Michael Nielsen – How science actually progresses

In a deep and expansive conversation on the Dwarkesh Podcast, Michael Nielsen explores the complex, non-linear nature of scientific progress, challenging the traditional view of science as a straightforward process of hypothesis testing and falsification. Drawing on historical examples such as the Michelson-Morley experiment and the development of relativity, Nielsen illustrates how scientific breakthroughs emerge from a web of competing interpretations, aesthetic judgments, and long verification loops—often taking decades to gain acceptance even after empirical support. He emphasizes that progress is driven not just by data, but by community dynamics, heuristic intuition, and the 'market for follow-ups,' where ideas gain momentum through sustained engagement. The discussion extends to the role of AI in science, noting its success in narrow domains like protein folding but its limitations in generating conceptual leaps, suggesting that future scientific advances may come from interpreting complex, non-transparent models through distillation and archaeological analysis. Nielsen also reflects on the contingent rise of quantum computing, shaped by technological advances and personal discovery, and advocates for open science as a reimagining of credit and collaboration in the digital age. He concludes with a powerful meditation on deep learning, arguing that true understanding requires demanding, active engagement—such as implementing algorithms or writing essays—rather than passive consumption or overreliance on AI, which can substitute for the hard, aversive work essential for intellectual mastery.