Summary: Productivity gains from AI will obliterate many jobs, but people adapt and create new work. Humans always crave more, inventing new products, services, and experiences. Many jobs will disappear within 10 years, too quickly for workers to retire or retrain leisurely. However, AI accelerates retraining, and AI tools increase learning rates and enable personalized tutoring at scale. Start now on building AI-driven education so that it’s ready when needed.
300 million jobs will evaporate worldwide thanks to generative AI, according to Goldman Sachs. Given the average 66% productivity increase we’ve already measured in early studies of fairly primitive AI tools (ChatGPT 3.5 etc.), I would not be surprised if productivity gains by the end of the decade reach 100%, with half a billion workers rendered superfluous in their current jobs. This is especially likely to happen if the next generation of industry-specific AI tools is driven by UX research and design instead of the purely engineering-driven “design” that plagues current AI tools.
When people read these economic predictions of jobs being taken over by AI, a natural first reaction is to worry about unemployment. Followed a few minutes later by a sense of panic as readers realize that their jobs are next on the line. In contrast to earlier rounds of automation which only hit manual labor, the new generative AI tools target highly-educated workers. Early data suggests that the productivity gains may be larger the more cognitively complex the job. You, my dear reader, are next in line for the dole.
Oh, the angst. Oh, the gnashing of the Internet’s collective teeth.
Internet angst over AI-induced mass employment of the former elite. (“Angst” by Midjourney.)
Wrong! The prediction that vast masses of unemployed wretches will result from productivity gains is the fixed-work fallacy. As the name says, this is erroneous. This doesn’t happen in the real world.
Yes, it could hypothetically be true that only a fixed amount of work needed to be done in the world. Then (and only then) would it follow that if workers can now produce twice as much as before, half of the previous workforce would become surplus to needs and be fired.
But this is not true.
There is not a predetermined fixed quantity of work that needs to be done. Humans have infinite inventiveness to develop ideas for new activities and desires. And humans also have infinite capriciousness for desiring new things. Nobody is satisfied with what they currently have. (If this were the case, people would routinely turn down salary raises because there would be nothing to spend that extra money on.)
Look at my ancestors, living 50 miles north of Copenhagen in the year 1600:
“Poor Danish farming family” generated by Midjourney.
Even though the big city is only an hour’s drive from their farm in a modern automobile, they never visited. The furthest they traveled was to the neighboring hamlet, a rare treat. They were dirt poor, and yet they look happy enough in the picture. (Of course, they are not my literal forebears, but invented by an AI.) If we asked these people what more they desired compared to what they already had, they would likely have petitioned for an extra pig. They would not ask for a commuter train ticket to visit Copenhagen, let alone for a car, since neither trains nor cars had been invented yet. Certainly, they wouldn’t want tuition for the young girl to be educated as a UX designer.
A list of the things, services, and experiences you enjoy, which my peasant ancestors couldn’t even envisage, will run longer than any reasonable word count for this article. The list of things your descendants will enjoy that you don’t currently have will be equally long.
(And when I say “things,” I mean products, services, experiences, and any other nicety that people might like to spend their money on, and thus create a demand for other people’s labor, bringing said niceties into the world.)
This infinite demand for new things is why the fixed-work fallacy is false.
Let’s look at history more generally than just a few of my Danish ancestors:
Until 10,000 years ago, everyone worked as hunters and gathers.
Then, around 10,000 years ago, several Edison-level inventors living in the Fertile Crescent of the Middle East, the Yangtze River basin of China, and the New Guinea highlands devised the radical notion to raise animals instead of hunting them and to cultivate plants in a specific place (later named a “field”) instead of roaming wide and far to gather them. Wow, agriculture, what an innovation.
It’s estimated that each hunter-gatherer could produce 2,000 to 4,000 calories daily. In contrast, Neolithic farmers produced 5,000 to 10,000 calories per day. Agriculture increased each stone-age person’s productivity by 150% — even better than the 126% productivity increase for programmers using Github’s AI-driven Copilot.
