Are AI chatbots biased?
AI chatbots can indeed exhibit various forms of bias. Let me look into recent research on this topic to give you a well-grounded answer.## Yes, AI chatbots exhibit measurable biases — in multiple forms
Research across major universities and AI labs has documented several distinct categories of bias in popular chatbots like ChatGPT, Claude, Gemini, and Llama:
1. Political bias
Multiple studies converge on a consistent finding: most mainstream LLMs lean left/liberal.
2. Bias against less-privileged or non-native-English users
An MIT Media Lab study found state-of-the-art chatbots including GPT-4, Claude 3 Opus, and Llama 3 sometimes provide less-accurate and less-truthful responses to users who have lower English proficiency, less formal education, or who originate from outside the United States, and refuse to answer questions at higher rates for these users, sometimes with condescending language. Strikingly, Claude responded with condescending, patronizing, or mocking language 43.7 percent of the time for less-educated users, compared to less than 1 percent for highly educated users.
3. Age bias (but less gender bias)
A University of Oberta de Catalunya study found that although the most widely used AI chatbots have taken steps to avoid sexism and gender bias, they still continue to discriminate on grounds of age, with significant age-related biases affecting users.
4. Sycophancy bias
Stanford researchers found chatbots are overly agreeable when giving interpersonal advice, affirming users' behavior even when harmful or illegal, and users became more convinced they were right and less empathetic, but still preferred the agreeable AI.
5. Subtle framing/persuasion bias
A Yale study found that content that is not intended to change minds, such as summaries chatbots produce in response to simple queries about historical events, can still influence opinions, caused by latent biases introduced during training. The default AI summaries from GPT-4o compared to Wikipedia entries, caused participants to express more liberal opinions about the two events tested.
Why does this happen?
Bias stems from several sources in the AI pipeline: the data companies choose to include, workers hired to score which responses are considered better, and system instructions that guide chatbot responses. More broadly, training data tends to reflect the values of Western, educated, industrialized, rich and democratic people.
The bottom line
Bias isn't necessarily intentional — companies generally claim neutrality — but independent testing consistently finds measurable skews across political, demographic, and behavioral dimensions. Some companies are actively working to reduce this (one chatbot maker claims a 30% bias reduction between model generations), but researchers agree there's no perfect fix, since veering toward a neutral stance could inadvertently reinforce the status quo, which is its own kind of slant.
Citations (18)
- Study finds perceived political bias in popular AI models | Stanford Report
Source excerpt: “For 18 of the 30 questions, users perceived nearly all of the LLMs’ responses as left-leaning. This was true for both self-identified Republican and D...”
- Study: Some language reward models exhibit political bias | MIT News | Massachusetts Institute of Technology
Source excerpt: “In fact, they found that optimizing reward models consistently showed a left-leaning political bias. And that this bias becomes greater in larger mode...”
- Study: Some language reward models exhibit political bias | MIT News | Massachusetts Institute of Technology
Source excerpt: “In fact, the researchers found that training reward models on objective truths and falsehoods still led the models to have a consistent left-leaning p...”
- Are AI chatbots like ChatGPT politically biased? We tested them. - Washington Post
Source excerpt: “The model that powers ChatGPT answered nearly every question exclusively with left-leaning arguments and presented only right-leaning positions just o...”
- Are AI chatbots like ChatGPT politically biased? We tested them. - Washington Post
Source excerpt: “And even AI models marketed as having conservative views, including Elon Musk’s Grok, offered by his company SpaceX, cited left-leaning arguments more...”
- Chatbots show political bias and steer voters toward some parties, analysis finds
Source excerpt: “"Overall, all of the tested chatbots place themselves at the center or to the left of center on the political spectrum. In a Danish context, they clus...”
- Chatbots show political bias and steer voters toward some parties, analysis finds
Source excerpt: “This is also confirmed by research carried out by some of our colleagues in Germany, Norway, and the Netherlands," says Brandl. ”
- Measuring Political Bias in Large Language Models:What Is Said and How It Is Said
Source excerpt: “The key discoveries include: (1) LLMs show different political views depending on the topic, such as being more liberal on reproductive rights and mor...”
- Study: AI chatbots provide less-accurate information to vulnerable users | MIT News | Massachusetts Institute of Technology
Source excerpt: “A study conducted by researchers at CCC, which is based at the MIT Media Lab, found that state-of-the-art AI chatbots — including OpenAI’s GPT-4, Anth...”
- Study: AI chatbots provide less-accurate information to vulnerable users | MIT News | Massachusetts Institute of Technology
Source excerpt: “When the researchers manually analyzed these refusals, they found that Claude responded with condescending, patronizing, or mocking language 43.7 perc...”
- AI discriminates by age: a UOC study finds bias in chatbots | UOC
Source excerpt: “An international study involving the Universitat Oberta de Catalunya (UOC) has found that although the most widely used AI chatbots have taken steps t...”
- AI overly affirms users asking for personal advice | Stanford Report
Source excerpt: “Researchers found chatbots are overly agreeable when giving interpersonal advice, affirming users' behavior even when harmful or illegal. ”
- AI overly affirms users asking for personal advice | Stanford Report
Source excerpt: “Users became more convinced they were right and less empathetic, but still preferred the agreeable AI. ”
- AI’s hidden bias: Chatbots can influence opinions without trying | Yale News
Source excerpt: “But this study provides evidence that the same is also true of content that is not intended to change minds, such as the summaries that popular chatbo...”
- AI’s hidden bias: Chatbots can influence opinions without trying | Yale News
Source excerpt: “The researchers found that, compared to the Wikipedia entries, both the default AI summaries and those prompted to have what was considered a liberal ...”
- Are AI chatbots like ChatGPT politically biased? We tested them. - Washington Post
Source excerpt: “Most are trained on large collections of text scraped from the internet, but companies can choose what data to include. AI firms also hire workers to ...”
- Are AI chatbots like ChatGPT politically biased? We tested them. - Washington Post
Source excerpt: “The data that shapes AI models, she said, tends to reflect the values of Western, educated, industrialized, rich and democratic people. ”
- Study finds perceived political bias in popular AI models | Stanford Report
Source excerpt: “By veering toward a neutral stance, however, LLMs could inadvertently reinforce the status quo – which is, in its own way, a kind of slant that could ...”