Fitzgerald's framework names the innovation stream no funder targets
MIT professor Eugene Fitzgerald's new book 'The Invisible Engine: Why Innovation Evades Control' argues that research investment splits into three streams and that one of them has no dedicated funder. According to Fitzgerald, the Merton C. Flemings SMA Professor in MIT's Department of Materials Science and Engineering, altruistic science funds people, strategic research funds a goal, and the 10- to 20-year practice of fundamental innovation, where technology, implementation, and market advance together, falls into neither bucket. His empirical case is strained silicon: discovered at AT&T Bell Laboratories in the 1990s, pushed through MIT and a startup, and ratified by an Intel settlement once the industry found the technology was needed to extend Moore's Law. The path itself is the argument: integration across institutions over decades, not isolated research funding, is what produces the surprise Fitzgerald calls fundamental innovation.
The typology's empirical weight rests on a 30-year arc. Fitzgerald turns strained silicon into a profile of how integration actually happens, and the profile spans discovery at AT&T Bell Laboratories, transfer to MIT, a venture-backed startup, and an Intel settlement once the industry concluded the technology was necessary to extend Moore's Law. According to Fitzgerald, none of the three investment streams captures this trajectory. Altruistic science produces people, not products. Strategic research, his example is the F-35, organizes multiple disciplines around a single customer who absorbs the uncertainty. The third stream, which he names fundamental innovation, has no institutional home and, in his argument, no targeted funding mechanism. Each successful path is assembled ad hoc.
The implementation variable in Fitzgerald's framework is where the current AI moment cuts hardest. Fitzgerald defines implementation as how a technology can actually be built and delivered, and the source describes it as one of three inputs that must run in parallel with technology and market from day one. The book is framed for a moment when artificial intelligence is reshaping how knowledge, research, and innovation happen, but the framework's central case, strained silicon, is a 30-year integration arc, and the source does not test whether faster implementation tooling breaks or accelerates that integration. The framework treats simultaneous progress on all three variables as the requirement; whether that requirement holds when implementation tools compress by an order of magnitude is unanswered in the source.
The institutional response Fitzgerald proposes is what he calls a 'third place,' a university-based convening institution that brings together investors, companies, and researchers to keep all three variables active through long horizons. He argues companies don't have enough time, government funding is too episodic, and universities are the natural site for sustained integration. The empirical anchor is MIT itself: the book draws on the MIT and Masdar Institute Cooperative Program and the MIT-Singapore Alliance for Research and Technology that Fitzgerald led. Whether a third place can be replicated outside an institution with MIT's specific resources, student pipeline, and convening power is the question the source does not address. The book uses MIT-as-success-case as the proof, which is selection-biased by construction.
Knight's surprise concept is load-bearing in the typology. According to Fitzgerald, surprise is the gap between what an entrepreneur commits to (an uncertain outcome) and what they ultimately deliver (a market-accepted product), and profit is the reward for absorbing that uncertainty across all three variables in parallel. The framework makes a competitive claim about which actors are best positioned to absorb long-horizon uncertainty: universities, in Fitzgerald's reading, sit closer to that position than companies optimizing for quarterly results or governments optimizing for procurement cycles. This is a typology, not an empirical comparison, and the source does not benchmark how each actor type actually performs on long-horizon uncertainty.
The strongest claim in the interview rests on a narrow definition of yield. According to Fitzgerald, altruistic science investing in academic institutions produces educated people, and the direct economic yield is 'basically zero' over the years. Treating that contribution as 'basically zero' requires excluding nearly everything except shipped products, and the framework's strongest empirical case, strained silicon, is itself the academic-converted-into-startup pathway the framing discounts. Universities produce the bulk of trained researchers who staff the integration process the book prizes, and the interview does not specify how that contribution is counted in the yield metric.
The framework asks the right diagnostic question about how innovation gets funded, but it does not test two things the diagnosis depends on. Whether the integrated 10- to 20-year process survives when AI tooling compresses implementation to weeks rather than years is one. Whether universities can serve as the missing institutional home for that integration with current funding flows unchanged is the other. The source uses MIT as the proof of concept and one semiconductor path as the empirical anchor, which is consistent with the framework's logic but not enough to validate either scaling question.