Lost in automation?
Caption

Lost in automation? / Getty Images

Earlier this year, a drumbeat of news headlines played into public anxieties about the safety of human jobs when Duolingo, a language learning app, became a prominent example of a company cutting workers and replacing them with artificial intelligence.

The most eye-catching job cuts were those for translators, who worked on some of the company’s less popular language education courses. Translators and interpreters are often near the top of media listicles as the jobs most likely to be killed by AI. When the stories about Duolingo’s job cuts circulated, they seemed to confirm that the inevitable AI jobs apocalypse had arrived.

In a recent conversation with Planet Money, the CEO of Duolingo, Luis von Ahn, downplayed the meaning of the cuts. It wasn't full-time employees. It was only 10% of their contractors. His company’s recent embrace of generative AI only played one part in the decision, and so on. More interesting, considering Duolingo’s official partnership with OpenAI, was von Ahn’s reaction to the company’s recent demonstration of its newest version of ChatGPT, GPT-4o.

In its live-streamed demonstration event announcing the launch of GPT-4o last month, OpenAI showcased how good its popular chatbot is at translating languages between people in real time. It showed two OpenAI employees, one speaking Italian, the other English, with a ChatGPT app on a smartphone audibly translating a conversation between them. The demo was short, with the employees asking and answering a single question: “If whales could talk, what would they tell us?” ChatGPT — not surprisingly, given this was a public marketing event — seemed to do a good job.

“It’s funny that was the demo,” von Ahn says. He says Google Translate could have done a similar demonstration like 8 years ago. The reality, he says, is that computer translation between the world’s major languages has already been “really good” for quite a while.

Indeed, AI has been supercharging the abilities of machines to translate foreign languages for close to a decade or more — which is why it’s an interesting case study of the potential effects AI will have on the job market. Contrary to some of the doomsayers, the great AI massacre of translator jobs has not arrived, including even at Duolingo. It’s proving hard to fully automate away translation jobs. But why isn’t AI killing these jobs? And, even if it isn’t, how is it reshaping them?

AI Hype

As far back as 2006, when Google launched Google Translate, the translation industry has been “speculating about the potential for AI to replace human translators,” says Bridget Hylak, a representative from the American Translators Association, the largest professional organization for translators and interpreters in the United States. “Since the advent of neural machine translation (NMT) around 2016, which marked a significant improvement over traditional machine translation like Google Translate, we [translators and interpreters] have been integrating AI into our workflows.”

So, yeah, translators have been grappling with AI for a while. Yet, despite the fact that anyone with a smartphone has long been able to use this machine translation technology for free or at a relatively low cost — there are still a ton of jobs for human translators and interpreters out there.

In fact, according to the US Bureau of Labor Statistics (BLS), the number of jobs for human translators and interpreters grew by 49.4% between 2008 and 2018, thanks largely to globalization. After 2018, BLS changed how they collect and measure occupational data, which makes it less reliable for measuring job growth over the last few years.

However, data from the Census Bureau, which began tracking growth in this occupation in 2020, shows the number of people employed as interpreters and translators grew 11 percent between 2020 and 2023. (Thanks to Sofia Shchukina, our new Planet Money fellow, for helping us sift through and crunch all the numbers!)

The reality is that, despite advances in AI, jobs for human interpreters and translators are not cratering. In fact, the data suggests they’re growing.

Tons of businesses and governments are currently hiring translators and interpreters. Honda, for example, is currently hiring a Japanese interpreter/translator for its factories in South Carolina. Starplus Energy, a manufacturer of batteries for electric vehicles, seeks multiple Korean interpreters/translators for their plant in Kokomo, Indiana. The City of San Francisco seeks a “Bilingual (English-Spanish) Translator/Proofreader and Phone Operator.” Languars Inc wants a “French Medical Interpreter.”

In fact, BLS projects the number of jobs for interpreters and translators will grow by about 4% over the next decade. While that would represent a slowdown from the tremendous job growth the industry saw over the last couple decades, it is still actually slightly faster than the average growth BLS projects for all existing occupations in the US economy.

So, if AI has gotten so good, and so good at translation in particular, why are there still so many jobs for translators and interpreters?

“Well, I don’t think it’s that good,” says Daron Acemoglu, a superstar economist at MIT who studies AI. “I think how good AI has become is often exaggerated.”

Acemoglu has a new academic paper out that sort of throws a wet blanket over the fiery excitement for AI. Sure, he says, AI can do some amazing things. “But there is pretty much nothing that humans do as meaningful occupation that generative AI can now do. So in almost everything it can at best help humans, and at worst, not even do that.”

Acemoglu says he believes translation is “one of the best test cases” for AI’s capability to take over human jobs “because, I think if it can do anything, it’s translation.” But, he says, even in this realm, the technology is just “not that reliable.”

Why AI Didn’t Kill The Translation Star (At Least Not Yet)

For a more bullish take on AI, back to Duolingo CEO Luis von Ahn. Von Ahn, like many other technologists, sees AI ushering in a dramatically different world. It, for example, is making his company’s mission of teaching people foreign languages with an app more effective by enabling users to have rich, improvised conversations with an interactive chatbot.

However, even von Ahn acknowledges that the technology is still limited. That’s why, despite recent headlines suggesting otherwise, his company still employs translators. “It’s still the case that computers make mistakes,” von Ahn says. “I don’t think you wanna fully rely on a computer if you're a translator for the army and you're talking to an enemy combatant or something like that.”

