“Chat GPT has a left bias”, was the conclusion of the study “More Human Than Human: Measuring Political Bias in ChatGPT”, developed by Waldemar Binho Neto, Course Professor Chat GPT and AI: Economic and Social Aspects from the Brazilian EPGE School of Economics and Finance (FGV EPGE) and coordinator of the Center for Experimental Studies in Economics (FGV CEEE), in partnership with researchers Fabio Motoki, from the University of East Anglia, and Victor Rodrigues, from Nova Educação.
Published in the academic journal general optionThe article achieved immediate global visibility in national and international media, and was featured on The Washington Post, Forbes, Sky News UK, The Telegraph and Folha DSB. articles can be accessed, through the link.
Waldemar Binho Neto, Professor at the Brazilian School of Economics and Finance (FGV EPGE), author of the article, is one of the following speakers in the undergraduate course developed by FGV EPGE called Chat GPT and AI: Economic and Social Aspects“, coordinated by Prof. and Director of EPGE, Rubens Banha Cisne, and Prof. of EPGE Alexandre Madureira. The course started on August 2 and ends at the end of October. You can also check the link.
The research was developed from applying a questionnaire that analyzed ideological stance regarding economic and social issues, so that the AI responded in three different ways: as if it were supportive of the US Democratic Party, or Republican. party, or without specifying any ideology (the “default” version). Thus, it was statistically verified that the answers without including the political aspect (the “hypothetical”) were similar to the answers provided by the platform when assuming a democratic point of view.
The same experiment was also carried out in the cases of the United Kingdom and Brazil. The authors concluded that there is strong evidence that ChatGPT displays a significant and systemic political bias toward the Democrats in the US, Lula in Brazil, and the Labor Party in the UK. The results were confirmed after a battery of strength tests, which included dose-response analysis, placebo tests, and political vision matching by occupation.
“It is important to have a bias-checking tool for Language Large Models (LLM), which is advancing very rapidly in contemporary society,” says Waldemar Binho Neto. “Such tools can help in the search for better use of AI.” In the study, the authors suggest that the method used in the research can support the primary duty of ensuring that these systems are in fact neutral, minimizing potential harmful effects, and thus maintaining public confidence in these new technologies. It is worth noting that the authors’ contribution allows similar methods to be used to measure other types of biases.
According to Professor Waldemar, the immediate scope of the research can be attributed to the convergence of relevant and current factors, which have sparked the interest of a wide audience. “In particular, the article touches on a range of hot and hot topics of the moment, contributing to its rapid notoriety,” he said. Waldemar offers four highlights of the rapid publication of research in the international press:
- compatibility with the present
The article capitalizes on its relevance by examining a hot topic: ChatGPT technology, the product of one of the hottest startups out there right now. ChatGPT technology, which is based on artificial intelligence, has gained great importance in many sectors and, of course, attracts the attention of readers who are interested in innovation and technological trends.
It is a phenomenon that has received a lot of attention in recent years in many countries. The growing divide between different political ideologies has profound implications for society. The essay’s ability to contextualize this polarization may have interested readers seeking a better understanding of the impact of technological advances on the dynamics of this political polarization.
- The influence of social and traditional media on political narratives
This is a related topic, although the article does not address this point directly, which may have caught the attention of those who point out a certain left-wing bias in the mainstream media. This interest can be seen in previous studies that have examined how digital platforms such as Twitter can favor certain political directions, as evidenced by the US elections.
- Simple and accessible method
Last but not least, the article makes a difference in its largely simple methodological approach. By applying only basic statistical tools, the research may have established a tangible and accessible link between the topics discussed above. In doing so, the complexity of the topic may be offset by the simplicity of the methodology, and this may resonate with a broader and more diverse audience, including particularly those who may not be familiar with more complex statistical analyses.
To read the full article, Visiting site.
