Article · AI translation
IT Home, November 2: Artificial intelligence is not going away. How should we coexist with it? More precisely: how will we make a living in the age of AI?
The usual answers boil down to two: “Learn to use AI,” or find a job that AI cannot replace and that will not be automated, preferably hands-on work.
According to Business Insider, Ethan Mollick does not think this approach will work. Mollick is a management professor at the Wharton School and an influential researcher who focuses on explaining what AI can do today and what it means for ordinary people who need to work alongside it. His new book, Co-Existence, argues that large models are advancing far faster than most people realize, and that the only way to keep up with their rapid development is to learn to coexist with them.
That means acknowledging that many of the abilities you have now will continue to lose value in the future, and focusing on developing personal qualities that AI cannot replicate.
IT Home noted that Peter Kafka spoke with Mollick on his podcast Channels about AI's impact on work, career choices for the next generation, and what will happen after traditional systems for developing talent break down under the pressure of AI. What follows is an edited excerpt from the interview.
Peter Kafka: People worried about their own jobs or their children's prospects are often advised to become AI experts or find jobs resilient to AI, such as plumbers or nurses. Is that useful advice?
Ethan Mollick: I don't think it helps much. For one thing, even if you choose one of those careers, things change if a robot like PlumBot 6000 appears someday. I'm not a robotics expert, but the robotics researchers I know have changed their views. They used to think, “Large language models and robots are completely different things; developing robots is extremely difficult.” Now they realize there really may be a way to connect the two.
I don't know where things will go, and I believe no one can predict it accurately. Coming from an AI researcher, that may sound reassuring, or it may make people anxious.
Programming is the industry undergoing the most dramatic change. We can already see programming jobs changing, and we cannot even be sure whether the total number of programmer jobs will rise or fall in the future. What is certain is that programmers' work will keep evolving.
Writing code with zero errors is already less important than clearly thinking through what we want to build. Many programmers enjoy their work more than they used to, but some feel a sense of loss. These jobs have always been evolving.
In my book, I describe four qualities that can give people at least some advantage. The first is deep expertise: immerse yourself in a field and draw on your accumulated knowledge to outperform AI. The second is breadth of knowledge: explore many different fields.
I've observed a pattern at many companies: programmers are starting to do design, designers are writing code, and marketing staff are drafting business plans. The more you know across different fields, the better you can use AI.
The third is aesthetic judgment. AI produces enormous amounts of content. To avoid shoddy results, you need judgment and the ability to distinguish the good from the bad.
The last is initiative. I think simply saying “learn to use AI” no longer means much. We've tested various ways of teaching people to use AI, and the techniques that worked in the past are gradually becoming ineffective.
Prompt engineering is becoming more like project management. Instead of relying on some magic incantation to direct AI, you define your requirements clearly: I want to make this product; what are the acceptance criteria, how will we test it, and how will it be delivered in stages? Give me five versions of a plan to choose from.
For midcareer workers with years of experience, perspective, and judgment, that is good news. But if you are just starting out, how do you develop those abilities in an environment where AI can readily look things up and draft documents for you?
There are two layers to the problem: one concerns individuals, and the other concerns our existing education and training systems.
At the individual level, humans are naturally inclined to outsource mental work; it is ingrained in us. Technology is meant to free us from tedious labor. Once people have a dishwasher, no one wants to insist on washing dishes by hand; nor does anyone want to churn butter themselves.
My children no longer learn cursive handwriting, and I don't know how to use a slide rule either.
But it is hard to tell which abilities we can safely hand over to tools. Students using AI to cheat on assignments is, at its core, a manifestation of this problem.
Then there are the various institutions that train people. One career learning path runs through middle school, high school, and university education. That system is admittedly a mess right now, but we still have room to adjust it. AI itself is an excellent tutoring tool, and we also have ways to assess students without AI: closed-book written exams and face-to-face oral exams. In the future, AI could tutor students after class while classrooms use active learning, making education better overall.
What really worries me is the other path to professional growth: apprenticeships.
In the past, newcomers joining banks, consulting firms, auditing firms, and car dealerships followed a model of professional development that had been passed down for thousands of years.
They did tedious, menial entry-level work for meager pay, sometimes while enduring harsh treatment; in exchange, they learned skills, built connections, and learned how to do the job through practice. Middle managers also had someone to handle routine tasks. Even if mentors were not good teachers, newcomers still became familiar with business processes; those who delivered results could advance, continually producing qualified professionals. This model worked for thousands of years.
But now that system has stopped working: the incentives of everyone involved no longer align. For newcomers, the easiest choice is to hand everything to AI, which often produces better work than they can themselves. But newcomers can easily misuse AI: they simply give Claude the assignment and submit its output unchanged. They have not yet developed judgment and cannot tell whether the report is any good.
Managers may also think: If AI can solve problems faster and better, why should I hire an intern? Company leaders, meanwhile, may ask why they should slow things down and accept work of inconsistent quality from interns.
The once virtuous cycle of talent development has broken down completely.
There are only two ways forward: either expand formal vocational education, so that many entry-level office jobs follow a vocational-school route like welding and plumbing; or companies must take mentoring seriously as a real part of the job. Training newcomers takes time and sacrifices efficiency in the short term, but it creates a steady supply of future talent.
What worries me most is that companies will choose to “hire only experienced people.” That is a dangerous path: the pipeline of talent would be cut off, leaving no one to take their place.