UPI Study › Courses › Introduction to Artificial Intelligence › Free lessons › Lesson 9.6 Free with a free account
How AI Reshapes Economies and Markets
Introduction to Artificial Intelligence · Chapter 9: AI Applications, Ethics, and the FutureSection 9.6 · lesson 67 of 6913 min · free with a free account
Free preview
Create a free account to keep watching
It takes a minute and costs nothing. Your free account unlocks all of Chapter 1 and 3 more lessons of Introduction to Artificial Intelligence.
Lesson complete
Earn the 3 college credits for Introduction to Artificial Intelligence
Enrolling adds a quiz after each lesson, the graded work and the official transcript. Recommended for credit by NCCRS.
Next lesson: The Evolving Workplace: AI's Influence on Jobs and Skills →← Emerging Directions in AI and RoboticsNext lesson: The Evolving Workplace: AI's Influence on Jobs and Skills →
In this lesson
Simple models of AI's effects on productivity, competition and investment.
- Explain why AI productivity gains may arrive late and unevenly, using the J-curve, compounding and bottleneck models
- Calculate the Herfindahl-Hirschman Index and describe forces that push AI markets toward concentration
- Analyze how market structure, adoption speed and discounting affect who gains from AI investment
- Identify warning signs of valuation bubbles and capacity overbuilding
Free study notes
Get this lesson's notes by email
The key points and 4 practice questions with answers, in your inbox. One email, free.
One email. Unsubscribe in one click.
Transcript
Everything said in the video, section by section (1,702 words). Read along, or search it.
Opening Question
Solow's Productivity Puzzle Returns
In 1987, the economist Robert Solow wrote that you could see the computer age everywhere, except in the productivity statistics. AI raises the same puzzle. Will it show up in economic output? Who will capture the gains? And could the boom around it be a bubble? This lesson builds small models of these forces. They aren't forecasts. They show why certain outcomes can happen, so you can judge the claims.
Productivity
Why Productivity Beats Extra Capital
Productivity is the amount produced per unit of input, like per hour worked. Over long periods, it's what raises living standards. A standard simple model writes output as A, productivity, times capital to the power 0.3, times labor to the power 0.7. AI could raise any term. But raise each by 10 percent, and productivity lifts output 10 percent, capital only 2.9, and effective labor 6.9. Extra capital has diminishing returns. That's why economists care most whether AI raises productivity itself, not just the amount of equipment.
The J-Curve
Gains Trail Adoption by Years
AI is a candidate general-purpose technology: it spreads across the economy, improves over time, and invites new inventions on top of it. Such technologies show up in output slowly. Factories took decades to reorganize around electric motors. The benefit needs complementary investments: new processes, retraining, better data. Those cost money before they pay off, so a firm's productivity may first dip, then climb. That shape is a J-curve. In a toy model, each adopting firm loses 5 percent for two years, then gains toward 25 percent. Half the firms adopt by year 10, but half the gain arrives in year 15. Nine in ten firms adopt by year 16, and nine-tenths of the gain comes by year 21. The gain trails adoption by about five years. Judge AI too early, and you'll misjudge it.
Compounding Growth
Small Growth Gaps Become Large
And small growth differences compound. Output per worker growing 1 percent a year is 35 percent higher after 30 years, and takes 70 years to double. At 2 percent, it's 81 percent higher, doubling in 35 years. At 3 percent, it's 2.43 times higher, doubling in about 23. Raise growth from 1.5 to 2 percent, just half a point, and output per worker ends 16 percent higher after 30 years. So forecasts a point apart aren't trivially different. There's also a measurement problem. Statistics count what's sold at market prices, so free digital services and better quality can make people better off without showing up much.
Bottlenecks
Slow Tasks Cap the Speedup
Gains are also capped by what AI can't speed up. If AI makes a fraction f of the work s times faster, the whole process speeds up by 1 over 1 minus f, plus f over s. Speed up 30 percent of the work tenfold, and the whole only gets 1.37 times faster. Even doing that 30 percent instantly caps it at 1.43. Speed up 90 percent tenfold, and the whole gets 5.26 times faster. The slow parts, physical tasks, approvals and human judgment, set the pace. So big gains need the whole process redesigned, not one step sped up.
Quick Check
Speeding Up 80 Percent of the Work
Quick check. AI makes 80 percent of a team's work four times faster, and leaves the rest unchanged. How much faster is the whole process? Pause, and work it out. Two and a half times. The untouched 20 percent still takes 0.2 of the old time, and the sped-up part takes 0.8 over 4, also 0.2. That's 0.4 of the old time in all, and 1 over 0.4 is 2.5. Four times is the trap: the slow fifth holds it back.
