Lately I’ve been thinking about the Collingridge dilemma.
When a technology first emerges we can easily shape its development. Yet we can't fully grasp its impact until it’s widely used.
When change is easy at the beginning, the need for it can’t be foreseen. Once the need for change is apparent, it’s become difficult, expensive, and time-consuming to implement.
Jun 23, 2025 07:52Trade-offs evolve over time. Regulators have a timing challenge: wait too long, and the technology becomes entrenched; act too early, and you regulate without understanding the full picture.
The same also applies to our use of AI tools. When you uncritically accept AI suggestions, you exchange the control you have whilst change is easy for much more challenging future work.
Understanding when and how to intervene—maintaining agency whilst leveraging AI's capabilities—is, I think, one of the defining technical skills of our age.
