# Mapping AI Capabilities to Human Expertise: A New Rosetta Stone > Coverage of lessw-blog **Published:** March 09, 2026 **Author:** PSEEDR Editorial **Category:** platforms **Content tier:** free **Accessible for free:** true **Word count:** 497 **Tags:** AI Evaluation, Epoch Capability Index, AI Forecasting, Human Baselines, Frontier Models **Canonical URL:** https://pseedr.com/platforms/mapping-ai-capabilities-to-human-expertise-a-new-rosetta-stone --- lessw-blog introduces an extension to the Rosetta Stone framework, anchoring abstract AI capability scores to concrete human expertise levels to better forecast when AI might reach top-tier human performance. In a recent post, lessw-blog discusses a critical extension of the Rosetta Stone framework and the Epoch Capability Index, aiming to solve a persistent problem in artificial intelligence evaluation: how to map abstract benchmark scores to real-world human expertise. As frontier AI models continue to scale rapidly, tracking their actual utility has become increasingly difficult. Traditional benchmarks often produce relative capability scores that lack intuitive, real-world anchors. For developers, policymakers, and enterprise leaders, knowing that a model scores a specific numerical value on an isolated evaluation is far less useful than knowing whether it performs at the level of a crowd worker, a skilled generalist, or a PhD-level expert. Establishing these concrete baselines is critical for assessing the real-world impact of advanced AI, guiding deployment strategies, and informing risk assessment-especially within the rapidly evolving ecosystem of AI agents, evaluation frameworks, and synthetic data generation. lessw-blog has released analysis integrating human performance baselines directly into the Rosetta framework. By mapping AI capabilities against human benchmarks-ranging from average individuals to top-tier performers-the author provides a highly interpretable scale for AI progress. The analysis reveals a striking timeline: current frontier models, when restricted to technical and scientific benchmark skills, have already surpassed Average Humans (late 2022) and Skilled Generalists (early 2024), and are currently crossing the threshold of Domain Experts in 2025. Looking ahead, the post forecasts that future AI models could reach Top-Performer human levels by October 2027 (with a 95% confidence interval spanning May 2027 to March 2028). However, the author is careful to note that this timeline comes with caveats. The analysis challenges the assumption of a single axis of capability by examining benchmarks specifically designed to be easy for humans but hard for AI systems. Furthermore, inconsistent and sparse human performance data remains a significant bottleneck in creating perfectly accurate projections. This work offers a highly significant metric for understanding the trajectory of AI progress, providing an actionable framework for future planning and policy decisions. For a deeper look into the methodology, the integration of human baselines, and the nuances surrounding these capability forecasts, [read the full post](https://www.lesswrong.com/posts/cfbdyJGbHkY8rPesE/mapping-ai-capabilities-to-human-expertise-on-the-rosetta-1). ### Key Takeaways * The original Rosetta Stone framework has been extended to include human performance baselines, providing concrete real-world anchors for AI capability scores. * Current frontier AI models have already surpassed Average Humans and Skilled Generalists on technical and scientific benchmarks, and are reaching Domain Expert levels in 2025. * Future AI models are forecasted to reach Top-Performer human levels by October 2027, offering a critical timeline for risk assessment and policy planning. * The analysis challenges the assumption of a single axis of capability by examining benchmarks designed to be easy for humans but hard for AI. * Inconsistent and sparse human performance data remains a primary bottleneck in accurately mapping AI capabilities to human expertise. [Read the original post at lessw-blog](https://www.lesswrong.com/posts/cfbdyJGbHkY8rPesE/mapping-ai-capabilities-to-human-expertise-on-the-rosetta-1) --- ## Sources - https://www.lesswrong.com/posts/cfbdyJGbHkY8rPesE/mapping-ai-capabilities-to-human-expertise-on-the-rosetta-1