Are AI models ‘woke’? The answer isn’t so simple

In a bold move, former President Donald Trump has signed an executive order that seeks to position the United States as a frontrunner in artificial intelligence. This directive aims to eliminate what Trump labels as 'woke' ideologies from AI technologies utilized by the federal government. The order, announced on Wednesday, prohibits the acquisition of AI systems deemed to carry partisan biases or ideological agendas, specifically targeting concepts like critical race theory. This initiative is part of a broader White House strategy to advance the U.S. in the realm of AI. The executive order mandates that AI models, particularly large language models that power popular chatbots like ChatGPT, adhere strictly to principles of 'truth-seeking' and 'ideological neutrality.' Trump emphasized that from this point forward, the government will engage only with AI that embodies fairness and impartiality. Experts are weighing in on the complexities of AI bias. They note that AI models reflect the data they are trained on, the feedback received during training, and the specific instructions given to them. This raises the question: Can AI genuinely possess ideological bias, or is it merely a reflection of the inputs it processes? Oren Etzioni, former CEO of the Allen Institute for Artificial Intelligence, explained that while AI lacks personal beliefs, it can exhibit biases based on the data it processes. The executive order outlines two key principles: 'truth-seeking,' which mandates that AI strive for accuracy and factual analysis, and 'ideological neutrality,' which requires that models remain nonpartisan and avoid manipulating output in favor of any ideological perspective, including diversity, equity, and inclusion (DEI) initiatives. The order also warns against intentionally coding partisan biases into AI responses unless explicitly prompted by users. Although the focus of this order is on government-acquired AI, it also reflects Trump's ongoing concerns about perceived biases within the tech industry. In previous instances, such as during his first term, the administration had urged citizens to report instances of alleged political bias on social media platforms. As the debate over AI bias continues, recent research has indicated that many Americans perceive certain AI outputs as leaning left. This perception may stem from how these systems are trained, often emphasizing accuracy and avoiding potentially offensive content. This raises questions about the effectiveness of Trump's new policy and its potential impact on innovation in the AI sector. The implications of this executive order could be significant for AI developers aiming to collaborate with the government. By imposing additional requirements, there is a concern that it may hinder the pace of innovation that the administration seeks to promote. As the landscape of AI continues to evolve, the balance between political influences and technological advancement remains a critical area of scrutiny.