Benchmarking AI productivity gains against smartphones, the internet, and the PC
Maybe we're all due for a dose of good news
Last week I proposed that AI is looking more and more like it’s going to be the platform shift in technology, and the first since the shift to mobile back in the mid-to-late 2000’s. Seven days later, this is looking more like an inevitability than just a theory.
While calling something a platform shift is interesting in theory, I’ve been wondering how you actually prove it. There are a few theories out there regarding this, but I’ve always found the most compelling ones to be backed up by hard data. While consumer usage and adoption is one compelling metric, I think the most notable and valuable metric always comes back to productivity.
When we think back to the greatest inventions in human history, they’ve typically all revolved around allowing us to do more with less: fire, the wheel, the printing press, and electricity have all raised the standard of living for humanity and helped to cover our basic needs so that we can spend our time elsewhere. What I consider to be the greatest invention in human history, language, enabled us to share our learnings with one another in order to be more productive.
In more recent years, as the standard of living has increased, breakthrough inventions have typically focused on enabling our workplace productivity, so that workers themselves can contribute more to our society. It’s not a perfect metric by any means (as has been well documented), but typically the cleanest and most comparable measure of quality of life / standard of living has been Real GDP per capita. In other words, the primary factor that drives prosperity is labour productivity.
Numbers! Data! Charts!
Below is a fascinating chart on US labour productivity that McKinsey pulled together in their latest productivity report (February 2023). What becomes immediately noticeable, is that there are clear peaks and valleys in labour productivity growth. While real compensation has typically grown less than productivity (as asset holders accrue more value than the working class), the two are certainly highly correlated:
The team at McKinsey has already done an excellent job annotating this data and providing some commentary on the trends through various economic eras. Interestingly, I don’t really see any correlation between recessions and labour productivity, so it doesn’t appear recessions themselves are much of a factor.
However, I thought it might be interesting to overlay the time periods of the last few technology platform shifts that we’ve discussed, to see if they can teach us anything about how platform shifts impact productivity gains.
First off, I’ve included the emergence of digital computing for the sake of completeness, but I think it’s probably best to ignore it. The 1940’s and 1950’s postwar boom saw productivity gains for a variety of reasons (repurposing of wartime factories, increasing automation, etc.), making it hard to ascribe all that value to the invention of the digital computer.
Secondly, it’s a bit of an inexact science to pinpoint when technologies started to really contribute to productivity gains. Based on very high level research, it seems like PC’s became commonplace in the 1980’s, most Americans started using the internet between 1995 and 2002, and most Americans adopted smartphones between 2010 and 2017. Again, not an exact science, but I think directionally correct.
If we highlight these ranges on the same chart from above, we end up with something that looks like the following:
There’s a lot of interesting trends going on in this data. First and foremost, it looks like there were very big increases in productivity in the last two years of the dataset (2020 and 2021). My understanding is that this occurred due to the productivity gains that emerged from employees working from home during the pandemic, which McKinsey also alludes to in their article. I think it’s safest to ignore these last two years as it’s hard to tell how productivity will normalize in the eventual 2022 / 2023 data.
Returning to the task at hand, it looks like the proliferations of the PC and internet coincided with periods of accelerating productivity gains, while the shift to smartphones coincided with a slowdown in productivity growth. It’s certainly too far reaching to say that these platforms shifts accounted for ALL the movement in productivity over these times periods, but I think it’s certainly a large factor that influences some underlying baseline. The McKinsey team sort of alludes to this later in their article, where they claim that there’s a 70% correlation between productivity growth and digital adoption across subsectors:
If we assume this correlation holds for this exercise broadly, one could interpret that technological innovation and adoption accounts for the majority of YoY changes in productivity gains. When we eyeball our annotated chart, that would means that PC and internet adoption each drove a ~200 bps increase in labour productivity growth, while smartphone adoption drove a ~150 bps decrease in labour productivity growth. The math here is all very high level, but it’s pretty interesting stuff.
Maybe smartphones and all that scrolling through social media actually made us less productive after all? It certainly feels plausible.
