The article asserts that the concern of AI running out of data is misguided. It argues that the real issue lies not in the scarcity of data but in the nature of the data itself. The internet is filled with written information, but much of human knowledge is tacit—gained through experience rather than through text. This tacit knowledge includes skills like catching a ball, sensing emotions, or determining when food is ready to cook. Current AI models primarily rely on textual data, which fails to capture the complexities of lived experiences. The piece emphasizes the importance of experiential learning in AI development. It points out that robots and autonomous systems are increasingly using simulations and real-world interactions to learn. For instance, robots improve their grasping techniques through trial and error, while virtual cars undergo extensive simulated driving before hitting the road. This hands-on approach generates rich data from various senses and feedback loops, contrasting sharply with the limited insights derived from written instructions. The article compares text corpora used in AI training to randomized controlled trials in healthcare, highlighting that while both offer valuable insights, they miss the nuanced realities that come from real-life experience. In healthcare, the article draws parallels between structured clinical trials and the need for real-world evidence to capture complexities. Just as RCTs can overlook important variables like social context, AI training based solely on text fails to account for the messy realities of human experience. The author argues for a blended approach, where AI learns from both curated data and experiential learning to better understand and navigate the world. This new perspective on data could lead to more capable and intuitive AI systems.