There’s been a lot of talk in recent weeks about a “white-collar blood bath,” a scenario in the near future in which many college-educated workers are replaced by artificial intelligence programs that do their jobs faster and better.
Dario Amodei, the chief executive of the A.I. company Anthropic, recently predicted that half of entry-level positions in fields like law, consulting and finance could meet this fate in just a few years. Mark Zuckerberg, the chief executive of Meta, has predicted that A.I. will replace many of Meta’s programmers within the next year or two.
Optimists push back with a different prediction, forecasting that A.I. won’t replace white-collar workers but will rather serve as a tool that makes them more productive. Jensen Huang, the chief executive of the computer chip maker Nvidia, has argued that “you’re not going to lose your job to an A.I., but you’re going to lose your job to someone who uses A.I.”
Both sides in this debate are making the same mistake: They treat the question as one of fate rather than choice. Instead of asking which future is coming, we should be asking which future we want: one in which humans are replaced or only augmented?
The decision will depend on companies like Anthropic and OpenAI and whether they want to build ethical, sustainable technologies — as they say they do. In that case, they should design A.I. that works hand-in-hand with humans, rather than trying to build autonomous systems to replace us. Equally if not more important are the employers who adopt A.I. systems: If they really want productivity gains, they too must embrace A.I. programs that augment rather than replace.
The distinction between augmentation and replacement can be subtle. Any technology — from the stone ax onward — replaces some human work in the course of augmenting it. The key question is whether the tool enhances our abilities while still leaving us in control of how to use it. As Steve Jobs once put it, a computer can be “a bicycle of the mind.” Just as a word processor allows writers to write without having to laboriously correct and retype manuscripts, so too A.I. should help humans devote ourselves to our most significant and interesting challenges.
A simple example is Copilot, an A.I. tool offered by GitHub, an online service for computer programmers. When you write a line of code, Copilot suggests the next line of code, as with the “autocomplete” feature in texting. That can increase efficiency (when the suggestion is indeed what you had in mind) and spark new ideas (when the suggestion isn’t quite what you had in mind), while keeping the human user in the driver’s seat.
Another example is the A.I. technology that helps radiologists do their job better by refining images, automating routine tasks and flagging potential medical abnormalities. These are not autonomous systems that replace humans — as Geoffrey Hinton, a prominent A.I. scientist, predicted in 2016 they soon would be — but rather tools that make medical professionals better at their jobs.
The augmentation approach to A.I. isn’t just about preserving jobs (though that’s important, too). It’s also about keeping human interests central to our future. That may seem like an obvious goal, yet it is alarmingly easy to lose sight of. All systems are vulnerable to mission drift — when they gradually, often imperceptibly diverge from their original purpose — but the risk is compounded if we allow autonomous A.I. systems to run the show.
There is also a strong practical argument for the augmentation approach: It helps maintain the quality and performance of the A.I. systems. For one thing, A.I. systems rarely work well without human involvement. Consider the financial-technology company Klarna, which last year boasted of automating its customer support, only to reverse itself this year and begin hiring back humans.
In addition, most A.I. systems are trained using data scraped from the web, much of which was created by humans. But if humans are cut out of the loop, A.I. will increasingly be using data drawn from other A.I. systems. This is a problem: As recent scholarship shows, when A.I. trains on the output of other A.I., the quality of the outputs can quickly degrade, a phenomenon known as model collapse.
That means, for example, that fully autonomous A.I. “journalists,” when posting to the web, would gradually pollute the data supply. The problem could be managed by humans, but unmanaged it would be likely to undermine any gains in journalistic productivity.
The risk of degradation goes both ways: Humans can atrophy, too, if they lack meaningful work. Numerous studies of the unemployed and of towns where factories have closed show major declines in mental and physical well-being.
Augmentation is not any kind of panacea: In yielding greater efficiencies, it will lead to job loss. But that was also true of augmenting technologies like the tractor and the personal computer, innovations that were worth their disruptive trade-offs.
There’s a world of difference between productivity gains and a systematic plan of industry-eliminating unemployment. Indeed, as the economist David Autor argues, A.I., done right, could help rebuild the middle class by giving a broader range of workers access to expertise such as software coding that is currently concentrated among higher-skilled workers. It all depends on how the technology is implemented.
In a classic 1960 paper, the computer scientist J.C.R. Licklider wrote that the present goal of computing should be “man-computer symbiosis.” This remains correct. Augmenting humanity has been the aim of technological development since at least the Bronze Age, and that should continue to be our goal even as we develop technologies that mimic human intelligence. The wholesale replacement of human work makes for good science fiction but a bad future.
Tim Wu (@superwuster) is a law professor at Columbia, a contributing Opinion writer and the author of the forthcoming book “The Age of Extraction: How Tech Platforms Conquered the Economy and Threaten Our Future Prosperity.” He served on the National Economic Council as a special assistant to the president for competition and tech policy from 2021 to 2023.
The Times is committed to publishing a diversity of letters to the editor. We’d like to hear what you think about this or any of our articles. Here are some tips. And here’s our email: [email protected].
Follow the New York Times Opinion section on Facebook, Instagram, TikTok, Bluesky, WhatsApp and Threads.
The post A ‘White-Collar Blood Bath’ Doesn’t Have to Be Our Fate appeared first on New York Times.