With a 150% productivity increase, the fixed-work fallacy would predict that 60% of the stone-age population had become unemployed hunters. (As farmers, 40% of the workforce could produce all the food that necessitated the work of everybody before.) Of course, that’s not what happened. Rather, the population grew, and civilizations emerged.
I can write another set of bullet points about the change from agriculture to the manufacturing industry, but let’s cut straight to the present day, where 2% of the U.S. workforce is engaged in farming. (The farming share of the workforce was 41% in the United States as recently as 1900 and was very close to 100% in all countries in the year AD 1.) With 2% of workers employed in farming, does this mean that 98% of the workforce is now on the dole, showing up each day at the farmhands’ union hall to see if, by some miracle, there’s a job that day to drive a tractor on some farm where the regular tractor operator called in sick?
No, no industrialized country in the world has 98% unemployment.
Productivity improvements don’t cause unemployment because every time the demand dwindles for workers to produce old stuff, engineers and entrepreneurs invent ventures that soak up the supposedly-useless workers and employ them to make new stuff.
What are those new things going to be this time around? It won’t be growing turnips like when the hunters stopped being needed. It won’t be manufacturing cast iron like it was when most farmers stopped being needed. This time, it might be user-experience professionals because the world needs more UX work to fix all the terrible design. However, mainly the fired people will be employed doing new jobs that haven’t been invented yet. (Just like the very role of UX Specialist didn’t exist 50 years ago.)
Whenever somebody says, “But this time it’s different,” guard your wallet and flee. Invest your money with anybody else, preferably somebody saying, “This time, the time-proven rules still hold, but with a tweak.”
When something has been the same for 10,000 years, it is a safe bet to assume that it will remain the same for the next 100 years.
UX Unemployment?
You may be forgiven for caring less about the general workforce than about your own employment prospects. Since I target an audience of UX professionals, let’s consider the impact of AI on UX jobs.
As I mentioned in my review of the existing research on AI’s productivity impact, I expect the UX profession to gain fewer benefits from AI than other, similarly cognitively demanding jobs. (In general, it seems that AI helps more on complex jobs, probably because of its function as “forklifts for the mind” where the AI can take over much heavy data manipulation.)
UX work will likely be helped less because many of our activities are anchored in human beings. Most importantly, the users and their workplace. But also our colleagues who invariably need convincing to embrace the best design. Much automation can be achieved, such as summarizing field study notes. But it will always take an hour to sit with another person for an hour, so there are no productivity gains in the human-facing side of our work.
Still, UX professionals are knowledge workers and will see substantial productivity gains from AI. (That’s why I urge you to embrace AI now before others do so and outcompete you.)
If we believed in the fixed-work fallacy, we would need fewer UXers in the future. But it’s absolutely not the case that there’s a fixed need for UX work that is identical to the amount of UX work delivered today. The world is severely underserved when it comes to design quality.
UX costs too much per unit of design quality. (“Coins” by Leonardo.AI.)
There are three reasons there’s currently too little UX in the world:
UX is too expensive per unit of design quality delivered. Only big companies (and small companies that care uncommonly about quality) can afford to pay our current prices. If and when the cost per unit of UX work drops in half due to better AI support, demand for UX work will surge.
UX is too unestablished in many companies that still live at the bottom half of the UX Maturity scale. This is a historical artifact that follows from the relative youth of the discipline. More resources will be allocated to UX as time passes, and more senior executives have personal experience with the value of UX work from their salad days as hands-on managers. More companies will claw their way up the maturity scale.
There are too few UXers in the world. Again because it’s a new field, not many people were trained in UX in the past. To this day, universities deliver fewer UX graduates than the market needs (especially BS/BA/BFA degrees which are more valuable than MS or Ph.D. degrees for aspiring UX pros).
All three reasons are gradually being addressed, so we’ll have more UX in the future. In particular, I think that the AI-induced price drop will generate more than sufficient incremental demand to soak up any UX professionals that may be made redundant in those companies that choose to deliver the same design quality as always, even when their UX teams have become more efficient. (I expect those companies will become the dredge of history, as other companies exploit the new opportunities for higher quality and thus win customers. But that’s a story for another day.)
