Censorship by Proxy: The Oversight Board’s Latest Warning

The Oversight Board, created by Meta as a self-proclaimed champion of free expression, has been seeking to expand its influence beyond the social media giant. A recent report from the board should give pause even to those who are ardent defenders of AI.

According to the board’s research, leading AI models are more likely to restrict users’ free expression when prompted with questions related to governments that have restrictive speech laws. This finding is based on testing 10 different models, including those from OpenAI, Google, and Anthropic, which all exhibited a similar bias towards restricting speech.

The problem lies not only in the AI itself but also in how these models are being trained and deployed. Many of the companies involved in this research rely on government data to train their models, leading to biased outcomes. Even when they’re not directly influenced by governments, these companies often cite local laws as a reason for restricting speech, even if those laws don’t apply in the jurisdiction where the request was made.

Censorship by proxy is a threat to free expression on a global scale. The Oversight Board’s report should be a wake-up call for AI companies, which have been quietly amassing power and influence without being held accountable for their actions. As Paolo Carozza, co-chair of the board, notes, social media companies have learned hard lessons about transparency and vigilance in the past, but AI companies seem to be repeating those mistakes.

The report’s focus on human rights law is particularly noteworthy. The Oversight Board recommends that AI companies publicly disclose their responses to government requests affecting model output and establish policies for handling content restrictions that may be inconsistent with international human rights law. This proposal is sensible, but it remains unclear whether AI companies will take it seriously.

One reason for skepticism is the lack of a formal structure for the Oversight Board to influence policy decisions at these companies. Even if they did try to exert their influence, it’s unlikely to have much impact, given that many of these companies are already pushing back against regulatory efforts.

The implications of this report are far-reaching and demand attention from policymakers, regulators, and the public. We need to start taking a closer look at the models being developed and deployed in our name, demanding more transparency from AI companies about how their models are trained and how they respond to government requests. Recognizing the real-world implications of censorship by proxy is crucial: it’s not just about speech; it’s about the erosion of trust in institutions and the spread of misinformation.

The Oversight Board’s report serves as a warning sign, but also an opportunity for us to get ahead of this issue before it spirals out of control. Will AI companies listen, or will they continue to ignore warnings like those given by social media platforms in the past? The answer will determine not only the future of free expression online but also the very fabric of our democracy itself.

The Oversight Board’s report is a stark reminder that we’re still living in a world where technologies can have profound impacts on people’s capacity to express themselves. As Carozza said, “one has to be really vigilant” – and right now, it seems like no one is paying attention.