Rodney Brooks warns technologies run out of steam

The author argues that transformative technologies—such as steam power, neural networks, and reinforcement learning—typically require decades of sustained, marginal effort before succeeding, and that successful technologies eventually run out of steam when faced with new problems. Therefore, the author cautions that society cannot predict which current technologies will solve future problems and must continue investing in diverse, early-stage research. The post illustrates this lifecycle through the multi-decade histories of steam engines and foundational machine learning techniques. Open article →