The Difference: Faster Pace
I mentioned before that “this time it’s different” are the most dangerous words in thinking about changes in the world. But, this time, things are indeed different in a critical way from all the 10,000-year-long precedents. Yes, there will be enough new jobs and professions to soak up all the otherwise unemployed. But the transition from old to new jobs will happen at an unprecedentedly accelerated tempo.
The AI Express waits for nobody. (“Speeding Train” by Midjourney.)
That’s the tweak: faster pace because the AI revolution waits for nobody. We face an express train hurtling down the tracks, not a cart hauled by a flea-bitten goat along a dusty trail.
Previously, 40 years could easily pass from when a role started to become irrelevant until all those jobs had been eliminated. During that time, most employees would have retired due to natural attrition — or at least they would have plenty of time to ponder alternate careers.
Now, accelerated technology diffusion makes it likely that most companies in rich countries (and many in mid-income countries) will have completed adopting AI tools in 10 years. In fact, 10 years seem preposterously lethargic considering how easily AI can be deployed to realize vast productivity gains. But the complete roll-out will take a decade because we need many domain-specific AI applications to be developed to make AI truly useful in many companies.
At present, a smattering of particular roles can use the existing rudimentary AI tools for a few tasks, but that will only double the productivity of some of the workforce. Employees may double their productivity while performing those tasks that are especially amenable to be performed with the help of low-usability AI tools, but this will account for a small proportion of each person’s workday in most cases.
Still, the complete diffusion of AI across the economy in 10 years represents 4 times as fast change as most previous workforce-shaking innovations. Even worse (from the perspective of unemployment), the AI-driven productivity improvements squeeze 57 normal years of economic growth into that single decade. Given this immense acceleration, most workers won’t have the luxury of early retirement or leisurely retraining for new careers. They’ll be rapidly booted from their obsolete old jobs and won’t have a comfortable readymade alternative waiting.
Leaving workers hanging in this brutal manner is unacceptable. Luckily, there’s a solution at our AI-created doorstep: hugely accelerated retraining to make redundant workers qualified for the many new jobs we know will spring forth.
AI excels at teaching. We know from the relatively limited empirical research into AI-driven productivity that people using an AI tool become proficient in a new profession much faster than people who do that same job without AI help. Well-designed AI-driven domain-specific applications act as forklifts for the mind and take over much of the heavy lifting in understanding complicated new concepts.
Thus, AI-supported jobs will be easier to learn as learners grasp concepts and skills at their own pace.
AI can’t do everything, so some learning will remain for most new jobs, meaning that workers will still need retraining courses. AI can also help with these courses because AI has immense potential for revolutionizing learning through individualized tutoring.
The pampered progeny of the wealthiest families has always benefited from personal tutors with a 1:1 ratio between learners and instructors. Only the proles suffer the indignity of being crammed into a classroom with 20+ other kids to receive undifferentiated instruction that doesn’t target their individual needs and learning level. (Sadly, the current education system treats 99% of the population to such substandard mass-produced schooling.) We can revert to the artisanal production of education with this luxurious 1:1 ratio between learners and teachers — as long as AI does the teaching. AI scales, humans don’t.
I expect there will still be some need for human teachers to supplement the AI, but likely primarily for motivation and less for teaching new concepts and skills, which will proceed at each student’s pace and cater to his or her personal interest and goals.
AI business applications are moving fast, so we need even more urgency and speed in perfecting AI for education: it’s a horserace, but the education horse isn’t out of the gate yet. (“Speed” image by Midjourney.)
Such AI-expedited instruction will be sorely needed in a few years when hundreds of millions of redundant workers will clamor for retraining. There’s no time to squander: we must start developing these AI-supported educational tools now. Pedagogy is a subtle art, and the initial attempts will surely fail. Let’s get past these failures before it’s too late so we have robust solutions ready when they become sorely needed in a few years.
Conclusion: The Future is Active, not Passive
Negative thinking about the future is driven by the sense that humans exhibit a lack of agency. We are passive weaklings to be crushed by the slightest change to our present cushy circumstances. People with this mindset envisage the future like the left image in the illustration below.