Market Concentration
Fixed Costs Favor the Giant
Now markets. Several features of AI push toward concentration, a few firms holding most of a market. Large fixed costs with tiny marginal costs. Data feedback loops: more users, more data, a better service, more users. Scarce computing power. And talent and customers clustered in a few places. The first is simple arithmetic. Say a service costs $100 million to build, and 50 cents a user a year to run. With a million users, each costs $100.50. With ten million, $10.50. With a hundred million, $1.50. The giant can charge a fraction of a small rival's price.
Measuring Concentration
The Herfindahl-Hirschman Index
A common measure is the Herfindahl-Hirschman Index. Square each firm's market share, in percent, and add them up. It runs from near zero to ten thousand, a monopoly. Four equal firms at 25 percent score 2,500. Ten equal firms score 1,000. One firm at 60 percent and four at 10 score 4,000. Shares of 50, 30, 10 and 10 score 3,600. Competition authorities watch this number, and how it changes.
Data Feedback Loops
Early Luck Picks the Winner
Now model a data feedback loop. Two equal firms. Each period, 100 new users choose, and each leans toward the firm with more data. One dial sets how strong that pull is. In 300 simulated markets, with no feedback the leader ends near 50 percent. At strength 1, about 55. At strength 2, the leader takes 94 percent, and at 3, about 99. The HHI heads toward ten thousand. And with strong feedback, which firm wins depends on early luck. The winner needn't be the better firm.
Quick Check
HHI for an AI Chip Market
Quick check. A market for AI chips has four firms, with shares of 40, 30, 20 and 10 percent. What's its HHI? Pause, and work it out. Three thousand. Forty squared is 1,600, thirty squared is 900, twenty squared is 400, and ten squared is 100. Twenty-five hundred is the trap. That's four equal firms. Among four firms, unequal shares always score higher than equal ones.
Both Sides
Benefits and Harms of Scale
Concentration cuts both ways. Scale can lower costs and prices, fund huge investments, and create standard platforms others build on. But once rivals are gone, prices can rise or quality fall, challengers can't enter with new ideas, and many firms depend on a few suppliers. Some forces push back: rented computing, openly shared models and tools, specialized niche models, and customers who use several providers.
Value Chain
Layers From Chips to Applications
AI products are built in layers. Chips and hardware need enormous design and factory costs. Cloud data centers are heavy in capital and energy. Models are costly to train, but cheap to run once built. Tools and platforms are mostly software. And applications win by knowing the customer, trust and distribution. Firms earn money in several ways: subscriptions, pay per use, advertising, AI embedded in existing products, outcome-based fees, and open models with paid services.
Falling Costs
Who Keeps the Savings
Who keeps the savings when computing gets cheaper? A pay-per-use service charges $2 per thousand requests, and spends $1.20 serving them, a 40 percent margin. Computing costs fall 30 percent a year. If the firm holds its price, its margin reaches 79 percent by year 3. If competition cuts prices 25 percent a year, the margin is about 51 percent, and customers pay 58 percent less. Market structure decides who gets the gain.
Investment Value
Net Present Value Depends on Adoption
Building AI capacity needs money now, with revenue later. Take a project costing 100 units, with returns over ten years, discounted at 10 percent a year. With fast adoption, it returns 510 and pays back in year 4. Its net present value is plus 191. Medium adoption returns 350, with a net present value of 88. Slow adoption returns 168, more than it cost, yet its net present value is negative: minus 9. The returns come too late to pay for the wait. The value of a huge investment hangs on how fast customers adopt, which nobody knows in advance.
Widening Gaps
Frontier Firms, Laggards and Diffusion
Gaps can widen between firms. Those with more data, skills, capital and well-organized processes adopt sooner and gain more. A frontier firm growing 3 percent a year, against a laggard at 1, starts equal. After 10 years it's 1.22 times as productive. After 30, 1.8. Between countries, picture a leader with four times a follower's income. Over 30 years, if the leader grows a point faster, the ratio rises to 5.37. If the follower grows a point faster, it falls to 2.99. These are scenarios, not forecasts. What matters is diffusion: whether the ability to use AI spreads. Countries need computing and energy, skilled people, data in their own languages, capital and trust in institutions, and reliable connectivity.
