AI systems can generate scientific claims and candidate solutions, but people still need to understand the work and examine the evidence. Google DeepMind’s Conjecture Machines identifies scientific validation as a growing constraint; OpenAI has organized outside experts to explain and examine AI-produced research. These are the organizations’ assessments and responses, not measurements that every field faces the same bottleneck. One part of the challenge is helping scientists acquire the background needed to evaluate unfamiliar results. How can that learning and examination reach more people while retaining experiments, proof checks and qualified scientific review?