A new AI framework helps machines better handle uncertainty by combining human judgment with machine learning

Scale AI, a data-labeling startup, has confirmed a major investment from Meta, boosting the company’s valuation to $29 billion. As part of the deal, Scale’s co-founder and CEO Alexandr Wang will step down to join Meta, focusing on its superintelligence initiatives. The investment, reported at $14.3 billion for a 49% stake, underscores Meta’s strategic move to strengthen its AI capabilities amid competition from rivals like Google, OpenAI, and Anthropic. Meta’s partnership with Scale AI aims to enhance collaboration in producing training data for AI models. Wang’s departure highlights the growing demand for expertise in AI development, as Meta seeks to close gaps in its own model releases. Jason Droege, Scale’s current chief strategy officer, will assume the role of interim CEO. The company emphasized its independence, noting Wang will remain on the board. The funding will prioritize returning capital to shareholders and accelerating growth. Scale AI has recently intensified hiring efforts, targeting PhD scientists and senior engineers to meet rising demand for high-quality data in frontier AI research. Last year, the startup raised $1 billion at a $13.8 billion valuation, with Amazon and Meta among its investors. Meta’s investment reflects the critical role of training data in the AI race, as companies compete to lead advancements. Wang’s transition to Meta signals a broader trend of talent and resources shifting toward large-scale AI projects. His new role will focus on superintelligence, a concept involving highly advanced AI systems capable of complex problem-solving. Scale AI’s work has been pivotal in supporting leading AI labs, including OpenAI, by providing annotated data for language models. The startup’s focus on uncertainty quantification—teaching AI systems to assess confidence in predictions—aligns with Meta’s goals to improve decision-making in ambiguous scenarios. This approach could address challenges in areas like medical diagnostics, where incomplete data requires balancing technical accuracy with human judgment. The company’s framework, developed by Willie Neiswanger, a USC professor, integrates classical decision theory with utility principles to enhance AI’s handling of uncertainty. His research, presented at the 2025 International Conference on Learning Representations, explores applications in sequential decision-making, such as optimizing drug discovery or logistics. By modeling human preferences, the system ensures recommendations align with real-world values, not just mathematical efficiency. Neiswanger’s team also aims to improve transparency, creating interfaces that explain AI decisions to users. This focus on “human auditability” could make AI systems more trustworthy, particularly in critical fields like healthcare or finance. The collaboration with Meta highlights the industry’s push to merge AI’s computational power with human-centric reasoning, addressing gaps in current models that often lack nuanced understanding of uncertainty. As AI becomes more integrated into complex decision-making, the ability to navigate ambiguity remains a key challenge. Scale AI’s investment and leadership changes mark a significant step in bridging this gap, with implications for both corporate strategies and the broader AI landscape.