In a world increasingly shaped by artificial intelligence, a critical question emerges: who truly dictates the information AI delivers? Campbell Brown, a veteran journalist and former news chief at Meta, isn’t just asking this question; she’s building a solution. After a career dedicated to accurate information, Brown recognized a familiar pattern in the rise of AI, reminiscent of social media’s early days, but with even greater potential for societal impact.
Driven by the personal stakes for her own children and a broader concern for public discourse, Brown founded Forum AI. Her mission is clear: to ensure the foundational models shaping our future are built on a bedrock of truth and nuance. She isn’t waiting for others to tackle this monumental challenge; she’s taking the lead to define what “accurate” means in the age of AI.
Taming the AI Wild West with Expert Evaluation
Forum AI is spearheading efforts to evaluate how leading AI models perform on what Brown calls “high-stakes topics.” These aren’t simple yes-or-no questions, but complex areas like geopolitics, mental health, finance, and hiring, where answers are often murky and require deep understanding. The company’s innovative approach centers on bringing together the world’s foremost human experts to create rigorous benchmarks.
Once these benchmarks are established, Forum AI trains sophisticated AI judges to evaluate models at an unprecedented scale. The ambitious goal is to achieve approximately 90% consensus between these AI judges and the human experts. This crucial threshold demonstrates a level of accuracy and nuance that Brown believes is essential for trustworthy AI.
For its critical work in geopolitics, Forum AI has assembled an incredibly distinguished panel of experts. This roster includes:
- Niall Ferguson
- Fareed Zakaria
- Former Secretary of State Tony Blinken
- Former House Speaker Kevin McCarthy
- Anne Neuberger, who previously led cybersecurity in the Obama administration
This commitment to deep domain expertise is what sets Forum AI apart, ensuring that complex subjects are judged by those who truly understand their intricacies.
Uncovering AI’s Blind Spots and Biases
Brown’s journey to founding Forum AI began during her time at Meta, coinciding with the public release of ChatGPT. She distinctly remembers the realization that AI would become a primary funnel for information, and critically, that its output was “not very good.” This observation, coupled with concerns for her children’s future learning, solidified her resolve to address the issue head-on.
Her frustration grew from observing that while foundation model companies excelled in areas like coding and math, the nuanced world of news and information was largely overlooked. Yet, for Brown, the complexity of information doesn’t make it optional; it makes it paramount. The initial evaluations conducted by Forum AI underscored her concerns, revealing significant shortcomings in leading models.
Brown cited several troubling examples, including Gemini’s tendency to pull from Chinese Communist Party websites for unrelated stories. Forum AI also identified a pervasive left-leaning political bias across nearly all models it tested. Beyond overt biases, the evaluations consistently revealed subtle yet critical failures, such as missing context, omitting diverse perspectives, and presenting “straw-manned” arguments without proper acknowledgment. Clearly, there’s a long road ahead, but Brown also sees some surprisingly simple fixes that could dramatically improve outcomes.
Reflecting on her years at Facebook, Brown drew parallels to the challenges of optimizing for engagement over accuracy. She openly acknowledged that many initiatives failed, including the fact-checking program she helped build. The enduring lesson from social media, often ignored, is that prioritizing engagement above all else can be detrimental to society and leave users less informed.
Enterprise Demand: The Unexpected Ally for AI Accuracy
Despite the current challenges, Campbell Brown remains hopeful that AI can break free from the patterns of social media. She believes the technology stands at a crossroads, with the potential to either cater to user preferences or, more profoundly, to “give people what’s real and what’s honest and what’s truthful.” While optimizing AI for truth might sound idealistic, Brown sees a powerful, and perhaps unexpected, ally: enterprise demand.
Businesses utilizing AI for high-stakes decisions—like credit assessments, lending, insurance, and hiring—have a vested interest in accuracy due to significant liability concerns. These companies will unequivocally demand that AI models “optimize for getting it right.” This imperative for correctness forms the bedrock of Forum AI’s business strategy.
However, translating this interest in compliance into consistent revenue remains an ongoing challenge. Much of the current market is still content with superficial “checkbox audits” and standardized benchmarks that Brown considers utterly inadequate. She candidly describes the prevailing compliance landscape as “a joke,” pointing to instances like New York City’s hiring bias law, where over half of audited AI systems had undetected violations.
Real evaluation, according to Brown, necessitates deep domain expertise to navigate not only known scenarios but also obscure edge cases that can lead to unforeseen trouble. This rigorous, time-intensive work requires specialists, as “smart generalists aren’t going to cut it.” Forum AI, which successfully raised $3 million led by Lerer Hippeau last fall, is uniquely positioned to bridge this critical gap.
Brown emphasizes the significant disconnect between the AI industry’s grand pronouncements and the everyday reality for users. While Big Tech leaders extol AI’s world-changing potential—predicting job displacement and cancer cures—the average person using a chatbot still frequently encounters “slop and wrong answers.” Forum AI is committed to bringing clarity and accountability to this evolving landscape, ensuring that AI lives up to its true potential for good.
Source: TechCrunch – AI
