A Response to OpenPhil's R&D Model

A few stray blurbs in critique of “Social Returns to Productivity Growth”, because they don’t have a comments section...

The Author concludes that R&D may not be worth funding, in general, for being many times below the Give Well threshold of “10x cash transfers.” There are a few problems with the toy model used, which change the result by more than an order of magnitude:

Estimate of the author: “R&D investment adds 1% to growth each year, from $2 Trillion spent.”

This is likely far below reality, for two important reasons. First: as time goes-on, we use-up the most readily available sources of materials. Those latter reserves are more expensive and difficult, if we are using the pre-existing technology. Yet, prices of raw materials are not impacted as much, because of R&D that improved that processing. The Author uses Total Factor Productivity, which is strictly “Outputs per Input”, and this does NOT account for inputs becoming successively more difficult to obtain, over time.

As an example: the oil wells tapped in the 40s and 50s were Saudi, hugely productive at any price, and readily available with the technology of the day. Now, we need to drill miles into the ocean floor, which was impossible without R&D. So, instead of asking “how much more output now, per input, compared to the past?” We must ask: “Using the PAST’s technology, how much output per input WOULD we be generating, given TODAY’s reserves?” That value is far lower—productivity will decay, without R&D to keep it treading water.

What would be a fair estimate of this Input-Inflation? Consider that 2% implies a 35-year doubling: “If we wanted to extract this modern, hard-to-get-oil using tech from 35 years ago, it would be twice as capital-intensive per barrel of oil as we are doing now, with modern tech.” That 2%, added to the Author’s 1%, triples the estimated return. Let’s address the other half of productivity, too:

Second, the Author gave a nod to the idea that “Spending $60K/yr on 1990′s products at 1990′s prices is not as good as $60K today, so the income growth from $40K/person in 1990 to $60K today is underestimating the value of R&D.” R&D often reduces the price of high quality, to the extent that a homeless person has access to the internet on their phone that is many times faster than the best dial-up modem from the 90s. Granted, making tech 10x cheaper does NOT make the user 10x more productive… yet, leaving it out entirely, as the Author does, is an enormous oversight for one particular reason: the Threshold of Viability.

As a technology deflates, providing more and more value at lower cost, then it becomes possible to create new businesses—even new business MODELs, built upon these prices. When the Author credits R&D with only 40% of growth, they are leaving-out the entrepreneurship and process-innovations that only became POSSIBLE with lower priced tech. I doubt we’d have so many new, valuable companies, if we never made the internet so cheap and ubiquitous. AI is possible with cheap GPUs, right? TFP doesn’t measure any of that. In essence “all the other TFP growth came from R&D in the past that created the viable opportunity.” So, that’s another 2 to 2.5x multiple.

The other major sticking-point in the Author’s assessment is the idea that “because new research is getting harder, roughly 2x for every 30% of economic growth, then productivity growth with flatten, eventually hitting zero, eliminating many of the lasting gains of research today, as opposed to research tomorrow.” That discounting of future productivity was what gave R&D an estimated “4.5% GHW threshold”—while “Scenarios 2 – 4 can multiply the bottom line by up to ~7X, or much more for some versions of scenario 4.” Well, 4.5% multiplied by 7 is 31.5% of the GHW threshold, which is 3x better than cash transfers. And, if the other factors above are included, then R&D as a whole category without any selection for benefits to the poor would still exceed the mark!

What is that “Scenario 4”? 4: “Maybe some trend-breaking future technology will allow us to avoid growth stagnation despite ideas getting harder to find?” The Author focuses upon the possibility of AGI granting us infinite research, eventually, and calls the entire category “unlikely scenarios with large upsides”. (emphasis mine) This is a gap in their creativity: even without Aritificial General Intelligence, the AI already deployed by Waymo and Amazon are good-enough for Robot-Only mining of Near Earth Asteroids. We don’t see it happening yet, because it takes a few years each to design and fund and prototype.

According to the Author’s own model of R&D’s diminishing returns, we still have a few centuries left before we hit a Zero-Growth wall—so, their claim that asteroid-mining is an “unlikely scenario” hinges upon NOBODY mining the asteroids that whizz between us and the Moon at any time in the next few centuries. The Author’s assumption sounds like the unlikely one.

So, if we have Asteroid Mining, then capital->capital autonomously, and productivity can continue to grow without ANY change in technology. That would give the 7x valuation the Author mentioned, above. Combine that with 2x for R&D fueling new entrepreneurs, making it possible to measure and eliminate misallocations, and making greater capital accumulation viable, mentioned above. Then, multiply by 3x for the gains which are hidden by resource-depletion, which I had explained first.

Combined, the original estimate of “4.5% GHW threshold” becomes 1.89 times higher than that GHW threshold. If you narrow R&D down to “projects which are likely to predominantly benefit the poor,” then there is a solid case for charitable research.

A Response to OpenPhil’s R&D Model