Recent incidents involving AI chatbots, such as Google's Gemini and Adobe Firefly's image creation tool, have sparked discussions on the intersection of AI bias, political correctness, and free speech. While debates over AI's political inclinations and efforts to combat bias are crucial, the focus often neglects a fundamental issue: the AI industry's stance on free speech and its adherence to international standards.
Policy researchers at The Future of Free Speech, based at Vanderbilt University, highlight critical shortcomings in the approach of generative AI, which creates content based on training data. They argue that major AI chatbot companies' use policies fail to meet United Nations standards on free speech. These policies, particularly concerning hate speech and misinformation, are vague and overly broad, leading to censorship of content deemed controversial by the companies.
Testing various prompts on contentious topics, researchers found that AI chatbots refused to generate content for 40% of inquiries, citing policies against hate speech and misinformation. However, such policies risk stifling legitimate discourse and diverse perspectives, undermining the principles of free speech and access to information.
While AI providers have the right to enforce restrictive policies, their significant market influence necessitates a balanced approach that respects free speech. The European Union's Digital Services Act, though imperfectly applied, underscores the need for tech giants to mitigate systemic risks, including threats to free expression.
The researchers advocate for AI companies to adopt a free speech culture guided by international human rights principles. Rather than outright refusal to generate content, they suggest alternative measures like providing context or countervailing facts. Such approaches not only uphold free speech but also mitigate the risk of driving users towards platforms that specialize in promoting hateful content and echo chambers.
More: https://techxplore.com/news/2024-04-ai-chatbots-controversial-output-free.html
