You have time to learn a new tool, but you’re not sure it’s worth it. You read that the technology is already changing work. Then you hear that its real benefits could take decades.
Neither claim answers the question you have today: would learning it help you do something you care about? Proof that a tool helps with one task is not the same as a forecast about the whole economy.
A long change can still bring early benefits. But having a technology doesn’t mean the improvements you want will follow. Know what a claim really measures. Then it’s easier to decide what to look into next.
This module is for working adults deciding how much spare learning time to give a technology. It covers commitments you can change or stop. It doesn’t cover technology your employer requires, uses where safety is at stake, or big decisions you can’t undo.
What you’ll get from this module
In this module
By the end of this module, you’ll have a clearer way to judge your next learning commitment. We’ll separate three things: when a technology arrives, how it helps with a task and what changes across the economy. You’ll apply those ideas to a situation from your own life. Then you’ll plan an amount of learning you can afford and decide when to review it. Factory history, research across many countries, an economic model, a workplace AI study and a rural electricity experiment show why these differences matter.
Ask Jing
Become a Jingius.
Different benefits run on different clocks
Economic history: a slow change, with useful early uses.
In a 1990 account, Paul A. David dates the broad gains from electric power in U.S. factories to the early 1920s. But switching to electricity meant more than swapping one power source for another.
Some factories used an electric motor to turn the existing shafts that drove whole groups of machines. Giving each machine its own motor offered a different way to lay out the factory. Making that work widely took machines that fitted, new factory designs and know-how.
David describes engineers slowly building up experience with the new approach. Replacing factories that still worked could also cost too much. Seeing what electricity might make possible was not the same as being able to use it widely.
Earlier benefits still mattered. Warren D. Devine, Jr.’s 1983 history reports an 1891 comparison: electricity was usually cheaper when only small amounts of power were needed. Large steam setups could still make economic sense.
Why it matters: These accounts describe different uses within one long change. Electricity could pay off for one factory while manufacturing as a whole changed much more slowly. The useful question is what made a particular use worthwhile, not how long everyone had to wait.
Read the historical account of factory electrification. Read the history of changing factory power systems.
Research across countries: arriving and being used are different things.
Diego Comin and Bart Hobijn’s 2010 study looks at how 15 technologies spread across 166 countries. It finds that newer technologies were taken up faster than older ones. It also finds big gaps between countries and between technologies.
That describes how quickly technologies spread. Diego Comin and Martí Mestieri’s 2018 study adds another point. It looks at many countries over two hundred years. Poorer countries caught up with richer ones in when technologies arrived. But the gap in how much they used those technologies grew wider.
Why it matters: So a technology could arrive sooner without being used as much. And knowing when people get it, or how much they use it, still leaves one more question: what useful result does it produce? Each of these measures a different part of the change.
Read the study of technology adoption. Read the study distinguishing arrival from intensity of use.
Economic model: early preparation can hide in the numbers.
Erik Brynjolfsson, Daniel Rock, and Chad Syverson’s January 2020 working paper explains why early effort may not show up fully in productivity figures. Their model, the productivity J-curve, looks at effort that is hard to count. That includes new ways of working and new skills built around a technology.
People spend time and money building those skills and ways of working. That effort can create something useful for later work, even when the figures don’t fully count it. In the model, early productivity growth then looks lower than it really is.
Later, those investments help produce more. The effort behind that output wasn’t fully counted. So growth can now look higher than it really is.
Why it matters: This is one possible explanation for weak early figures. It doesn’t mean every disappointing start will lead to a rebound. To understand a delay, we need to know what is being built and how it could help. Simply saying that more time is needed doesn’t explain anything.
Read the working paper on the productivity J-curve.
Workplace study: useful gains on a task came soon after the tool arrived.
Erik Brynjolfsson, Danielle Li, and Lindsey Raymond studied AI help for customer-support agents at one business-software company. The November 2024 version looks at data from 5,172 agents, who got access at different times.
The main rollout wasn’t random: agents weren’t given access by chance. The researchers compared what changed around the time each agent got access. They allowed for differences between agents and over time. They estimated that agents resolved 15% more issues per hour on average, with gains showing in the first month.
This was an early improvement in one workplace task. The clock started when agents got access, after the tool had been built and set up. It doesn’t tell us when a whole economy will benefit.
Why it matters: That is the useful contrast with factory history. We can check whether a task is improving now, while the technology’s wider effects are still playing out.
Read the author preprint on AI in customer support.
