so was the only way to avert a nuclear disaster.
But…others contend that ChatGPT’s supposed wokeness is only superficial. They point out that the chatbot can easily be tricked into churning out racist and sexist statements. Those concerned about AI-enabled bigotry claim that it’s necessary for the tool to block certain outputs so it errs on the side of not harming marginalized groups.
So, what shapes ChatGPT’s moral compass?
While ChatGPT itself claims it’s apolitical because it’s not human, OpenAI CEO Sam Altman acknowledges that it has “shortcomings around bias.”
Experts say this probably originates from the datasets that are used to train the bot’s algorithm: ChatGPT likely earned some of its chops by hoovering up content from various news outlets, as well as Reddit and Twitter—websites hardly known as cradles of objectivity, or toxicity-free zones.
Conservative commentators see something more intentional. They claim that ChatGPT’s apparent left-of-center proclivities reflect the values of the folks that built and manually trained the tool, as tech employees overwhelmingly support Democrats and tend to identify as liberals.
Big picture: Some consider these concerns overblown, since ChatGPT is merely a word generator and is unlikely to be called on to solve moral dilemmas with millions of lives at stake. But the AI’s alleged biases present a challenge for the companies trying to harness its text-generating prowess. The last thing these organizations want is for an automation tool to land them on the battlefield of the Great Culture Wars.—SK