
Research Economist @AnthropicAI | PhD student @MITEcon | Prev @UChicago
San Francisco, CA
Joined March 2021
- Tweets18
- Following332
- Followers423
- Likes125
Rating exposure over the past 50 years shows how robots have advanced: each year, they have become able to do 2% of the physical work they previously could not.
We also find that robot-exposed jobs saw wages and employment fall compared to other jobs in the same industries. 5/6
1
15
1,688
Cost also limits robot adoption.
We estimate total annual costs for robots to produce what human workers do in a year on a task.
Robots are cost-competitive for just 0.3% of work today. It will take 40 years to reach 10% if price declines follow past trends. 4/6
Capabilities are a big limiting factor. Robots can do 74% of physical tasks, or 34% of all tasks by working hours, but mostly in very controlled settings, unlike where humans work.
Today, driving and warehouse jobs are highly exposed. Nurses and mechanics are not: robots lack dexterity today, and some physical work is also interpersonal. 3/6
Our premise: robots are more likely to affect jobs in the future that they can already do today.
Using Claude, we measure exposure by where robots work. Robots usually need controlled environments. AI is bringing them out of factories and to warehouses and construction sites. 2/6
New report with @AnthropicAI: What work can robots do today?
We find that robots extend AI exposure far beyond digital work. Over 80% of working hours in the US are exposed to robots or LLMs, but robots are cost-competitive for just 0.3%.
If robot price declines follow past trends, it will take 40 years to reach 10%. 1/6





