Innovation is the beating heart of economic progress, but its lack of manifestation in productivity statistics has given economists heartache for the last couple of decades. Notwithstanding better performance recently, the picture post-2005 is one of a marked productivity slowdown. The Bureau of Labor Statistics computed that this slump cost the US economy over $10 trillion compared to a scenario where productivity growth would have matched its post-WWII average. Among the diagnoses floated by economists at the bedside of this patient is that innovation is harder than it used to be. Encouragingly, there are early indications that AI could be a game changer.
As early as 2009, economist Ben Jones observed that it is getting harder for researchers to ingest the state of the art and push the frontier. PhDs are taking longer to complete; Nobel Prize winners generally achieve their breakthroughs later in life; and academics have had to become more specialized. Since innovation often arises by combining insights from different fields, scientists have increasingly relied on collaboration as part of progressively larger teams. The notion that ideas are getting harder to find is further documented in a famous 2020 paper by Nicholas Bloom and colleagues, evidencing falling research productivity spanning multiple industries, products, and firms. According to their estimates, the United States needs to double the amount of research effort every 13 years just to sustain constant growth in GDP per person.
While this backdrop is challenging, there are indications that AI is poised to revolutionize research productivity. The most tangible manifestation of this is Google DeepMind’s AlphaFold system, whose transformative impact was recognized by the award of the 2024 Chemistry Nobel Prize to Demis Hassabis and John Jumper. AlphaFold solved a 50-year-old scientific challenge known as the protein-folding problem, unlocking predictions of the structure of some 200 million proteins based on their sequences of amino acids alone. Prior to AlphaFold, investigating the structure of a single protein could take years and cost hundreds of thousands of dollars. This tool is now available for free to the scientific community and has already been used by millions of researchers from more than 190 countries, supporting discoveries in areas such as malaria vaccines and cancer treatments.
One of the key contributions of AI is turbocharging idea generation. This is because many problems take the form of finding a needle in the proverbial haystack. Scientists seeking to formulate a small molecule drug face more than 10^60 potential options; in the case of a protein with 400 standard amino acids, they face 20^400 options. These represent overwhelming search spaces at human scale, but if AI can predict which solutions are more promising, the challenge becomes a lot more manageable - an approach that is already leading to breakthroughs from mathematics to computer chip architecture.
Science fiction movies frequently depict genius protagonists co-innovating with their trusted AI assistants. Emerging AI breakthroughs and applications suggest that this vision of the future may be around the corner. In fact, AI co-scientist systems that help researchers generate novel hypotheses are already available - impressing scientists at Imperial College with their performance. But realizing the full potential of AI for science will not come automatically and requires deliberate actions on the part of the private and public sectors. Google recently released a set of recommendations for policymakers to empower AI-based scientific research. These include improving access by scientists to AI infrastructure through initiatives such as the US National AI Research Resource, boosting public investment to complement private initiatives, and articulating pro-innovation regulatory regimes giving scientists the clarity they need to move forward with their research.
Unleashing the spark of science is only one step in rekindling the fire of productivity across the economy - but it is a critical one. And just as important as the economic gains that stand to be realized are the human ones from advances in health and wellbeing. This is a generational opportunity.