Experts have long imagined that blue-collar workers like truckers or factory workers would be the first to lose jobs to artificial intelligence. Former presidential candidate Andrew Yang suggested that a possible collapse of trucking industry jobs, caused by the perfecting and popularization of self-driving vehicles, even had the potential to cause civil turmoil as countless workers suddenly found themselves without a paycheck. Still, the idea never got a lot of media attention—perhaps because job loss to AI has largely been considered a working-class problem.
Eventually, some people surmised that those whose jobs revolved around pattern recognition, like radiologists, might also be replaced by a computer algorithm able to spot a tumor or a cyst with much higher accuracy than human eyes (though the prediction of imminent unemployment among radiologists turned out to be somewhat premature). Kai-Fu Lee, AI expert and the author of AI Superpowers, has gone even further, predicting that in the next few decades most doctors would find themselves relegated to a vestigial function as the “warm” human customer service representative for largely AI-conducted treatment, gently chiding MDs to work on their people skills. Numbers crunchers like accountants were also thought to be in some danger.
However most such analyses have largely ignored the dirty realities of AI implementation. The release of the ChatGPT tool by artificial intelligence company OpenAI last week should be prompting all of us to revise our ideas about the trajectory of AI’s impact on labor markets. Specifically, we should be rethinking the question of whose jobs will be first to the chopping block. (Spoiler alert: The answer is writers).
If you haven’t been following recent developments, let me fill you in.
ChatGPT is a terrifyingly robust chatbot, capable of spitting out thousands of words on any given subject based on your broad prompt. It’s equally adept at composing a version of “Bohemian Rhapsody” about a woman with a pathological fear of fish as it is writing business proposals, news items, middle-brow opinion pieces, or Wikipedia-style essays. While there are clearly some guardrails written into the software—ChatGPT will refuse to write legal contracts for example, and is also supposed to be incapable of generating racist or offensive content—they are very easy to circumvent.
Fool around with the chatbot for a few minutes, which is fun and easy to use, and it should become very clear that many of us who make our livings from writing are in grave danger of losing our jobs. Content marketers and paralegals are two that immediately come to mind here, but there are many other examples.
What should scare the shit out of anyone who writes for a living is how unbelievably flexible and easy to use this tool is. Why pay someone $30 dollars an hour to write inane search-engine-optimized marketing content, or a mind-numbing legal brief, if ChatGPT can do nearly as good a job in moments? (The chatbot’s work does require a little proof-reading; it seemingly inserts factual inaccuracies, perhaps intentionally in order not to be too scary to us meat puppets).
The key point to consider when thinking about what part of the job market AI implementation is likely to strike first is acceptable margin of error.
When it comes to jobs that take place in the real world, especially jobs that put the lives of others on the line—like driving a vehicle or searching an MRI for a patient’s tumors—almost no margin of error is acceptable. The slightest inaccuracy in the guidance of a self-driving semitruck has the potential to kill numerous people and cause massive damage. A missed tumor can be the difference between life or death for a patient. For AI to be able to take someone’s job in high-consequence occupations like these, the possibility of error has to be reduced well below the margin of actual human error. This means that these professions, and others that involve operating heavy machinery or making health care decisions, are likely to be safe—at least for a while.
Not so with writing. The stakes in allowing a typo, inaccuracy, or syntactical guffaw produced by an AI writer into a text are not very high. ChatGPT, off the shelf, is pretty much ready for use to produce low-level written content. Which means we should revise our projections for whose jobs will go to Skynet first by thinking about where AI implementation is the least risky. We thought AI would come first for Walmart shoppers, but it might well be us white-collar types with our jobs on the block.
That said, I don’t expect to see rioting at Trader Joe’s quite yet. For now, higher-tier jobs are probably safe. In its current iteration, ChatGPT is imitative, not inventive. People who write marketing content should be very worried; people who design broad-scope marketing campaigns and crunch analytic data into client presentations are probably safe for now. The paralegals who write the lengthy contract of lawyerly text are in danger, but the exponentially better-paid attorneys who then insert legal loopholes and clauses have a moat around their jobs for a while longer. AI can’t negotiate contractual provisions either. And while ChatGPT can debug code and make basic websites, the higher-level functions of computer programmers are still well beyond it. But the machines are coming.
What will this brave new world look like? It doesn’t seem probable that our government will launch a Butlerian Jihad in the style of Frank Herbert’s Dune to destroy the thinking machines and protect our jobs—if only because to do so would cede competitive advantage to companies in countries with looser regulations.
Furthermore, a reading of history suggests that the prospect of our leaders preempting the coming unemployment crisis by offering Andrew Yang–style UBI is equally unlikely. The introduction of social welfare programs almost always comes on a delay as a reactive measure to widespread suffering; indeed, I struggle to think of a single instance of government ever providing proactive aid in this way.
So if the machines are going to come for our keyboard-based jobs, what’s a person to do? Well, some experts say that skilled manual professions like plumbing may turn out to be resistant to automation. The varied situations a plumber has to work in, the sheer amount of different kinds of toilets they contend with, all pose a considerable challenge for AI, as Futures Today Institute CEO Amy Web pointed out in an interview earlier this year.
Will trade schools become the new coding bootcamps? Is it time for us English majors to pick up a wrench and look for a union job? Only time will tell. Meanwhile, if you’ll excuse me, I’ve got a deadline to meet.