Should I Ditch the Dictionary and Drive a Dredge?
About a month ago, I wrote a piece called “Seven ‘T’ Jobs for the Future” in which I tried to predict which jobs would be most likely to endure as AI develops. That analysis was based on how I thought AI would be used in the future. I stand by that (for now).
But a juicy new study released this week puts a different lens on that question and turns up some slightly different findings. Unlike my piece, which attempted to project how AI could be used, this study looked at how AI is actually being used by the 40% of us that are now using it in one way or another. A group of researchers at Microsoft, led by Kiran Tomlinson, took a close and enlightening look at more than 200,000 interactions individuals had with Microsoft’s Bing Copilot, one of the many large language models seeking to do “generative AI.”
The researchers then looked at which of 342 different work activities people were asking Copilot to help them with and which ones they were asking Copilot to do for them. Then they analyzed whether people rated the interaction with AI as successful or not.
And that…. resulted in a list, a ranking of which occupations were most and least likely to be impacted by AI. I love me a list.
The formula the researchers used was sort of complicated: it looked at the tasks required to do each kind of job, then analyzed which of those tasks people were using AI for and which ones they thought AI helped with.
There’s been a huge debate over what the job implications of AI will be and how we should respond to better prepare for change. A list like this is red meat for people who are trying to come up with answers. I suspect you will enjoy it as much as I did.
Let’s start with which occupations the researchers say are most at-risk from AI. For each of the jobs on the list, I include a number for the percentage of job tasks people are already using AI for and the number of people working in those jobs in the US. (If the job you are wondering about is not on this list, there’s a link to the full study in my notes at the end):
MOST AT-RISK:
1. Interpreters and translators (Copilot users are already employing AI for help on 9image 8% of interpreting and translation tasks; there are a total of 51,560 people in the US in those jobs)
If AI can do 98% of the tasks in your interpreting job, this might be a good time to consider developing new skills (image from First Legal).
2. Historians (91%; 3,040)
3. Passenger attendants (80%; 20,190)
4. Sales representatives of services (84%; 1,142,020)
5. Writers and authors (85%; 49,450)
6. Customer service representatives (72%; 2,858,710)
7. CNC (computer numerical control) tool programmers: (90%; 28,030)
8. Telephone operators (80%; 4,600)
9. Ticket agents and travel clerks (71%; 119,270)
10. Broadcast announcers and radio DJs (74%; 25,070)
I did a deeper dive into the top 40 most vulnerable professions and it looked similar to the top 10. Some 20 of the 40 jobs researchers rated as most vulnerable had something to do with translation or analysis – the kind of work that interpreters and historians do, but also a lot of what management consultants and teachers do. Another 9 of the top 40 jobs fell roughly into the “sales” category – sales reps, but also travel agents, ad sales, financial advisors. Then there was a group of jobs in the customer service category – 5 of the 40 most vulnerable jobs fell into that category. Another three jobs were programming or data science-related.
LEAST AT-RISK:
Dredge operators (people are currently using AI for 0% of dredging-related tasks; there are 940 people currently working as dredgers in the US).
Can’t touch this! AI can’t replace any of the key tasks of dredge operators — yet (Image from Owlguru)
2. Bridge and lock tenders (0%; 3,460)
3. Water treatment plant and system operators (0%; 120,710)
4. Foundry mold and coremakers (0% 11,780)
5. Rail-track laying and maintenance equipment operators (0%; 18,770)
6. Pile driver operators (0% 3,010)
7. Floor sanders and finishers (0%; 5,070)
8. Orderlies (0%; 48,710)
9. Motorboat operators (1% 2,710)
10. Logging equipment operators (1%; 23,720)
When I looked deeper into the top 40 least at-risk jobs, they were overwhelmingly made up of jobs in the skilled and unskilled trades: in addition to the occupations listed above, you’ll find truck drivers, maids, roofers, dishwashers, and glass installers (these are the kinds of jobs I lumped under the “Tool and Die” category in my piece last month). There are also a number of health care occupations the researchers consider to be insulated from the challenges of AI: phlebotomists, nursing assistants, massage therapists, orderlies and others (these are the jobs I put in the “Touch” category in the earlier piece).
So how do we think about a study like this one from Microsoft?
I think it will perform an important role in getting more people to talk about and plan for a future where AI will become increasingly important to us. We need more people in more places seriously discussing what AI is doing and will do to the future of work.
I just have three cautions:
Be skeptical: Don’t take this to the bank. This is a study focusing exclusively on US users of a minor player in the generative AI market, Bing Copilot. In spite of all the interesting stuff it tells us, it is not able to distinguish whether people are using the AI for help with work projects or for personal use, or whether Copilot users are different from users of other larger AI tools.
Be realistic: You could read this study and conclude that you should ditch your customer service job to become a dredge operator. But while a lot of the 2.8 million customer service jobs will almost certainly disappear in the future, there will still be a lot more of those jobs than dreading jobs. My prediction: ten years from now we’ll still have about 940 dredge operators in the US.
Be flexible: Just because AI can perform many of the tasks in a given occupation doesn’t mean that the jobs will go away. It does mean those jobs will change. Maybe dredge operators aren’t using AI to determine how long it will take to clear the silt from a channel; I bet they will be in the future. Historians who ignore AI’s capacity to locate and review historical documents will put themselves at a disadvantage. It’s the same for salespeople — those who don’t use AI’s increasing ability to research and analyze sales leads will struggle. Attorneys and programmers who don’t use AI to help with parts of their work will struggle, slowly falling behind those that are adapting.
So take the study with a grain of salt, but enjoy the conversation it’ll spark. Quiz your friends to see how many of the top 10 most vulnerable or least vulnerable occupations they get right.
And use that as an excuse to start thinking about whether and when and how you need to change the way you work.
-Leslie
Next time: I take my own advice, testing AI to see if it can replace me in writing this blog!
Notes:
Microsoft study of AI job vulnerability: https://arxiv.org/pdf/2507.07935