Field experiment: being connected didn’t bring the hoped-for gains.
Kenneth Lee, Edward Miguel, and Catherine Wolfram studied homes being connected to electricity in rural Kenya. Their April 2019 manuscript describes discounts on the cost of connecting, offered at random across 150 communities, each served by a transformer. A comparison group was offered no discount.
The researchers checked back about 16 and 32 months after homes were connected. They found little evidence of broad gains, economic or otherwise. One exception was that people reported being more satisfied with life. Newly connected homes used very little electricity.
These results are about those homes and that period. They don’t settle what happens over a lifetime or in other places. They show why getting access and getting a particular benefit are separate questions.
Why it matters: Set beside the AI study, the lesson isn’t that one technology is fast and another is slow. The people, the conditions and the results differ. A useful claim about timing needs to tell us what improved, for whom and from what starting point.
Read the author manuscript on rural electrification.
What this means for you
Focus on the next learning commitment you can choose. You can stay unsure about a technology’s future and still decide whether learning something about it is useful now.
Name the benefit before choosing the clock
Start with a task or skill you care about. Describe what better would look like, including what quality is good enough. “Learn the new tool” names an activity; “produce a usable summary” gives that activity a purpose.
Say you’re thinking about learning a spreadsheet feature to prepare a summary you make regularly. A usable result might have correct totals and a layout that’s easy to read. Choose the result that matters in your situation.
Now ask what the evidence measures. A demonstration that the feature exists tells you it’s available. Results from similar tasks tell you more about whether it could help with your summary.
Your goal doesn’t have to be getting more work done. Name it before you judge the result. Otherwise, a change of goal could make a failed attempt look like it achieved what you first wanted.
Look for evidence that fits your situation
Compare the people, the task, the result and how long it was watched with your own case. A result from a customer-support system that was already set up tells you about that system. It doesn’t predict that every new tool will help you within a month.
Tell a demonstration apart from real use. A demonstration can show something worth exploring. On its own, it doesn’t show what happens in the conditions you would face.
Write down the evidence that applies and the biggest way your case differs. Ask whether that difference changes what a trial could tell you. If a missing fact would change how much effort you put in, finding it can be your next useful step.
Identify what still needs to change
The factory history and the J-curve model show why preparation deserves attention. Name one thing that’s missing: access you can actually use, running costs you can afford, a tool that does the task reliably, or the right training.
Then describe what would show that thing is getting better. “The technology needs more time” gives you nothing to review. “I need to learn the formula this summary needs” is something you can look into.
In the spreadsheet example, not knowing the formula could justify learning it. But if the results are still wrong after the learning you planned, adding more time needs a reason. Is there a specific skill you still need, or is the result failing even after the preparation?
When a missing piece is sorted out, that can support another commitment. Repeated failures, or total costs that keep rising, can argue against putting in more effort. Naming what’s missing doesn’t guarantee it will improve.
Make the next commitment affordable and informative
Choose an amount of learning you can afford and can change or stop. Say what finishing it should show you. Make the trial big enough to test the task, without spending more effort than you’re willing to risk.
For the spreadsheet example, you might learn the operation you need and try it on a summary. Then check the totals and layout against the result you wanted. The point is to learn something that can change your decision.
If the preparation would take more effort than you can afford, pause or pick a smaller question. A trial too small to show the benefit you expect may leave the answer open. Write down that limit instead of treating it as proof of failure.
Decide what would justify continuing
Choose when you’ll review the decision before you start. It could be at the end of the learning you planned, or when a specific change in the evidence happens. Compare what happened with the result you wanted and the missing pieces you named.
Better results on the task can justify another small commitment. So can steady use in similar settings, or a missing piece being sorted out. Before you put in more effort, explain what the next commitment would achieve.
If the result you agreed on doesn’t appear, pause or change the commitment unless you have a specific new reason to carry on. Write down that reason. Don’t keep putting off the review without new evidence. That makes it hard to judge whether the first decision helped.
You can keep an eye on a technology while choosing not to spend more time on it now. Deciding your next commitment doesn’t require a final verdict on its future.
How about your daily life?
Where are you deciding whether to spend more time learning a technology? You might be trying a new way to prepare a report, organize personal information or build a creative skill.
Think of an optional commitment you can change or stop. What would you like to do better, and what uncertainty makes the choice hard?
