AI now helps doctors read scans. But in 2025, hospitals offered a record number of radiology training spots, and pay went up.
TakeawayAI changed the job. It didn’t erase it.
Smart people make big predictions about AI and jobs. First, look back at one. Then see what researchers can measure right now. Then look ahead.
Remember Geoffrey Hinton from Trail 1? He’s one of the scientists who made modern AI possible. In 2016 he said:
“People should stop training radiologists now.”
(Radiologists are doctors who read X-rays and scans.) He thought AI would read scans better than doctors within five years.
So what happened? Read all four stories before you decide what they mean. Tap each card to see its takeaway.
AI now helps doctors read scans. But in 2025, hospitals offered a record number of radiology training spots, and pay went up.
TakeawayAI changed the job. It didn’t erase it.
In 2025, the head of the tech company Salesforce said AI helpers let it shrink its support team from about 9,000 people to about 5,000. Some workers moved to sales jobs.
TakeawaySome jobs really do shrink.
A 2026 study from the Federal Reserve, the government agency that tracks the U.S. economy, found it got harder for 18- to 24-year-olds to find jobs. The biggest reason was fewer job openings overall. AI explained about a third of it.
TakeawayAI is part of the story, not the whole story.
Americans applied to start a record 3.2 million new businesses in the first half of 2026. But fewer of those businesses plan to hire workers.
TakeawayMore people are starting something. It’s too soon to know how many will last.
What the % means: out of every 100 hours people in these jobs spend working, how many go to tasks text AI could do (green) or is doing (dark, in the next step).
Two research teams at Anthropic, an AI company, measured how much of the work in different jobs a machine can do today. The first team studied text AI: AI that reads and writes, like ChatGPT or Claude.
Tap one box. You can change your pick before you hit the button.
Tap any label to compare groups. Use the checkboxes to hide a layer.
Same % as before, with one change. The small dark shape counts only work text AI is doing on the job today, not what it could do. It comes from millions of real work conversations with Claude.
The gap between the two shapes is the story. Text AI could do most office work. So far, it is doing a small slice of it.
What the % means: out of every 100 hours people in these jobs spend working, how many go to tasks a machine could do. Green counts text AI only. Orange counts text AI or robots.
Six months later, a second team asked a different question: what if we count robots too? A robot here means a machine that senses and acts on its own, like a self-driving car or a warehouse robot.
Which group of jobs do you think changes the most when robots are added?
Tap one box. You can change your pick before you hit the button.
Text AI and robots expose almost opposite kinds of work. Put together, text AI or robots could do tasks that fill about 81% of all work hours. No job group is under 40%.
These are the ten most exposed jobs from each report. Read both lists. Then look for a job that shows up on both.
% = out of every 100 work hours in this job, how many go to tasks text AI is doing today (the dark shape, not the green).
This list uses a score, not a %. It runs from 0 (no robot can do any task) to 3 (robots can do every task anywhere). The tag says where a robot can already do the example task.
Before you decide what these charts mean, ask the three questions you have been asking all course.
Anthropic builds Claude, an AI product. Its own economists wrote both reports, using Claude's own usage data. That does not make the numbers wrong. It does shape what they can see. They can see their users. They can't see people using other AI tools, or people who never use AI at all.
They measured tasks, not jobs. A job is a bundle of tasks. "Exposed" means a machine can do some of the bundle. Nobody here measured who lost a job. That is a different study.
Nobody knows what 2030 will look like. Economists working with Anthropic, an AI company, built a tool to explore three possible futures. As you read them, ask: who made this, and what might they want?
AI matters about as much as the internet did. The economy grows a little faster. Most people barely notice.
AI does half of all office work. That’s a bigger change than the internet or the railroad. Office workers’ pay stays flat while other workers’ pay grows.
AI beats people at most office work. The economy booms, but many office workers lose jobs and their pay drops. Pay for hands-on workers rises by about a third. The tool’s makers say this would take AI that can improve itself.
Before you click, here are three words you’ll see:
Now make your five predictions:
The tool shows you the economy your predictions would create, and how your guesses compare with other people’s.
Wheel numbers are read from two published charts and rounded to whole percents; a few groups may be off by a point or two. Sources: Massenkoff & McCrory, Labor market impacts of AI (Anthropic, March 2026) and Legate-Yang & Massenkoff, What work can robots do? (Anthropic, September 2026). Task ratings for text AI come from Eloundou et al. (2024), Science. Job and pay data come from the U.S. Bureau of Labor Statistics.