How AI could restart the stalled gears of innovation

With a supercomputer in every pocket and the world's information at our fingertips, you would think we are living in a scientific golden age. The data says otherwise.

Two of the biggest drivers of human progress — productivity and innovation — have been mysteriously stalling for decades. But generative AI might finally be getting the rusty gears of growth moving again.

Science is drowning in its own paperwork

Innovation is slowing across the board. A recent landmark paper revealed that the pace of breakthrough invention is dropping by roughly 50% every 13 years across virtually every discipline, from cancer biology to agriculture.

To keep per capita GDP growth constant, the United States has had to double its research investments every 13 years just to overcome the difficulty of finding new ideas.

A major culprit is the crushing "burden of knowledge." Before a graduate student can even reach the frontier of a field, there is simply too much existing literature to digest. That is why half of all pioneering contributions now happen after age 40, a milestone that used to belong to much younger researchers.

At the same time, startup rates among STEM PhDs have dropped 38% over the last two decades. Projects have gotten so complex that young scientists need giant teams and bureaucratic support just to operate, driving them straight into giant corporations.

More papers are published each month than ever, but researchers compensate by citing canonical, highly-cited old papers rather than discovering new work. AI is stepping in as an intellectual sieve: new studies show machine learning can pinpoint high-potential research paths from past literature, while tools like ChatGPT are already taking over painful tasks like literature reviews and abstract drafts.

The desk job that 1980 left behind

Then there is the frustrating slowdown in productivity gains.

Between 1865 and 1980, the average British worker went from clocking 124,000 lifetime hours down to 69,000, cutting their work life from 50% of their existence to just 20% while enjoying a much higher standard of living.

Since 1980, however, that progress has stalled. Even with personal computers, enterprise software, and high-speed internet, the last decade delivered historically sluggish productivity growth.

Much of managerial and knowledge work has barely changed in forty years: people still spend hours manually wrestling with Word, Excel, and PowerPoint. Generative AI fundamentally shifts that baseline by drafting documents, building prototypes, and assembling slides on command.

The clearest early proof comes from software engineering. In a recent randomized controlled trial run by Microsoft on GitHub Copilot, developers finished programming tasks twice as fast when paired with the AI assistant.

Why it matters

For everyday knowledge workers, creators, and engineers, AI is not just a parlor trick; it is an immediate tax cut on cognitive drudgery. If you write code, analyze data, or draft briefs, using LLMs right now is a no-brainer just to chop the first 50% of prep work off your plate.

For the broader economy, this could finally resolve the modern tech paradox: decades of digital tools that never actually moved the macro productivity needle. If AI genuinely cuts project turnaround times in half, companies will be forced to restructure teams around judgment and validation rather than manual assembly.

We think the smart move here is treating AI as a high-speed filter for human overload rather than an autonomous replacement. The looming risk, however, is clear: if cheap text generation merely floods journals and inboxes with low-effort slop, the burden of knowledge will get even heavier. AI will only rescue productivity if we use it to synthesize noise, not to amplify it.