I’m a Member of the Meta Oversight Board: Urgent Need for AI Safeguards | Suzanne Nossel

The rapid pace of artificial intelligence (AI) advancement has become truly astonishing. Unlike previous technological revolutions—be it the advent of radio, the discovery of nuclear fission, or the rise of the internet—it’s noteworthy that governments are not taking the lead in this AI revolution. Concerns about AI’s potential dangers are becoming increasingly evident. For instance, chatbots have been known to give alarming advice to vulnerable individuals, such as teenagers contemplating suicide, and there are reports suggesting that these technologies might soon guide people on creating biological weapons. Yet, unlike industries like pharmaceuticals, where entities such as the Federal Drug Administration rigorously evaluate safety before new products hit the market, AI faces no such regulatory body. In fact, in the tech landscape, there’s often little obligation for companies to disclose significant breaches or dangerous incidents. The immense lobbying power of the tech industry, alongside the deep political divisions in Washington, combined with the intricate nature of rapidly evolving technologies, have effectively kept comprehensive federal regulations at bay. In Europe, officials face opposition to regulatory measures that some argue might hinder the region’s competitive edge in the global market. While several states in the US are testing the waters with AI regulations, they are currently enmeshed in a hesitant and inconsistent framework, with former President Donald Trump trying to invalidate them.

Ceos of leading AI platforms, such as OpenAI’s ChatGPT and Google’s Gemini, insist on their commitment to safety. However, the quest for AI supremacy entails investing hefty sums into technologies that are often beyond the comprehension of even their creators. Key decisions, such as the integration of ads, and functionalities that the Pentagon is now seeking from Anthropic, inherently increase the associated risks. Anthropic presents itself as a vanguard in responsible AI development, claiming that their models are designed to ponder how a thoughtful senior employee would weigh usefulness against potential risks. This philosophy harkens back to earlier critiques of Silicon Valley’s approach, suggesting that major tech firms have been shaping global user experiences from insulated positions, leaving consumers uneasy. Alarmingly, a survey indicates that 77% of Americans now perceive AI as a potential threat to humanity.

Currently, we find ourselves at a crossroads where neither robust government regulations nor effective self-policing by the largest corporations prevails. Nevertheless, independent oversight presents a crucial opportunity to mediate the tension between AI’s expansive possibilities and its significant threats. By adopting independent oversight mechanisms, AI firms could prove their genuine commitment to earning public trust through transparency and accountability.

The rationale behind independent oversight is uncomplicated yet compelling. Regardless of the corporate executives’ good intentions, their responsibilities towards shareholders and investors often dictate a focus on short-term profitability over long-term safety. This tendency may lead them to prioritize revenue generation, thereby compromising ethical considerations. Recent historical examples illustrate how the disruptive power of technology, particularly in social media, has sometimes overlooked glaring warning signs—contributing to consequences including violence, electoral interference, and deteriorating mental health among youth.

Implementing independent oversight for AI could help reveal, evaluate, and mitigate risks, granting communities and advocates a greater role in shaping the societal impact of these technologies. A pertinent illustration comes from the realm of social media. Following severe criticism for purportedly exacerbating the Rohingya crisis in Myanmar, Meta, previously known as Facebook, instituted an oversight board as a form of self-regulation. The move aimed to distance the company from its past missteps. In the subsequent year, Meta declared a policy to adhere to human rights standards. While the oversight board has not fully met every expectation—often referred to as a potential “supreme court of Facebook”—its existence provides valuable insights on the feasibility and necessity of effective independent oversight in the AI domain.

Effective oversight hinges on the inclusion of varied perspectives. Meta’s global presence means it has users in nearly every country. Decisions made within the company, shielded in its Menlo Park headquarters, often lack the nuances of understanding diverse local cultures, leading to unintended omissions and backlash. The oversight board consists of 21 members who bring extensive cultural and professional insights to complicated matters of content moderation, such as determining whether a violent clip should be shared for informational purposes or removed due to its potential to offend the victim’s dignity. Members come from a range of backgrounds, including conservatives and liberals, journalists, legal experts, and even notable public figures like a former Danish prime minister and a Nobel Peace Prize laureate.

Employing Meta’s own guidelines, the oversight board evaluates whether specific posts violate established community standards that prohibit harmful content, bullying, or support for terrorism. This oversight ensures that Meta remains true to its promises regarding human rights, particularly in accordance with Article 19 of the International Covenant on Civil and Political Rights, which guarantees freedom of expression. AI corporations should similarly adopt such commitments and establish oversight mechanisms to uphold them. By leveraging international human rights principles, a common standard transcending borders can be established. This framework could guide decisions regarding AI, such as assessing whether a bot’s refusal to provide information unjustifiably infringes upon a user’s rights.

Accessibility, transparency, and public engagement are vital for effective oversight. The oversight board at Meta is open to public appeals, shares which cases it will assess, seeks public commentary, and collaborates with experts and affected communities. To date, it has rendered over 200 detailed decisions that have influenced courts around the globe.

The strength of a voluntarily established oversight body is inherently linked to the powers bestowed upon it by its parent organization. The oversight board at Meta wishes for expanded authority but has nonetheless gained recognition for significantly transcending the superficial advisory groups that other tech companies often establish and then dissolve. While the board can make authoritative decisions on whether specific pieces of content should remain visible or be removed—an undertaking that may seem akin to battling a wildfire with mere smoke blowers—its most substantial impact lies in selecting pertinent cases that spotlight larger issues, offering rationales for its conclusions, and delivering recommendations that Meta must acknowledge. Notably, it has adopted around 75% of the board’s more than 300 suggestions, as declared in December, prompting significant transformations that affect billions of users.

Changes have included ensuring users receive notifications about which policies they allegedly violated when content is removed, safeguarding rhetorical expressions and satire from being inappropriately flagged, and bolstering resources during crises such as natural disasters or armed conflicts. Furthermore, the board addresses broader policy matters, offering detailed guidance on critical issues like Meta’s policy adjustments for high-profile users or managing the removal of misinformation as the COVID pandemic subsided. Even though the board operates independently, its effectiveness hinges on the quality of information it receives from Meta regarding content management decisions, whether human-driven or automated, as well as insights into errors experienced during content removal. AI firms will need to ensure at least an equivalent level of transparency for their oversight to be genuinely meaningful.

Financial considerations are paramount. Meta secures the oversight board’s budget in an irrevocable trust to prevent sudden funding cuts, yet broader and more stable financial support would augment the board’s independence and efficacy. Oversight of cutting-edge technology demands financial resources, which include expert staff to facilitate thorough analyses and decision-making, as well as consultants with enhanced cultural or linguistic competence. Given the astronomical investments being funneled into AI development, the costs associated with robust oversight are a mere drop in the bucket.

As AI becomes increasingly integrated into educational institutions, corporate environments, and various facets of our lives, it is imperative for AI organizations to adopt independent oversight practices. This is the minimal foundational step they can take to ensure that, whether intentionally or unintentionally, they do not infringe upon our fundamental rights as individuals.