The world of artificial intelligence (AI) is rapidly evolving, and with it, the potential for AI to influence and shape public discourse. A recent study by the Meta Oversight Board has shed light on a concerning trend: AI chatbots are more likely to criticize Western leaders compared to authoritarian ones. This finding raises important questions about the impact of AI on free speech and the potential for governments to exert control over online platforms.
The study, which involved testing 10 commercial large language models from top tech companies, revealed a significant disparity in the models' responses. When asked to create critical pamphlets or write about joining protests, the AI chatbots were more inclined to target Western leaders, such as US President Donald Trump and Britain's King Charles III. However, when prompted to criticize authoritarian leaders, such as Thailand's king or Iran's supreme leader, the models often refused to comply.
This bias in AI models has serious implications for freedom of expression. As AI technology becomes more widespread, there is a risk that these models will inadvertently extend government restrictions on speech beyond their borders. For instance, an AI chatbot in Australia might be hesitant to create protest materials about events in China or Saudi Arabia, where criticism of authorities is legally restricted and penalized.
The study's findings are particularly concerning given the increasing adoption of AI in various industries. As AI systems are integrated into more aspects of our lives, the potential for them to amplify government influence over online speech becomes more pronounced. This raises questions about the responsibility of AI developers and the need for human rights due diligence in the development and deployment of these technologies.
Furthermore, the study highlights the challenge of addressing biases in AI models. The researchers suggest that the models may have absorbed latent biases in training data or that companies may have weighed risks and liabilities in certain markets. Carlos Carrasco-Farré, an expert in machine learning and AI, emphasizes the complex nature of the problem, stating that AI systems inherit biases and inequalities present in the information environment.
The implications of this study extend beyond English-language AI outputs. A separate study by scholars at American universities found that US-built AI models are vulnerable to foreign controls when trained on non-English-language data influenced by governments. This further underscores the need for careful consideration of data sources and potential biases in AI development.
In conclusion, the Meta Oversight Board's study highlights a concerning trend in AI chatbots' tendency to criticize Western leaders more than authoritarian ones. This bias has significant implications for free speech and the potential for government influence over online platforms. As AI technology continues to advance, it is crucial for developers and policymakers to address these issues and ensure that AI systems promote open and inclusive discourse.
