Dark mode switch icon Light mode switch icon

FAQ: Does automation lead to political polarization?

9 min read

Background: A long time ago, I wrote the FAQ section for reddit’s economics forum on automation and its effects on the labor force. I still sometimes get questions about it and sometimes go way over the top in answering them.
This is one of those cases, and I thought it’d be useful to share it.

Here is reddit user shellfish_bonanza’s question:

Hi VodkaHaze! Thanks for writing the article about automation.

One thing I read recently is that automation is not a Pareto Improvement: most people are better off, the workers displaced are not Source

Do you know what actually the data actually says happens to workers displaced? I’m trying to understand more.

The reason why I’m so curious is that in the linked paper and several others, it says that automation exposure is a key reason in political polarization.

So with the up to 50% of jobs exposure to automation in the next 10 years, I’m worried that if whoever is President doesn’t address this issue, the country will be even more polarized and the displaced workers/communities in shambles. This may lead to an even worse demagogue to take Trump’s place.Source

Thanks for taking time to respond!


Hey shellfish_bonanza!

Thanks for appreciating the article :) Do note that I’ve been meaning to rewrite it for over a year now, I think the framing is bad in some places and it lacks a lot of nuance in others.

One thing I read recently is that automation is not a Pareto Improvement: most people are better off, the workers displaced are not.

Note that almost nothing is a Pareto improvement – if it was a Pareto improvement we usually have implemented it already. Even if you take an existing product and strictly improve it, you make people profiting off the existing system worse off.

For example, you make a miraculous clean energy source that instantly fixes global warning. Even that isn’t a Pareto improvement, because you presumably made Charles Koch or the BP stakeholders worse off.

So if we’re talking about an improvement, we need to think about the distributional effects (how the gains of the improvement are distributed across actors in the economy). For instance, it seems like the gains in the western economy in the last few decades have generally had inequality-increasing distributional effects.

The Autor, Acemoglu and friends’ studies you link support that stylized fact: if jobs polarize between high and low skill, we’ve increased inequality even if the average income increased. Looking at the increasing gap over time between the average and median income is an easy way to spot-check the increasing income inequality.

Do you know what actually the data actually says happens to workers displaced?

For automation specifically it’s hard to say because it’s very heterogeneous depending on the business sector we’re talking about.

One way we can get a feel for it though is looking at a related topic: trade and immigration. Increasing free trade has a generally similar effect as automation does on the workforce: you can view trade with the new country as some part of the supply chain being automated. In economic terms, there’s a substitution effect and an income effect which will depend on the job and sector you’re in.

The bad news is that low income workers who get displaced don’t seem to do well at all in this case. Autor’s China shock paper is a famous study showing that rural US towns basically never recover.

The reason why I’m so curious is that in the linked paper and several others, it says that automation exposure is a key reason in political polarization.

In a sense here by “automation” we’re referring to almost everything causing productivity growth in the economy. That has little to do with, say, recent AI advances, and more to do with general economic trends. For example, a huge shipping company moving from manually-maintained excel spreadsheets to an integrated computer service is automating much more than the latest algorithm I deployed which increases conversion rates on advertisements by another 2%.

David Autor’s recent paper “Work of the past, Work of the future” goes over the economic trends related to education and rural/urban location. The picture it paints is fairly bleak – almost all economic gains in the workforce are going to college-educated urbanites.

Non Work Income

This picture is even worse if you consider non-work income. Specifically, urban landowners and capital gains will far outpace income growth, even for the privileged college educated urbanites who are still doing well in the current trends.

Since the productivity gains happen in the city, we see mainly rural -> urban migration. This is not necessarily a bad thing: metropolitan areas are more efficient in many ways than vast rural landscapes. However, if the supply of housing can’t increase in accordance with the increased demand from the emigrants, then housing’s price will increase.

This means two things for workers. First, housing will eat up bigger share of your salary, meaning your effective disposable income is lowered. Second, landlords get comparatively richer, so the inequality between renters and landowners increase sharply.

The solution to this problem is mainly political: we need to do away with restrictive zoning regulations, which enables building more housing, to reduce this problem.

Another trend that will emerge with higher automation is that the gains of capital will increase in comparison to the gains by labor. You can think of it in the microeconomics 201 Cobbs-Douglas model where a firm can use capital and labor to produce its goods. If we’re seeing an increase in worker displacement, what’s happening is that firms are substitution labor for capital. This means that the return to capital is higher in comparison to the return to labor –

The solution to this problem is through a better taxation system. Especially one that captures correctly the complexities of income for the very rich (who shift their income between corporate expenses, shares, bonuses and direct compensation on an international playing field to optimize for the lowest taxes). I’m not an expert on this topic, but there are approaches like a universal consumption tax which I see making sense. Either way, there needs to be large-scale structural changes to the tax system to reduce this form of increasing inequality.

So with the up to 50% of jobs exposure to automation in the next 10 years, I’m worried that if whoever is President doesn’t address this issue, the country will be even more polarized and the displaced workers/communities in shambles. This may lead to an even worse demagogue to take Trump’s place.

The good news is that the sort of recent political extremes (Brexit, Trump, Le Pen, Bolsanaro, etc.) we’ve seen aren’t only tied to economic reasons. Two non-economic factors I think are relevant are the voting systems we have in place and social media.

Voting Systems

It should be a surprise to no one that over the long run, different voting systems lead to different political equilibria. For instance, voting systems with a plurality rule system (like First-Past-the-Post in the US and Canada) naturally lead to a two part system over time. This is known as Duverger’s Law and is nicely explained in this video.

This has huge implications: the massive tribal split in the USA of every opinion, belief or behavior between Republican and Democrat is effectively just and artifact of how we designed the voting system.

Radiolab has an episode on the issue, talking about how Rank-Order based voting systems tend to lead to fewer political extreme candidates elected and much more positive electoral advertisement (rather than attack ads).

There are other problems in the on top of this, like gerrymandered districts and the distribution of electoral votes not keeping up with rural-urban migration. This all means very few votes effectively count in a federal election in a country like the USA or Canada: voters in marginal “swing states” hold everyone else hostage to their particular issues.

All in all, even though voting systems are generally seen as boring nerd discussions, they have massive effects downstream on how a country functions. It’s important to passionately argue for necessary change even though it has the emotional appeal of paint drying compared to immigration or crime issues.

Social Media

Even better voting systems don’t necessarily protect from populist demagogues. Brazil, for instance has a two-round system with proportional representation yet voted in crazy right wing conspiracy theorist Jair Bolsanaro in 2018. One reason for Bolsanaro’s popularity is his being boosted by the Youtube recommendation algorithm.

Voting system can’t protect against tainted beliefs in the electorate by efficient propaganda.

This is an issue of particular concern to me because I work as a data scientist, and my entire field of work is deploying black box algorithms at societal scale without understanding the implications (as describe by Cathy O’Neill in her book Weapons of Math Destruction).

In the last ~10 years, the delivery of media has almost entirely become filtered through opaque machine learning algorithms. Those algorithms optimize over some metric of the users’ behavior (likelihood to click, probability of commenting, time spent on site, or a mix of “engagement” metrics). Political extremes generally rank better on those metrics than the comparatively boring middle-of-the-road beliefs. This means algorithms will eagerly propagate extreme beliefs and filter out comparatively normal beliefs.

I’m not sure what the answer to this problem is, because it seems every social media service eventually develops strongly connected clusters of crackpots, and their recommendation engine is more eager to spread those than comparatively low engagement, “normal” communities.

Originally published on by Matt Ranger