The Arc of AI: Patterns Across Seventy Years

AI's history is often told as a sequence of winters and springs. That captures the funding rhythm but misses the deeper story: ideas that seemed central in one era often survive as substrates in the next, even when the overall paradigm changes.

Patterns that repeat

• Hype overshoots, disappointment undershoots; the truth lies underneath both
• Compute and data catch up to old ideas and suddenly make them practical
• Narrow wins generalize slowly and surprisingly
• The naming ceremony matters; AI, machine learning, and deep learning each reset the cultural clock

“History doesn't repeat itself, but it often rhymes.”

Three open questions for the next decade

1. 1Does scaling keep paying off, or does progress become idea-limited again?
2. 2Do reasoning and agentic AI deliver robust general capability, or plateau at brittle demos?
3. 3Who governs systems that increasingly mediate knowledge, work, and public life?

The big idea: AI's story is longer, stranger, and more cyclical than the current headlines suggest. Knowing the history is one of the best tools for navigating whatever comes next.