Bubbles
Valuations Built on Growth Expectations
Finally, bubbles: prices that detach from fundamentals as expectations run ahead of reality. A cash flow growing forever at g, discounted at r, is worth next year's cash over r minus g. With r at 10 percent, expected growth of 4 percent gives a value of 16.7, and 6 percent gives 25. At 8 percent, it's 50. So if expectations slip from 8 to 6 percent, the value halves, though nothing about today's cash flow changed. Only the story about the future did.
Overbuilding
Extrapolated Demand and Idle Capacity
Capacity can overshoot too. In a toy model, demand follows an S-curve to 100, and firms order data centers by extrapolating current growth. They arrive two years later. Extrapolate 3 years ahead, and they build 101: 99 percent used. Extrapolate 5, and they build 114: 87 percent used. Extrapolate 7, and they build 132, with 76 percent used. Railways and fiber networks were overbuilt, and proved valuable later, but not to the investors who paid for them. Warning signs: valuations that need years of very high growth, borrowing or circular deals, dependence on a few firms, and stories that every business will soon be transformed.
Back to Solow
What the Models Can and Cannot Show
Back to Solow's puzzle. The models show why AI may look invisible in the statistics at first: J-curves, slow adoption, bottlenecks and unmeasured benefits. They show why gains may concentrate, why progress can feed profits or lower prices, and why a real technology can still bring a boom and bust. What they can't do is give dates. They show which assumptions a claim depends on.
Key Takeaways
AI Economics in Summary
To sum up. Productivity gains from AI likely arrive late, because of the J-curve, and unevenly, because the slowest tasks limit the whole. Small differences in growth compound into large ones. Large fixed costs and data feedback loops push markets toward concentration, measured by the HHI. Whether AI widens or narrows gaps between firms and countries depends on diffusion. Valuations and capacity built on extrapolated growth are fragile, so a good technology can still produce a boom and bust.
Think About It
Doubling One Department's Productivity
Here's something to think about. A firm announces that AI will double productivity in one department. Using the bottleneck model, what would you ask about the work AI doesn't speed up? What share would AI have to handle, and how fast?
About Introduction to Artificial Intelligence at UPI Study
Introduction to Artificial Intelligence (Computer Science 373) is a self-paced online course of 69 video lessons in 9 chapters, from intelligent agents and search to game strategy, neural networks, logic and reasoning, machine learning, language, robotics and AI ethics. It carries a NCCRS credit recommendation for 3 semester credits.
Every lesson transcript is free on pages like this one, and the videos are free with a free UPI Study account. Enrolling adds a practice quiz after each lesson, the graded work and final exam that earn the credit, lifetime access, and an official transcript on completion. UPI Study has helped over 130,000 students with college credits. ACE and NCCRS have over 2,100 cooperating colleges that will consider accepting credits; the final decision is with the universities.
Enroll now → Course details Download the syllabus All Computer Science courses Every course with PRO →
Compare: AI vs Current Trends in CS and IT
Why take it here
Why students take Introduction to Artificial Intelligence with UPI Study
Where the credit goes
Universities that work with NCCRS
Introduction to Artificial Intelligence carries a NCCRS credit recommendation. These universities work with NCCRS; some are UPI Study's signed partners. ACE and NCCRS have over 2,100 cooperating colleges that will consider accepting credits; the final decision is with the universities.
Thomas Edison State UniversitySigned partner
University of the PeopleSigned partner
Charter Oak State CollegeSigned partner
Strayer UniversitySigned partner
Columbia Southern UniversitySigned partner
Indiana TechSigned partner
Penn StateWorks with NCCRS
Southern New Hampshire UniversityWorks with NCCRS
SUNYWorks with NCCRS
+many more cooperating universitiesFind my college →Find my college → Not listed? Ask Athena in the corner of this page, or book a 15-minute call.
FAQ
Questions students ask us
About How AI Reshapes Economies and Markets
What is the productivity J-curve for AI?
How do you calculate the overall speedup when AI speeds up only part of the work?
How is the Herfindahl-Hirschman Index calculated?
Why do AI markets tend toward concentration?
About the Introduction to Artificial Intelligence course
How many college credits is Introduction to Artificial Intelligence?
How much does Introduction to Artificial Intelligence cost and what is included?
Which universities accept UPI Study credits?
How long does Introduction to Artificial Intelligence take?
How are the exams done?
How do my credits reach my college?
Ready for the full course? Earn 3 credits in Introduction to Artificial Intelligence.
69 video lessons, a quiz after each one, and the graded work that earns the credit. Recommended for credit by NCCRS.
3 credits · NCCRSEnroll →