Guestimating AI’s impact on productivity
Returning to the task at hand, how do we expect the adoption of AI to affect worker productivity? The Wall Street Journal recently reported on a study conducted by researchers at the University of Pennsylvania and OpenAI, that claims 80% of the U.S. workforce could have at least 10% of their work tasks “exposed” to LLMs, while 19% could see at least 50%. Based on my read of the study, it sounds like the hurdle in which the researchers circled to be exposed was a “potential 50% reduction in time required to complete a specific DWA or task while maintaining consistent quality”.
The study goes on to state that 14-15% of all tasks in the entire dataset are exposed to LLM’s. If we conservatively apply a 50% reduction in time to all of the exposed tasks, that would already represent a 750 bps increase in worker productivity, more than triple the impact of what we saw occur from the adoptions of the personal computer and the internet (technically, this math is undercounting tasks in which more than 50% of time would be saved, and completely ignoring tasks in which less than 50% of time would be saved).
In researching this article, I also came across this report from Goldman Sachs that reaches similar conclusions. Based on Goldman’s analysis, they believe that the adoption of AI will increase labour productivity in the United States by ~150 bps per year. If we assume that it take AI roughly the same amount of time to reach mainstream adoption as prior technology platform shifts (which, per the sources above, seems to take approximately seven years), this would result in a 1,050 bps increase in labour productivity. Sounds like Goldman analysts are even more bullish on the impact of AI, which is a helpful gut check on our math. If we pull everything together and plot it on a single chart, we get something like this:
Needless to say, this would be incredible if true. Yes, it’s highly unlikely that AI solutions reach 100% adoption in the near term, and there are bound to be regulatory hurdles that slow deployment, but it’s certainly an exciting prospect in theory: even using our math that’s more conservative than Goldman’s, it seems very likely that AI will have a larger positive impact on labour productivity than the adoptions of the PC, the internet, or mobile.
The adoption of the PC and internet drove huge amounts of productivity gains, and therefore drove large increases in the global standard of living, global prosperity, and human innovation. If AI can even repeat the PC / internets level of success, there’s bound to be a wave of prosperity that follows. It’s important to be cautious about the dangers of AI, but I think that it’s also well positioned to make the world a much better place if leveraged properly.
The elephant in the room: the productivity-pay gap
As one could see in the first chart, real wages are certainly correlated with productivity gains, but they often don’t share in 100% of the upside. Since the 1980’s, we’ve seen a decoupling between the two in the US as the majority of productivity gains have flowed to asset holders, while labourers / the working class have fallen behind (chart per the Economic Policy Institute):
Most economists make the argument that this divergence arose from Ronald Reagan era (“trickle-down economics” or “Reaganomics”) public policy changes in the 1980’s that were made to suppress wage growth, but the divergence certainly persisted in the US post 1989 as well (in other G7 countries, this decoupling has always appeared to be much more muted). There are, however, early signs to be optimistic: it seems like real wage growth tracked productivity growth in both 2020 and 2021.
As any new technology proliferates and creates economic value, it’s important that everyone shares in the gains and benefits from increasing prosperity. If AI is currently well positioned to increase productivity growth even more than what we saw from the proliferations of the personal computer and the internet, this is more important than ever. Much of this will come down to regulation, subsidization, and reskilling programs as some jobs are more affected than others.
Where are we going from here?
While it’s easy to point to a Skynet / doomsday AI future, I’d like to think there’s potential for some very prosperous outcomes for all as AI removes the need for humans to complete some of the most manual, monotonous, and unenjoyable jobs out there (I’m looking at you, accountants).
Increasingly, there seems to be more and more global support for some version of Universal Basic Income as technologies like AI proliferate at warp speed. As the most unfulfilling jobs disappear, and if everyone’s basic human needs are provided by the government, one could certainly imagine a future where people are free to pursue more fulfilling, creative pursuits, ultimately leading to a new renaissance era for all of humanity.
Or, y’know… Skynet. TBD!
I wonder what contributed to the peak in the mid 60s. My guess is that the space war spurred many tech innovations that had positive ripple effects for productivity.