Duolingo, von Ahn says, still uses human translators to double-check that machine-generated translations don’t make mistakes in the company’s learning content. But, he says, translators at his company mostly work on more high-value aspects of the business, where the extra cost of employing a human is really worth it. “If it’s things like the user interface of Duolingo, where a button on the app says ‘quit’ or ‘purchase now’ or whatever, that translation is all done with humans. We spend a lot of effort on that because each one of those features is highly valuable. We just cannot have a mistake.”

And it’s not just about mistakes, von Ahn says. The company also uses human translators to ensure consistency in the company’s style and tone throughout their app. Turns out, AI can’t consistently master “the same playful voice” Duolingo wants to communicate to users. So, for that, von Ahn says, “we still employ humans.”

Daniel Sebesta, another representative from the American Translators Association, suggests this is a common reason why companies and governments still employ human translators. “AI still struggles with complex linguistic tasks that require creativity, cultural sensitivity, and the ability to understand subtle nuances in meaning, especially in low-resource languages (i.e., languages that don’t have millions upon millions of high-quality translated words that can be used to train AI),” Sebesta says. “Companies continue to hire human translators and interpreters because they understand that AI cannot replace the expertise and judgment that these professionals bring to the table. This is particularly true for high-stakes projects in fields like legal, medical, but also literary translation, where accuracy and cultural appropriateness are paramount.”

In realms where mistakes could mean lawsuits, embarrassment, injuries, or even deaths, it makes a lot of sense why so many companies, non-profits, and government agencies still want humans overseeing and editing AI-generated translation and interpretation. There’s also considerable demand for human translators and interpreters due to regulations. “In the United States, the Title VI of the Civil Rights Act of 1964 bars discrimination based on language, so some entities — like courts and schools — are simply mandated to provide language services,” Hylak says.

"Despite the widespread use of translation software, having a human expert in the loop is still necessary to ensure reliable and accurate translations,” says Javier Colato, an economist at BLS. “Human translators will also be needed to handle more complex translations, such as technical documents and literary works. Therefore, considering the strong underlying demand for translations and continued need for human translators, some employment growth is still likely for the occupation."

The Wages Of Cyborg Translators

Everyone we spoke to stressed that these days, human translators and interpreters use AI as a tool to become much more productive. “We see a future — for many, in fact, a present — where AI-powered tools and human translators/interpreters collaborate, with AI handling more routine tasks and humans focusing their cognitive effort on the more creative and nuanced aspects of conveying meaning,” says Sebesta.

Von Ahn says he believes this human-machine collaboration in translation is one reason why demand for translation services is so strong. “What you’re seeing today, for translation in particular, is this combo, this hybrid between humans and computers,” von Ahn says. That has made translation a lot faster and cheaper, and, as a result, he says, “there’s a lot more demand.”

So, great, more demand for translation services as they get cheaper. And AI is proving, at least so far, incapable of doing much of this work without an important role for humans. But that doesn’t necessarily mean the humans doing these jobs are thriving in this changing translation economy. AI automation of much of their work may, in fact, be devaluing their skills, since, thanks to the assistance of machines, more people can do more translation better and faster.

Acemoglu’s research suggests that the effect of automation on wages is, well, complicated and not universal. Sometimes automation can enrich workers. Think doctors who no longer have to spend as much time on paperwork thanks to computers. Instead, they can focus more on their bread and butter skills of treating patients. These skills are scarce, in demand, and therefore very valuable, and focusing more on them, doctors can be more productive and get even richer.

But other times automation can hurt an occupation’s wages by devaluing its core skills. Even if automation doesn’t kill the job, maybe what was a high-skilled job in the marketplace becomes a more lower-skill job as machines enable a whole lot more people to do it.

And, sure, these now-low-skill workers may be way more productive than they were before technological advances. But, Acemoglu stresses, this doesn’t mean they’ll necessarily share in the fruits of that productivity. Factory owners — or the owners of AI algorithms — may get all the money. Historically, Acemoglu’s research suggests, workers have had to turn to strikes, unionization efforts, or elections of pro-labor politicians to pass policies like minimum wage laws to share in the new riches created by machines and increase their standard of living.

Data from BLS — which is usually the best data source for stuff like this, but again, might not be well-suited to track changes over the last few years — suggest that, if anything, the wages of the typical translator and interpreter are actually growing. As of 2023, the typical interpreter and translator made $27.45 per hour, or about $57,090 per year, which is slightly higher than the typical pay for all American workers (about $48,000 per year).

When it comes to incomes, Sebesta foresees a growing disparity between translators who master AI and those who don’t. “The incomes of the former group will rise and the practitioners will feel empowered,” Sebesta says. “The other group will likely feel left behind and exploited and will miss out on the opportunities.” It’s why, he says, he sees his organization, the American Translators Association, as having an important mission in helping translators adapt to technological change and thrive in the age of AI.

Acemoglu, the MIT economist, glancing at the economics of the translation business, believes that the incomes of most translators and interpreters will likely take a hit as technological change sweeps the industry. For him, it boils down to the laws of supply and demand. If AI enables a flood of translation supply, that likely means the price of translation goes down. Translation services get cheaper. Good for consumers. Probably bad for the incomes of many translators. Although, he says, maybe more elite workers in the profession — like translators of books or high-level interpreters working in diplomacy — will be exempt from this downward pressure on their wages.

But, even if this scenario manifests itself, it would not mean an existential threat for the jobs of most human translators and interpreters anytime soon.