But it’s not true! Humans are active and adaptive. That’s why we could populate the planet from the hottest to the coldest climates thousands of years ago. And it’s even more true now when we don’t have to worry about getting enough food but can focus our mental energies on conceiving ways of improving everything. We think and tinker endlessly, as visualized in the right image below.
Passive vs. active futures (“hopeless unemployment line” and “inventors around a table” images generated by Midjourney).
Midjourney struggled to create evocative pictures of a hopeless unemployment line in the future; it probably did a little better with the group of inventors building the future. In any case, you get the point of passive versus active humans. We invent things continuously and adapt to thrive in changing environments, so the righthand picture is the more fitting metaphor for our increased productivity from deploying AI across the economy.
More info: Marc Andreessen wrote a lengthy article about why the AI doomsayers are wrong, countering other dubious complaints about AI. Marc invented the first widely used graphical user interface to the web (Mosaic) in 1993 and has been a technology leader ever since, so he’s worth listening to, even though I think his headline is overstating matters (“Why AI Will Save the World”).
Quiz: Check Your Understanding of This Article
Check your comprehension. Here are 7 questions about ideas and details in this article. The correct answers are given after the illustration below.
Question 1: What is the "fixed-work fallacy" as described in the article?
A. The belief that AI will create more jobs than it eliminates
B. The assumption that there is a predetermined, fixed amount of work that needs to be done in the world
C. The idea that AI will only affect manual labor jobs
D. The notion that AI will not significantly impact employment rates
Question 2: According to the article, how does AI impact the pace of job transition compared to previous workforce-shaking innovations?
A. AI slows down the pace of job transition
B. AI does not significantly impact the pace of job transition
C. AI accelerates the pace of job transition
D. The impact of AI on the pace of job transition is still uncertain
Question 3: How does the author suggest AI can help in the transition from old to new jobs?
A. By slowing down the pace of job elimination
B. By accelerating the retraining process through AI-driven education
C. By creating more jobs than it eliminates
D. By reducing the need for human labor in certain industries
Question 4: According to the article, how will AI impact the field of User Experience (UX)?
A. AI will make UX jobs obsolete
B. AI will not significantly impact UX jobs
C. AI will help increase productivity in UX jobs, but will not eliminate the need for UX professionals
D. AI will completely automate UX jobs
Question 5: What is the author's perspective on the future of work in the face of AI advancements?
A. The author believes that AI will lead to mass unemployment
B. The author believes that AI will create more jobs than it eliminates
C. The author believes that while AI will eliminate many jobs, humans will adapt and create new work
D. The author believes that AI will not significantly impact employment rates
Question 6: According to the article, what is one of the reasons there is currently too little UX in the world?
A. UX is too expensive per unit of design quality delivered
B. There are too many UX professionals in the world
C. UX is not valuable to companies
D. UX is too established in many companies
Question 7: What is the author's perspective on the role of AI in education?
A. AI will replace human teachers entirely
B. AI will have no significant impact on education
C. AI will revolutionize learning through individualized tutoring, but human teachers will still be needed for motivation
D. AI will make education less accessible to many people
“Take the Test” image by Wepik.
Quiz Answers
Question 1: What is the "fixed-work fallacy" as described in the article?
Correct answer: B. The assumption that there is a predetermined, fixed amount of work that needs to be done in the world
Question 2: According to the article, how does AI impact the pace of job transition compared to previous workforce-shaking innovations?
Correct answer: C. AI accelerates the pace of job transition
Question 3: How does the author suggest AI can help in the transition from old to new jobs?
Correct answer: B. By accelerating the retraining process through AI-driven education
Question 4: According to the article, how will AI impact the field of User Experience (UX)?
Correct answer: C. AI will help increase productivity in UX jobs, but will not eliminate the need for UX professionals
Question 5: What is the author's perspective on the future of work in the face of AI advancements?
Correct answer: C. The author believes that while AI will eliminate many jobs, humans will adapt and create new work
Question 6: According to the article, what is one of the reasons there is currently too little UX in the world?
Correct answer: A. UX is too expensive per unit of design quality delivered
Question 7: What is the author's perspective on the role of AI in education?
Correct answer: C. AI will revolutionize learning through individualized tutoring, but human teachers will still be needed for motivation