As AI redefines information consumption, Campbell Brown, former TV journalist and Facebook’s pioneering news chief, is on a mission to prevent a repeat of past platform failures. She believes the accuracy and reliability of foundational AI models are paramount, especially for “high-stakes topics.”
Key Takeaways:
- AI’s Accuracy Crisis: Campbell Brown’s Forum AI tackles the critical issue of foundational models providing “slop and wrong answers” on complex, high-stakes topics, a problem she views as an existential threat to informed society.
- Expert-Driven Benchmarking: Forum AI uses a unique methodology, recruiting world-renowned human experts to architect benchmarks and train AI judges to evaluate models, aiming for 90% consensus with human judgment.
- Enterprise as the Unexpected Ally: While idealism suggests AI should optimize for truth, Brown believes the practical demands of enterprise liability in areas like finance, hiring, and insurance will ultimately drive AI companies to prioritize accuracy, despite the current “joke” of compliance.
Campbell Brown has always chased truth. First, under the unforgiving glare of television cameras as a celebrated journalist. Then, in the turbulent digital sphere as Facebook’s inaugural and only dedicated news chief, navigating the complexities of information at scale. Now, with artificial intelligence rapidly reshaping how humanity accesses and processes information, Brown sees a familiar, disturbing pattern emerging: history threatening to repeat itself. This time, however, she’s not merely observing; she’s building the solution.
Her latest venture, Forum AI, founded 17 months ago in New York, is a direct response to this looming information crisis. As she articulated to TechCrunch’s Tim Fernholz at a recent StrictlyVC event in San Francisco, Forum AI’s core mission is to rigorously evaluate how foundation models perform on what she terms “high-stakes topics.” These aren’t simple yes-or-no questions, but rather subjects riddled with nuance and complexity: geopolitics, mental health, finance, and hiring. These are domains where the consequences of inaccurate or biased information are not just inconvenient but potentially catastrophic.
Brown traces the genesis of Forum AI to a specific, almost visceral moment. “I was at Meta when ChatGPT was first released publicly,” she recalled, “and I remember really shortly after realizing this is going to be the funnel through which all information flows. And it’s not very good.” The implications struck her not just professionally, but personally. “My kids are going to be really dumb if we don’t figure out how to fix this,” she vividly remembers thinking. It was an existential alarm bell, sounding a warning about a future where foundational knowledge could be fundamentally flawed.
What she found most frustrating was the apparent disinterest in accuracy among the very companies building these powerful new tools. Foundation model developers, she observed, are “extremely focused on coding and math,” while the messy, subjective, and often contradictory world of news and information is largely sidelined. Brown’s unequivocal argument? Just because it’s harder, “doesn’t mean optional.” The challenge of ingesting, synthesizing, and presenting factual, contextualized information is immense, but essential if AI is to become a trusted, rather than detrimental, source.
Forum AI’s approach to evaluating these models is both innovative and deeply rooted in human expertise. Recognizing that “smart generalists aren’t going to cut it” for complex domains, Brown has assembled a formidable roster of the world’s foremost experts. For their geopolitics work, for instance, she’s recruited intellectual titans like Niall Ferguson and Fareed Zakaria, alongside seasoned political figures such as former Secretary of State Tony Blinken, former House Speaker Kevin McCarthy, and Anne Neuberger, who spearheaded cybersecurity in the Obama administration. These experts don’t just review; they architect the benchmarks, defining what ‘good’ looks like in their respective fields. Forum AI then trains AI judges to evaluate models at scale, with the ambitious but achievable goal of reaching roughly 90% consensus with those human arbiters – a threshold Brown confidently states Forum AI has been able to consistently reach.
The early findings from Forum AI’s rigorous evaluations have not been reassuring. Brown cited instances of leading models, like Gemini, pulling information from Chinese Communist Party websites “for stories that have nothing to do with China.” Beyond overt misinformation, a more subtle, yet pervasive, left-leaning political bias has been detected across nearly all models. The failures extend further, encompassing missing critical context, overlooking diverse perspectives, and straw-manning arguments without proper acknowledgment. “There’s a long way to go,” she conceded, but quickly added, “But I also think that there are some very easy fixes that would vastly improve the outcomes.” The problem, she implies, is not intractable, but requires a deliberate shift in priorities.
Brown’s deep experience at Facebook serves as a potent cautionary tale. For years, she witnessed firsthand the consequences of a platform optimizing for the wrong metrics. “We failed at a lot of the things we tried,” she candidly told Fernholz, noting that the ambitious fact-checking program she painstakingly built no longer exists in its original form. The stark lesson from the social media era, often ignored, is that prioritizing engagement above all else has been “lousy for society” and left countless individuals less informed, more polarized, and increasingly distrustful. This painful past informs her present drive to ensure AI doesn’t repeat the same mistakes, amplifying them through its unprecedented scale and authority.
Her hope, therefore, is that AI can break this destructive cycle. “Right now it could go either way,” she mused; companies could continue to give users what they *want* – often engagement-driven, emotionally charged content – or they could “give people what’s real and what’s honest and what’s truthful.” She acknowledged that the idealistic vision of AI singularly optimizing for truth might sound naive. However, she believes enterprise demand may be the unexpected, pragmatic ally in this quest. Businesses leveraging AI for critical functions – credit decisions, lending, insurance underwriting, and hiring processes – are acutely aware of liability. “They’re going to want you to optimize for getting it right,” she asserted, seeing this commercial imperative as a powerful catalyst for change.
This enterprise demand is precisely what Forum AI is banking its business model on. Yet, translating genuine compliance interest into consistent revenue remains a significant challenge. Much of the current market, Brown observes, is still satisfied with superficial “checkbox audits” and standardized benchmarks that she considers woefully inadequate. The existing compliance landscape, in her scathing assessment, is “a joke.” She pointed to New York City’s pioneering hiring bias law, which mandated AI audits, only for the state comptroller to discover that over half of these audits failed to detect actual violations. Real evaluation, she stressed, demands deep domain expertise to navigate not only known scenarios but also the perilous, complex “edge cases that can get you into trouble that people don’t think about.” And such thorough, nuanced work takes considerable time and specialized knowledge.
Brown, whose company secured $3 million in funding last fall led by Lerer Hippeau, is uniquely positioned to articulate the growing disconnect between the AI industry’s self-proclaimed grandeur and the frustrating reality experienced by most users. “You hear from the leaders of the big tech companies, ‘This technology is going to change the world,’ ‘it’s going to put you out of work,’ ‘it’s going to cure cancer,'” she observed. “But then to a normal person who’s just using a chatbot to ask basic questions, they’re still getting a lot of slop and wrong answers.” This chasm fuels a widespread skepticism, with trust in AI languishing at extraordinarily low levels – a skepticism Brown believes is, in many cases, entirely justified. “The conversation is sort of happening in Silicon Valley around one thing, and a totally different conversation is happening among consumers.” Forum AI aims to bridge that gap, forcing the conversation toward a shared understanding of accuracy and reliability.
Bottom Line:
Campbell Brown’s journey from confronting media bias to tackling AI’s accuracy problem underscores a consistent commitment to informed truth. With Forum AI, she’s not just flagging deficiencies; she’s building an essential framework to hold AI accountable, hoping to leverage the pragmatic demands of enterprise liability to ensure that the next generation’s “information funnel” is built on a bedrock of reliable, nuanced, and expertly validated data, preventing a digital future where our collective intelligence is sacrificed for speed or engagement.
When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.
Source: {feed_title}