Let us know about free updates
Just sign up for Technology Sector Myft Digest – Delivered directly to your inbox.
The author is the author of “How Progress Exting: Technology, Innovation, ant the Fate of Nations” and an associate professor at Oxford University.
With each terror of AI-driven unemployment burns, optimists reassure us that artificial intelligence is a productivity tool that supports both workers and the economy. Microsoft's chief Satya Nadella believes that autonomous AI agents can name goals while the software is planning, running and learning on all systems. Dream Tools – If efficiency alone is enough to solve productivity problems.
History says that isn't the case. Over the past half century, we have filled our offices and pockets with faster computers, but the growth in labor productivity in the developed economy slowed from about 2% per year in the 1990s to about 0.8% in the last decade. Even China's former bold product per worker is stagnant.
The marriage of computers and internet shotguns promised more than increasing office efficiency. It envisioned a golden age of discovery. Breakthroughs should have been increased by placing world knowledge in front of everyone and combining global talent. However, research productivity is declining. The average scientist now has fewer groundbreaking ideas per dollar than his 1960s counterparts.
What was wrong? As economist Gary Becker once noted, parents face trade-offs between quality and quantity. The more children you have, the less you invest in each child. The same can be said about innovation.
Large-scale studies on creative output confirm the results. Researchers are less likely to bring about groundbreaking innovations by juggling more projects. Over the last few decades, scientific papers and patents have become increasingly progressive. The greats of history understood why. Isaac Newton said, “I kept the problem that was constantly in front of me, until the first doughing slowly opened slowly, slowly, into a complete, clear light.” Steve Jobs agreed, “Innovation says no to a thousand things.”
Where precedents are weak, human ingenuity will flourish. If the 19th century was focused solely on better looms and ploughs, we enjoyed cheap fabrics and abundant grains, but no antibiotics, jet engines or rockets. Economic miracles stem from discovery and do not repeat tasks at a more rapid rate.
Large-scale language models are drawn to statistical consensus. The model trained before Galileo would have parroted the Earth-centric universe. Given the 19th century text, it would have proved that human flight was impossible before the Wright brothers were successful. A recent review of nature has shown that while LLMS lightens everyday scientific chores, the crucial leap of insights still belongs to humans. Even Demis Hassabis, the team at Google Deepmind, produced Alphafold, produced a model that predicts protein shapes and perhaps the most famous scientific feat of AI.
In the interim, AI mainly increases efficiency rather than creativity. A survey of over 7,000 knowledge workers found that heavy users of generator AI reduced weekly email tasks by 3.6 hours (31%), but collaboration was unchanged. However, if anyone delegates email responses to ChatGPT, it could increase the volume of your inbox and disable the increased initial efficiency. The short revival of America's productivity in the 1990s teaches us what new tools can come from, whether it's a spreadsheet or an AI agent, unless it involves groundbreaking innovation.
AI can still fire a productivity renaissance, but only if you use it to dig deeper for new, previously unthinkable efforts, rather than simply drilling holes. It means rewarding originality over volume, supporting riskier bets and restoring autonomy. The algorithm may be ready soon. Our institutions must now adapt.
