Imagine a man whose desire for an easy life is stronger than his sense of ethics. And imagine that this person has a cutting-edge computer program that can quickly answer complex questions. Then imagine that person is presented with a difficult question. Instead of answering himself, he taps into the computer, then relaxes for a while. Finally, he relays the computer’s response and confesses to the many hours of hard work he didn’t do.
That’s a pretty good description of a 12-year-old boy I know who entered a homework question into ChatGPT, played on his Xbox all evening, and then turned in the computer work to a teacher who graded it perfectly. “Outstanding effort,” was the teacher’s comment – which, when you think about it, is true.
It’s also a good description of how at least one accountant reacted to one of the first digital spreadsheet programs, circa 1980. As Steven Levy reported in his 1984 In the Wired article, A Spreadsheet Way of Knowledge, this accountant, when given an urgent assignment, sat down with my micro and spreadsheet, finished it in an hour or two, and left it on my desk for two days. He then pitched to a former client and received all sorts of accolades for overtime.
ChatGPT is unparalleled in its ability to generate reliable answers to a wide range of questions. But it has very clear precedents in other ways, from the shearing frame to the calculator and satellite navigation. Those precedents give us some clues about what might happen next.
The first insight is that if a technology works well enough, it can be quickly adopted. I’ve often written about how it took more than three decades for the electric motor to catch on. It took a lot of rethinking, retraining, and retooling before plant owners could unlock its benefits.
But not every technology requires such epic transformations. The digital calculator has taken the business world by storm in about five years. It was just too good and too easy to use compared to the handwritten alternatives.
Second, new technologies do not necessarily destroy jobs, even in the industries that are most affected. The Planet Money podcast estimates that since 1980 (around the time digital calculators first came into commercial use) by 2015, accounting lost 400,000 jobs in the US and gained 600,000. The jobs lost were often accountants whose role it was to grind arithmetic through calculators. The jobs that were acquired were for more—dare we say it? – creative accountants.
But it’s the third insight that intrigues me the most: different technologies tilt the playing field in different directions. A calculator multiplies the skills of an advanced user, but satnav is different; it is an alternative to competence.
The cutting frame has turned the lives of skilled textile workers upside down, as almost anyone can achieve the complex, highly skilled task. Its use was despised by the Luddite rebels because, like satnav, their knowledge had become redundant.
The digital spreadsheet is an example of “skills-oriented technological change” that helps productive people be even more productive. For about half a century, skill-based technological change has been the norm and an important reason for decades of rising income inequality. However, as satellite navigation and the shear frame show, some new technologies are increasing the productivity of less experienced workers. This will not automatically reduce inequality – the shear frames may have helped unskilled workers a little, but they have mainly benefited capitalists.
So what about generative AI systems like ChatGPT and Bard? Do they increase the output of elite workers, or do they provide the most help to those in need? It’s too early to be sure, but the early evidence is intriguing.
One study by economists Erik Brynjolfsson, Danielle Li and Lindsey Raymond, investigated what happened when an AI-powered chat assistant was added to a team of more than 5,000 customer service agents at a software company. These employees typically engaged in lengthy text conversations with frustrated customers trying to resolve technical issues. Meanwhile, the chatbot would scan the conversation and suggest possible responses for the customer service agent to use, ignore or adapt.
Brynjolfsson and his colleagues found that chatbots helped, with employees solving slightly more customer problems and doing so 14 percent faster. Also, the chatbots were not skill-oriented: the best, most experienced agents had no benefit from the chatbot, while the least experienced and skilled agents handled 35 percent more inquiries per hour. Those inexperienced workers also learned and improved faster than those who didn’t have access to the chatbot.
Another study by economists Shaked Noy and Whitney Zhang, gave people writing assignments. Half of them had access to ChatGPT, half didn’t. Again, the least skilled people enjoyed the greatest benefits. The Homer Simpsons of the world, long estranged from technology, may finally find invention on their side.
I am still concerned about the damage new generative AI systems could do to our already damaged information ecosystem and the upheaval they could bring to the world of knowledge work. But I’m also encouraged by the glimmer of hope that they can improve the working lives of some long-marginalized people.
Homer Simpson famously proposed a toast: “To the booze!” The cause and solution to all life’s problems. The residents of the ChatGPT house may soon feel the same way.
Tim Harford’s children’s book The Truth Detective (Wren & Rook) is now available
Follow up @FTMag Twitter to be the first to hear about our latest stories
“Bacon fanatic. Social media enthusiast. Music practitioner. Internet scholar. Incurable travel advocate. Wannabe web junkie. Coffeeaholic. Alcohol fanatic.”