Patrick Collison and Michael Nielsen wrote in The Atlantic in 2018 that maintaining the pace of science is getting more expensive, at least as measured by survey judgements of the relative importance of discoveries. Specifically, they find that despite an exponentially increasing input in the form of research dollars and PhD students, discoveries have been declining in survey-assessed importance over time.
The same trend can be seen in more concrete measures of scientific output. In developed countries agricultural yields rose at roughly 2%/year from 1961-2003, and R&D spending rose at a pretty similar pace over the same period. At first glance that seems reasonable, we got 2%/year improvements in exchange for 2%/year increases in scientific inputs, but it actually means that every additional 2% improvement costs more than the last!
Bloom+2020 find that similar trends are common in scientific fields with concrete, measurable outputs (apart from publications), and lead to the conclusion that the next increment of science is getting more expensive than the last. At the aggregate level, this manifests by steady or slowing productivity growth in the economy despite massive increases in total R&D spending, shown in the following figure from Bloom+2020 in blue and green respectively:
![Screen Shot 2021-02-15 at 5.05.39 PM](fractal_value.assets/Screen Shot 2021-02-15 at 5.05.39 PM.png)
When I first saw this data I thought a few things might be going on:
- Something structural has gone wrong in how we do/fund science, making it harder to do science.
- Science is working less and less on problems that turn into products, so it doesn’t show up in productivity numbers.
- Good ideas are harder to find, because the ones that were easier to find have already been found.
I think there could be elements of each of these, but I want to propose an alternative framing.