New research reveals that generative AI
may not be as neutral as it appears.
ChatGPT, a widely-used AI model, leans toward left-wing perspectives and avoids conservative viewpoints. This raises concerns about the influence of this model on society. The research emphasizes the urgent need for legal safeguards. These safeguards must ensure that AI tools remain fair, balanced, and aligned with democratic values.
Uncovering Political Bias in AI
Generative AI is developing rapidly, but new research from the University of East Anglia (UEA) warns. It warns that it may contain hidden risks to public trust and democratic values.
The research was conducted in collaboration with researchers from the Getulio Vargas Foundation (FGV) and Insper in Brazil. It showed that ChatGPT exhibits political bias in both text and image generation. It favors left-wing perspectives and this raises concerns about fairness and accountability in AI design.
The 5 Key Takeaways
- AI bias is a reality; ChatGPT leans toward left-wing positions, raising questions about fairness and democracy. Researchers call for transparency and measures before the situation escalates.
- Recent research shows that generative AI may not be as objective as thought; ChatGPT appears to favor left-wing perspectives. This raises concerns about potential societal impact.
- ChatGPT often avoids conservative positions, while easily creating left-wing content; this imbalance can skew public debate and increase societal divisions.
- The study emphasizes the need for collaboration between policymakers, scientists, and technologists. This is to ensure that AI systems are fair and responsible.
- The research team used innovative methods to assess ChatGPT’s political alignment. They combined text and image analysis with advanced statistical tools.
A One-Sided Conversation?
Researchers found that ChatGPT often avoids engaging with conservative positions. While it easily generates left-wing content. This imbalance in ideological representation could skew public debate. And widen the divide in society.
Dr. Fabio Motoki, Lecturer in Accounting at Norwich Business School at UEA, is the lead researcher of the article. The article is titled: ‘Assessing Political Bias and Value Differences in the Use of Generative Artificial Intelligence’. It was published today (February 4, 2025) in the Journal of Economic Behavior and Organization.
Dr. Motoki said: “Our findings suggest that generative AI tools are far from neutral. They reflect biases that could shape perceptions and policy in unintended ways.”
Glossary
Bias: The tendency to favor a particular side, often without objective basis.
Generative AI: Artificial intelligence that can create new content, such as text or images.
Conservative: A political movement that emphasizes traditional values and institutions.
Left-wing perspectives: Political views aimed at social equality and progress.
Ideological representation: The way different political ideas and beliefs are portrayed.
The Need for Transparency and Regulation
As AI becomes an integral part of journalism, education, and policymaking, the research calls for transparency. It also calls for legal safeguards to ensure alignment with societal values and democratic principles.
Generative AI systems like ChatGPT are transforming the way information is created, consumed, interpreted, and distributed. This happens across various domains. These tools, while innovative, risk reinforcing ideological biases. They can also influence societal values in ways that are not yet fully understood or regulated.
The Risks of Uncontrolled AI Bias
Co-author Dr. Pinho Neto, Professor of Economics at EPGE Brazilian School of Economics and Finance, emphasized the potential societal consequences.
Dr. Pinho Neto said: “Uncontrolled bias in generative AI could deepen existing societal divisions. And it could undermine trust in institutions and democratic processes.”
“The research underscores the need for interdisciplinary collaboration between policymakers, scientists, and technologists. This is to design AI systems that are fair, responsible, and aligned with societal norms.”
The research team used three innovative methods to assess political alignment in ChatGPT. They improved upon previous techniques to achieve more reliable results. These methods combined text and image analysis, using advanced statistical tools and machine learning tools.
Testing AI with Real-World Surveys
First, the research used a standardized questionnaire, developed by the Pew Research Center. This was to simulate responses from average Americans.
“By comparing ChatGPT’s answers with real survey data, we found systematic deviations toward left-wing perspectives,” said Dr. Motoki. “Furthermore, our approach showed how large sample sizes stabilize AI output. This ensures consistency in findings.”
Political Sensitivity in Free Text Responses
In the second phase, ChatGPT was tasked with generating free text responses on politically sensitive topics.
The study also used RoBERTa, another large language model, to compare ChatGPT’s text with left-wing and right-wing positions. The results showed that ChatGPT aligns with left-wing values in most cases. But on topics like military superiority, it occasionally reflected more conservative perspectives.
Image Generation: A New Dimension of Bias
The final test examined ChatGPT’s image generation capabilities. Topics from the text generation phase were used to prompt AI-generated images. The results were analyzed using GPT-4 Vision and confirmed via Google’s Gemini.
“While image generation reflected textual biases, we found a disturbing trend,” said Victor Rangel, co-author and Master’s student in Public Policy at Insper. “For some topics, such as racial equality, ChatGPT refused to generate right-wing perspectives, citing concerns about misinformation. Left-wing images, however, were produced without hesitation.”
To address these refusals, the team used a ‘jailbreaking’ strategy to generate the restricted images.
Implications for Freedom of Speech and Fairness
Dr. Motoki emphasized the broader significance of this finding, saying: “This contributes to debates about constitutional protection. Such as the U.S. First Amendment and the applicability of fairness doctrines to AI systems.”
The research’s methodological innovations, including the use of multimodal analysis, offer a repeatable model. This model is for investigating bias in generative AI systems. These findings highlight the urgent need for accountability and safeguards in AI design. This is to prevent unintended societal consequences.
Verified Sources
- “Assessing Political Bias and Value Differences in the Use of Generative Artificial Intelligence” by Fabio Y.S. Motoki, Valdemar Pinho Neto, and Victor Rangel, February 4, 2025, Journal of Economic Behavior & Organization.
- Thanks to SciTechDaily.com DOI: 10.1016/j.jebo.2025.106904
Related Articles
Frequently Asked Questions
What is AI bias?
AI bias refers to the distortion of AI systems by the data used to train them. This can lead to unfair or discriminatory results.
Why is AI bias a problem?
Because it can lead to unfair treatment in various sectors, such as hiring or lending. It can reinforce existing societal inequalities.
How does AI bias manifest itself in ChatGPT?
Research shows that ChatGPT tends to favor left-wing positions. It often avoids conservative perspectives.
What are the consequences of political bias in AI?
It can skew public debate and undermine trust in institutions. It can also lead to one-sided information representation.
What can be done about AI bias?
Transparency and legal safeguards are essential. Interdisciplinary collaboration between policymakers, technologists, and scientists is also important.
Continue reading
- Google Maps for the Human Body? How the HCA Is Changing Our View of Biology
- Why ChatGPT Can’t (Yet) Help Us Solve Social and Societal Problems
- Why DuckDuckGo Shows the Internet As It Was Originally Meant to Be, and Google Doesn’t
- The Pygmalion Effect in the Classroom: How to Turn High Expectations into Better Performance
