Artificial Intelligence, Tasks, and the Future of Work
Jobs are bundles of tasks. AI substitutes for some and complements others, so analyze work at the task level rather than predicting which jobs disappear.
Reading
0%
Time left
~18 min
Quiz score
0/4
What this means
Most public argument about artificial intelligence asks the wrong question. "Will AI take my job?" is close to unanswerable, because a job is not one thing. A job is a bundle of tasks, and a technology almost never affects every task in that bundle the same way.
Economics already has the right vocabulary for this, and it comes from ordinary demand analysis. Two things are substitutes when one can be used instead of the other; they are complements when using one makes the other more valuable. Coffee and tea are substitutes. Coffee and a mug are complements. The same distinction applies to a technology and a worker, but it applies at the level of individual tasks, not whole occupations.
Consider a paralegal. Reading through thousands of documents to find the ones relevant to a case is a task where AI is a strong substitute; software does it faster and does not get tired. Interviewing a nervous witness, deciding which of the relevant documents actually matters to the argument, and signing off on work that a court will rely on are tasks where AI is at best a complement. If document review shrinks from most of the job to a small part of it, the paralegal's remaining hours shift toward the tasks where human judgment is the scarce input, and the value of that judgment can rise.
Now hold the two possibilities side by side, because the direction is genuinely uncertain and depends on specifics. If AI substitutes for most of the tasks in a job and the remaining tasks do not expand, demand for that kind of labor falls and wages or employment come under pressure. That is real, and pretending otherwise is not kindness. If AI substitutes for the routine portion while the remaining human tasks become more valuable, the same worker becomes more productive, can handle more cases or clients, and may earn more. Both patterns have occurred with past technologies, sometimes inside the same industry.
There is a third effect that is easy to miss because it happens somewhere other than where the disruption is visible. Technologies create demand for work that did not previously exist. Somebody has to specify what a system should do, check whether its output is correct, decide who is accountable when it is wrong, integrate it with the rest of a business, and handle the customers who need a person. These are not consolation-prize jobs; several of them require exactly the judgment the technology lacks. But they are frequently in different firms, different places, and different skill sets from the tasks that were displaced, which is why aggregate job creation offers little comfort to a specific displaced worker. Both facts are true, and the tension between them is the actual policy problem.
Why it matters
You are choosing courses, majors, and first jobs during a period when the task composition of many occupations is being rearranged. The useful response is not to guess which occupation will survive, since that guess has a poor historical track record. The useful response is to learn to read any job as a task list and ask, task by task, which ones a machine performs well and which ones it does not.
That analysis also points toward what to invest in. Human capital that pairs well with a powerful tool is worth more than human capital that competes with it directly. A worker who can direct, evaluate, and take responsibility for a system's output occupies a stronger position than one whose value came from performing the step the system now performs.
Real-world example
Look at what happened inside bank branches after automated teller machines spread. The core task of dispensing cash was substituted directly, and tellers per branch fell. But cheaper branches meant banks opened more of them, and the remaining teller role shifted toward selling accounts, resolving problems, and relationship work that the machine could not do. Total teller employment did not collapse in the way the initial substitution suggested it would. The lesson is not that technology never costs jobs, because it plainly cost the specific task. The lesson is that you cannot predict the employment outcome from the substituted task alone; you have to look at what happens to the rest of the bundle and to the cost of the service overall.
Try it
- Choose a specific occupation you might realistically enter. Be precise: "registered nurse in a hospital emergency department," not "healthcare." Precision matters because task bundles differ sharply within an industry.
- Build a task inventory of eight to twelve tasks that make up that job. Use the United States Department of Labor's O*NET database, which publishes detailed task lists by occupation, rather than guessing from imagination. Cite the occupation code you used.
- Sort every task into one of three columns: likely substituted by current AI, likely complemented by current AI, or largely unaffected. For each task, write one sentence of justification naming what the technology would actually have to do.
- Look up real employment data for your occupation using the Bureau of Labor Statistics Occupational Outlook Handbook. Record current employment, the projected change over the coming decade, and the reason BLS gives for the projection. Note whether that projection agrees with your own task-level analysis, and where it does not, say which you find more convincing and why.
- Estimate the effect on the worker. Given your three columns, does this job most likely shrink, grow, or transform in content? Write a short argument, and explicitly name what would have to be true for your prediction to be wrong.
- Design an adaptation plan for someone already working in that occupation. Identify two specific skills they should build, one specific task they should stop treating as their main value, and one concrete way they could gain experience directing or supervising AI output in their field. Avoid generic advice such as "learn technology."
- Identify durable human skills. List three capabilities that remain valuable in your occupation regardless of how the tools improve, and for each, explain the reason it resists substitution: is it physical dexterity in unstructured environments, accountability that a person must legally carry, negotiation and trust, ethical judgment under ambiguity, or something else? A skill is only on your list if you can name the reason.
- Present a five-minute case to the class. Compare your findings with a classmate who analyzed a different occupation and identify one pattern that holds across both jobs and one that does not.
Teacher note
The single most important move in this lesson is forcing analysis down to the task level, because students default to whole-job predictions and whole-job predictions are mostly noise. Step 2 should not be skipped or done from memory; O*NET task lists routinely surprise students by revealing that jobs they think of as technical are heavily interpersonal, or the reverse. Expect two opposite failure modes in step 3. Some students mark nearly everything as substituted, which usually means they are reasoning from headlines rather than from what the task physically requires. Others mark nearly nothing, which usually means they have not tried current tools. Pushing them to name the specific thing the technology would have to do in each case corrects both. Step 7 is where the strongest thinking appears; require the reason, not just the skill, because a list that says "creativity, communication, critical thinking" without justification is a slogan rather than an analysis. Keep the tone calibrated in discussion. This topic invites both dismissiveness and alarm, and neither is supported by the evidence, which shows genuine displacement of specific tasks alongside genuine growth in complementary work. A student has it when they can take an unfamiliar job, break it into tasks, and defend a substitute-or-complement call on each one.
Check yourself
A radiologist uses AI software that flags suspicious regions on scans, after which the radiologist reviews the flags, integrates patient history, and makes the diagnosis. What is the relationship between the technology and the radiologist's remaining work?
Why do economists argue that asking 'which jobs will AI eliminate' is less useful than asking about tasks?
Which skill is most likely to resist substitution, and for the right reason?
AI adoption in an industry displaces a set of routine tasks while total employment in the economy continues to grow. What does this tell us about the displaced workers?
Analyze work at the level of tasks, not job titles: technology substitutes for some tasks and complements others, and your position depends on whether the tasks you are best at pair with the tool or compete with it.