AI Job Losses Are About Whole System, Not Just Individual Industries. Or Follow the Money and the Power

Timothy Lee at Understanding AI has a generally positive take on how translators have dealt with AI intrusion into their workplace that I think largely misses the mark.

Lee does not think that AI is going to be disruptive enough fast enough to significantly damage working people as compared to the benefits AI will bring to working people as consumers. Lee uses the translation market as his proving ground and rests his contention on three points: AI is still not good enough to handle a significant portion of technical translations (law, medical, etc.), even at the low end; AI still needs human help and so people will be present to assist even if their wages go down; and automation benefits consumers and since workers are also consumers, they benefit from automation as well as suffer from it. None of these points are compelling on their own and they miss the larger issue: how AI distorts the political economy, which in itself worsens the lot of ordinary people.

Lee’s first point is his most compelling and he makes a strong case that there are simply things that AI cannot translate in any meaningful way. He talks with several experienced translators who make a strong case that in certain technical fields, AI currently lacks the ability to understand context well enough to translate them. That means that for the foreseeable future, those translators will have access to those relatively well-paying jobs. AI has not in fact moved that far up the human intelligence chain (because it’s not intelligent and it cannot learn, but that is an ongoing rant for another post). But Lee admits these are a minority of translation jobs and require a high level of training and specialization. In other words, it is a continuation of the worsening income inequality — a few winners and a vast pile of losers.

What of the more common translation jobs?

Lee seems to think those are under more threat, but still not completely endangered. In his words, the jobs are cheaper but faster, so translators get paid less but can do more jobs. In theory, this can help them make up their losses. In practice, however, Lee admits that his hasn’t really worked out. By Lee’s own numbers, median translator wages dropped by about eight percent between 2010 and 2021. He also argues that the change is happening so slowly that people have time to adjust. That might have been the case ten years ago, but this Washington Post article shows that many copywriting jobs are being swallowed by ChatGPT and other tools almost overnight. None of this sounds like the gradual transition Lee thinks is happening.

Finally, Lee argues that automation is good because it inevitably helps consumers. He argues that more poor translations, for example, are better than fewer better translations, and that the lower prices of Uber and Lyft are better for consumers than the higher prices taxis used to charge because of their specialized navigation skills taxi driving required before GPS.

Note the sleight of hand there? He asserts that workers are consumers and benefit from automation but then doesn’t discuss how automation worsening working conditions and pay for workers is offset by lower costs in those areas for those workers. That is probably because he cannot.

Increases in productively have famously not translated to increases in pay for most workers. Recent economics research has shown pretty clearly that automation has driven increases in economic inequality and actually destroyed jobs, not created new jobs. Which highlights the largest problem with Lee’s position — it ignores the systematic effect of AI driven automation.

As the links above show, the primary effect of automation in the last few decades has been to concentrate wealth away from working people. It does not do that by itself, of course. Political economy is complicated, but it certainly contributes. But the trend is real and undeniable. As a result, more power is concentrated in the hands of capital, making it less likely that we as a society will take care of the people left behind by these trends.

These trends have been evident for years, decades, and yet we have done nothing about them. We could have universal health care, but we do not. We could have a jobs guarantee, or a universal basic income (something that has been shown to increase social welfare and not decrease work everywhere it has been tried), but we have neither. Heck, we even let a minor child benefit for parents expire after one year despite its obvious poverty fighting effect because of nonsense about encouraging dependency.

All of this is in large part because our politics are heavily influenced by money. And thus the more money we shift from working people to capital, the less likely we are to takes steps to actually mitigate these issues. Even if Lee’s optimistic takes are correct, he is still missing the actual dangers. Because even he admits that AI is likely to be widely disruptive in the next five to ten years. If it is, the U.S. is not well positioned to deal with the consequences if we wait until they upon us. By being sanguine about the AI future, by ignoring how these trends affect the politics economy, we make it much more difficult to deal with the fallout for real people. Because capital won’t want to give up any of its power.

The problem with AI is not just how it affects writers, or translators, or artists, or any given sector of the economy. The problem with AI is how it distorts the entire politic system. You cannot understand the real potential dangers and impacts unless you are looking at the whole system. In theory, automation could lead to robotic socialist utopia. In practice, it concentrates the money and power in the hands of people who are more interested in neo-feudalism. Ignoring that misses the robotic forest for the artificial trees.