TB: How Long Do We Wait for New Inventions? (Potter)

Brian Potter used AI to analyze a corpus of about 190 major inventions and ask: how long after a technology becomes technically possible does it actually appear? The answer is shorter than the popular "we should have had X 50 years earlier" framing suggests — most inventions arrive within decades of feasibility, and the lag has compressed sharply since 1900. Inventions wait on technology more than on science.

• 64% of analyzed inventions appeared within 50 years of when they first became technically possible; using a "straightforward" possibility date, 90% appeared within 50 years.
• Post-1900 inventions show dramatically shorter wait times: for the 60 post-1900 cases, every one had a "straightforward" invention date of 50 years or less, and 75% appeared within 10 years.
• John's annotation in Blogs.org: "Research on AI output … Answer is not long" — the relevance to AI is that "feasible-but-undiscovered" gaps may be small, not large.
• Medical inventions have the longest delays among categories — institutional, ethical, and clinical-trial friction matter.
• Electronics show the shortest development timelines — the category where pure technological feasibility translates almost immediately into product.
• Technological barriers (manufacturing, materials, instrumentation) outweigh scientific ones as the binding constraint on invention timing.
• A small tail of inventions waited 1000+ years between possibility and instantiation; eight cases were that extreme.
• Mechanism for compressing wait times: faster knowledge cross-pollination between fields and rising background technology level.

• "We mostly don't wait all that long for new inventions."
• "For the 60 post-1900 inventions, every one has a 'straightforward' invention date of 50 years or less."
• "Why, then, was Towne's insight so novel? The answer appears to be that in 1951, physicists and engineers in the United States were not yet sufficiently acquainted with each other's territory." — historian Joan Bromberg, quoted by Potter

A useful prior for thinking about the AI-research-acceleration thesis: if "the inventions we don't have are technically blocked" rather than "stuck behind insight we've already had," then AI-accelerated research changes the constraint that actually binds. The piece also bears on John's working interest in how to navigate large search spaces — Potter's companion essay on information and technological evolution argues that intermediate stepping-stones, not random recombination, are what make those spaces tractable.

• Information and Technological Evolution (Potter) — same author; complementary thesis on how inventions get built
• The Beauty of Slow Research Assistants (Fletcher) — what the new AI cycle is and isn't likely to compress
• AI and Research Papers (Kling) — same theme of AI as research-process accelerator

How Long Do We Wait for New Inventions? — Brian Potter, Construction Physics (Substack), 2026-05-07