Write down possible situations. One good idea is enough.
| A situation I could explore | What I would like to improve |
|---|---|
| ____________________ | ____________________ |
| ____________________ | ____________________ |
My chosen situation: ____________________
Choose the situation that would be most useful to understand now. You can change your choice. Take it into the exercise below; choosing it doesn’t yet mean more learning is worth it.
Plan your next learning commitment
Start with your chosen situation. If you came straight here, choose a technology you’re thinking of spending spare learning time on. Pick a commitment you can change or stop.
This exercise doesn’t cover technology your employer requires, uses where safety is at stake, or big commitments you can’t undo. Those decisions need limits and evidence beyond this exercise.
- Describe a useful result. Name the task or skill and the improvement you want. Include what quality is good enough, or another sign of benefit you could see.
My useful result and how I’ll recognize it: ____________________ - Identify relevant evidence and a missing condition. Compare the people, tasks, results and how long each was watched. Tell a demonstration apart from real use. Name what still needs to change, such as access, cost, reliability or training.
If you’re missing a fact that would change the decision, write down that gap and make finding it your next step.
Evidence, important differences and missing information: ____________________
Missing condition and what would show improvement: ____________________ - Set an affordable commitment. Choose the effort you can afford and what finishing it should show you. Match it to the task. If that’s more than you can afford, pause or choose a smaller question, and say plainly what stays unanswered.
What I’ll commit and what I expect to learn: ____________________ - Set the review point and possible decisions. Review at the end of the commitment, or when a specific change in the evidence happens. Say what would justify putting in more effort, keeping it the same, changing it, pausing or stopping.
Review point and evidence that would change my decision: ____________________ - Use the result to choose the next commitment. Compare what happened with the useful result and the missing condition. If the result you agreed on hasn’t appeared, pause or change the commitment unless a specific new reason supports carrying on.
Write down that reason before you carry on. If the trial couldn’t answer the question, leave the answer open. Don’t treat time passing on its own as a sign of progress.
Result, next decision and reason—or unresolved question: ____________________
Ask Jing
Become a Jingius.
Your next step
The plan is ready when you know what you’re trying to learn, how much effort you’ll put in and when you’ll review it. Afterwards, keep the result and the reason for your next decision with the plan.
A note on the evidence
These sources measure different kinds of change: productivity in factories, technology spreading across countries, how well workers do a task, and what happens in homes. We can’t combine them into a personal waiting time. The model explains one possible pattern in the figures; it doesn’t tell us which disappointing technologies will succeed in the end.
How we prepared this module
This module brings together history, research across countries, an economic model and two studies of technology in use. A made-up spreadsheet situation shows the decisions in action. The exercise helps you organize one learning commitment and review what it shows you.
Sources
- Paul A. David. (May 1990). The Dynamo and the Computer: An Historical Perspective on the Modern Productivity Paradox. American Economic Review, 80(2), 355–361. Published article; pp. 357–359, factory electrification and engineering learning.
- Warren D. Devine, Jr. (June 1983). From Shafts to Wires: Historical Perspective on Electrification. The Journal of Economic History, 43(2), 347–372. Published article; p. 355, electric line shaft drive and the 1891 comparison; pp. 366–367, motor compatibility.
- Diego Comin and Bart Hobijn. (December 2010). An Exploration of Technology Diffusion. American Economic Review, 100(5), 2031–2059. Published abstract, both paragraphs.
- Diego Comin and Martí Mestieri. (July 2018). If Technology Has Arrived Everywhere, Why Has Income Diverged?. American Economic Journal: Macroeconomics, 10(3), 137–178. Published abstract, opening paragraph.
- Erik Brynjolfsson, Daniel Rock, and Chad Syverson. (October 2018, revised January 2020). The Productivity J-Curve: How Intangibles Complement General Purpose Technologies. National Bureau of Economic Research (NBER), Working Paper 25148. January 2020 revision; abstract and introduction.
- Erik Brynjolfsson, Danielle Li, and Lindsey Raymond. (November 2024). Generative AI at Work. Author preprint, arXiv v2; abstract, sections 3.1–3.2 and 4.1, printed pp. 12–16. Published in The Quarterly Journal of Economics, 140(2), May 2025, 889–942.
- Kenneth Lee, Edward Miguel, and Catherine Wolfram. (April 2019). Experimental Evidence on the Economics of Rural Electrification. Author manuscript; abstract, introduction p. 4, sections IV.A–C, pp. 9–11; results pp. 18–19 and Table 3, panel D. Published in Journal of Political Economy, 128(4), April 2020, 1523–1565.
