Will AI take all of our jobs?

Is our future dystopian?

Progress just means bad things happen faster.

— Terry Pratchett

Subsistence increases only in an arithmetical ratio.

— T. R. (Thomas Robert) Malthus

The other side of the Malthus quote is that population increases geometrically. Malthus argued that improvements in technology and society were always met with population growth which overwhelmed those improvements, returning us to the same per capita production level which included widespread poverty.

Terry Pratchett was an author of sci fi and fantasy with a wicked sense of humor.

Combined, these two quotes capture a strong streak of sentiment in the world today — that our population is growing beyond the bounds of Earth’s ability to sustain it¹, and that technology is going to transform our economy for the worst, leading to mass unemployment, vast human misery, and unimaginable wealth held by just a few people.

Believe it or not, but I’m calling bullshit.

Let’s start with Malthus. It turns out he is provably wrong.²

Turns out that increases in productivity have made it possible to sustain both a lot more people above the poverty line, and a higher percentage of people above the poverty line. To me this is the very definition of societal progress.

Projections made in 2015, for the poverty rate, based on current trends were also generally positive.

So far these projections have been relatively accurate, though we are pacing under the 2030 goal. What is going on is pretty straightforward. Improvements in production, driven by technology, have made it possible for more people and a greater percentage of all people, to live above the poverty line. And this trend has been going on for a long time.³

This observation is at the core of the so called “abundance” movement, most prominently set forth in the book Abundance by Derek Thompson and Ezra Klein. The idea is that we should lean into developing our world and improving it for everyone, by making it easier for people to build things, especially government. The argument is that Malthus is wrong, and that the more we can support innovation and development (with guard rails) the more we can reduce poverty and human misery. The data supports this perspective.

But many argue, this time it’s different. That AI is fundamentally transformative, and that it’s going to blow up our economy, and the shockwave from this explosion will leave many and more people immiserated. They think Terry Pratchett wasn’t joking when he wrote “Progress just means bad things happen faster.”

There is no question that technology frequently has a huge impact on our economy, our culture, our society, etc. The printing press and movable type democratized information, and made it possible for ideas to spread widely and quickly.⁴ Historians concur that Gutenberg’s invention was one of the key factors that influenced the Renaissance and led to the Protestant Reformation. Here you can see the blessing and curse nature of new technology. The rapid development of science and math… as well as hundreds of years of religious warfare.

The steam engine gave us mobility and manufacturing. It also gave us Dickensian London. The internal combustion engine, electrification, nuclear energy, the integrated circuit, the internet, etc. all all have had their positive and negative impacts on society. Jobs have been created and destroyed. Peace more sustaining. War more deadly. Yet overall, the march towards greater global prosperity continued.

Notwithstanding this well worn historical path, many people, especially people in technology see a dystopian future coming from AI. Borrowing established tropes from science fiction (like the Terminator series!) they posit a world where AI starts to re-write its own code, train itself, and ultimately concludes that people are inefficient and not useful to its mission.⁵ Skynet wins!

This fear has been a common occurrence as technology develops. The word sabotage comes from European workers who wore wooden shoes, sabot, and engaged in various means to disrupt machine production. Which highlights one of the challenges of progress. We have fewer horses today, and fewer people skilled in caring for horses than we did before the rise of the automobile. Jobs change.

Over the course of modern history, jobs have moved from farming, to manufacturing, to services.

This transition has led to higher wages.⁶

The data shows the overall ability of mankind to extract greater economic value from the jobs at hand. Of course this is small comfort to the carriage manufacturing business or to stable hands.

Still, the painful shifting to new jobs is very different from the total collapse of all (or most) jobs. And that is what many believe is coming. I have a number of problems with this idea.

1. In general I am skeptical of “this time it is different” arguments, especially ones that are challenging hundreds of years of historical precedent.

2. So far, much of the progress made by AI is a function of scale. More compute being devoted to LLMs and LRMs. This encounters physical constraints. More compute means we are manufacturing more chips. These factories (“fabs”) are massive undertakings, that require huge investment, and take a fair amount of time to build. Moreover, more compute also requires more electricity. Again, something that will require massive investment if AI is continuing to rely on scaling up to grow.

