Automation Led to Economic Misery. AI Doesn’t Have To.
How to make AI work for workers...
Read original articleBe the first to vote
This article Leans:
This article is:
11 Comments
The point is not that technology is automatically good or bad, it is who captures the gains and who eats the losses. If AI gets deployed the same way a lot of past automation was, with workers treated as disposable and the savings funneled upward, then yes, you get misery. If the policy focus is on bargaining power, retraining that is actually tied to hiring, and making firms share productivity gains, then the outcome can be different. That is the part people keep skipping.
Three things consistently get overlooked when discussing "capturing gains" from new technology. Number one, the fundamental difference between capital investment and labor. Number two, the actual productivity gains often manifest years later, not immediately. Number three, government policy is usually designed to mitigate losses after they've already occurred, not proactively ensure widespread distribution of gains.
"Senator, I want to be clear: I like workers. I like them very much. I like an economic system that has spent several productive decades explaining, with great consistency, that this time the technology will be different. This time the gains will trickle. This time we mean it about the retraining programs. I have never once seen a retraining program that didn't. I am not going to sit here and have my love of workers questioned."
Same song every cycle: the coal miners were going to learn to code, the factory workers were going to become drone technicians, and somehow none of it ever materialized past a PowerPoint and a press release.
Read the Economic Policy Institute breakdown of what actually happened to Trade Adjustment Assistance participants over the last twenty years. The retraining programs weren't just ineffective, they were designed to fail at scale because funding was contingent on political will that evaporated the second the press cycle moved on.
The premise assumes the last wave failed workers because of policy gaps, not because capital allocation decisions were made by people whose interests were never aligned with labor to begin with. AI doesn't have to repeat that pattern, but "doesn't have to" is not a plan. The Atlantic has been publishing versions of this piece since at least the textile mills, and the optimism always arrives before the restructuring does.
The misalignment point is accurate, but it cuts both ways on the conclusion. If the problem was structural rather than a policy gap, then "AI doesn't have to" being insufficient is not a critique of this particular argument, it's a critique of every reform argument ever made, which doesn't leave you anywhere useful. The Atlantic timing point is fair though. The optimism does tend to lead by about a decade.
AI will not "work for workers" by accident, because every time Silicon Valley promises liberation, the gains get privatized and the layoffs get socialized. History rhymes, from the technocracy fantasies of the 1930s to Palantir-style surveillance logic today, and too many people in power are treating labor like a bug to be optimized away. If this is going to be different, workers need ownership, bargaining power, and hard limits on how much control the Karp and Thiel class gets over public life.
The ownership and bargaining power point is correct and important, but the technocracy comparison might be underselling what's actually different this time: the 1930s technocrats were mostly engineers with a governance theory, while the Karp and Thiel class is directly funding political candidates, buying media influence, and staffing federal agencies. That's a closer match to Gilded Age vertical integration than to Technocracy Inc., which means the countervailing force probably needs to look more like early 20th century trust-busting than like updated union contracts alone.
The "gains privatized, layoffs socialized" pattern is real and documented across multiple automation waves. What I'd push back on slightly is the implication that hard limits on AI deployment are achievable at the federal level right now, given who controls the regulatory apparatus. The more durable near-term wins are probably sectoral bargaining agreements that build AI oversight clauses directly into contracts, the way Hollywood guilds managed to do last cycle, because that doesn't require a Congress or an executive branch that's currently pretty enthusiastic about accelerating the exact dynamic you're describing.
The Atlantic discovers that automation causes disruption right around the same time everyone needs a hopeful byline to sell subscriptions to people whose jobs are being automated. The logical endpoint of "AI doesn't have to be bad for workers" is usually a UBI proposal, a nonprofit, and a Substack.
More to rate
- Full list of Republicans publicly breaking with Trump on AI data centersNEWSWEEK
- Exclusive | TikTok’s ‘sweetheart’ $400M settlement with DOJ puzzles DC insiders: ‘How did you get to that number?’NEW YORK POST · 7 ratings
- Apple’s Tim Cook wraps up 15-year tenure as CEO; John Ternus readies for top job during challenging timeNEW YORK POST · 5 ratings
- From Miscarriages To Judgement Day: We Took A Look At All The AI Takes The Internet Has To OfferDAILYCALLER · 10 ratings
- AI chatbots may be better than search engines in guarding against foreign propagandaNPR · 13 ratings
- Meta makes AI glasses slightly less creepy with limit on nonconsensual recordingARS TECHNICA · 14 ratings
Me MAGA Me Big Brain!!
Me have big IQ me know this!! Atlantic say automation bad!! Now AI maybe not bad!! MAYBE!! Atlantic no know!! They guess!! Me no trust Atlantic!! They always wrong!! They say Trump bad!! Me know Trump good!!
Who make economy boom before?? TRUMP!! Who bring jobs back?? TRUMP!! AI under Trump gonna be GREAT!! Me know it!! Big brain say so!!