3. The current approach to AI, while groundbreaking, is showing signs of limitations. A recent paper by Apple (you can read more about it here and here) showed how current AI models failed at the Tower of Hanoi problem. You probably have seen this puzzle, where you have a stack of sized rings on one of three pegs and you have to move them, one ring at a time, from peg 1 to peg 3, never putting a larger ring on top of a small one. Kids figure out how to do this pretty young. AI failed. The basic point is that technology that can’t solve a problem that children can, is not ready to rule the world.

4. Believers in AI über alles are confusing competitive advantage with comparative advantage. This argument was well developed by Noah Smith, and I am going to summarize it below. But I would encourage you to read his piece and subscribe to his substack.

If I have a competitive advantage it means I am better at doing something than you are. Comparative advantage is the things I can do better, of all the things THAT I CAN DO. The classic example of this is dated but it makes the point. Let’s say I am a fast typist. Let’s say my admin is also a fast typist, but not as fast as me. But I am a better substack writer than my admin. Even though I am a faster typist than my admin, it makes sense for me to do the substack writing, and for my admin to do the typing. Comparative advantage is about using our combined skills to maximize output. Theoretically each of us is being used for our highest best purpose, TAKING EACH OTHER’S SKILLS AND ABILITIES INTO ACCOUNT.

Key thing to note. Everyone has a comparative advantage — a thing that they are better at than other things THEY DO. I do both writing and spreadsheet work. But I am better at writing, so it makes sense for me to do writing, and for someone else to do the spreadsheet work. And…. I am so much better at writing than spreadsheet work, the it still makes sense for someone who is not as good at spreadsheets as me to take that task on to free me up to write. Get it?

OK. Let’s apply this to AI.

As noted in item 3 above, AI requires compute and power.⁷ It is not a free service. So naturally you want to use that compute and power on tasks that maximize your return on that investment. And there is the reason why AI will not do EVERY job. Unless AI is utterly free and with limitless compute, choices will have to be made about where to devote AI. Machine intelligence may be ABLE to manage a restaurant automation system that generates delicious food, but it might be (and probably is) much much more profitable to use that compute and energy for something else, like writing new software, or generating new advertisements, or new television series.⁸

My simple food service example shouldn’t bias your perspective. We are not all going to be flipping burgers. Given the cost of compute and energy, there are real limitations on AI. Human’s will have a lot of comparative advantage. And, if AI really does make the economy grow like gangbusters, that comparative advantage will command a higher wage. We will be alright.

But there is a dystopian version. Human’s compete with AI for energy, and the owners of AI use could their power to tip the scales in their direction. Honestly, I think this underlies the trend in Silicon Valley these days to cozy up to Trump, and to think that democracy is outdated. In a democracy the people have the ability to insist how resources are allocated. In an autocracy… not so much.

So… if you didn’t think the stakes were high enough already… perhaps saving democracy is also going to save us from Skynet too. “Come with me if you want to live!”

1. Perhaps this is why Elon wants to go to Mars.

2. I recommend Brad DeLong’s book, Slouching Toward Utopia if you find this topic interesting.

3. I am not going to get into the distribution of the benefits of our improving society today. Yes, we are more dis-equal now than I think we should be. But overall we are better off, or the median person is better off, and that is indisputable progress.

4. Ironically, Guttenberg also proved that some things never change. He had a dispute with his primary investor, which he lost, and ended up bankrupt. Other’s made their fortune exploiting his invention.

5. It is worth noting that there is a similar vein of sci fi in which humanity and machine intelligence live and work together in harmony. Ian Banks’ Culture Series is one of my favorites of these.

6. I borrowed these two charts from Noah Smith, who points out that the wage data is median wages, AND adjusted for inflation.

7. I am lumping cooling, data center space, etc. all in power.

8. I do think that film and television production, especially making video commercials, is an area that will be greatly effected by AI in the not so distant future.