What future will AI bring: unprecedented growth or large-scale unemployment? Will everyone share technological progress, or will wealth become more concentrated?
We cannot give a definite answer yet.
Anthropic recently published research on AI and the economic future. Rather than predicting one outcome, it models three possible US economic scenarios through 2030 using different assumptions about capability, adoption, autonomy and the speed of workers' occupational transitions.
The deeper question is not only GDP growth:
If AI makes society wealthier, could people find it harder to earn income through work?
Work is a collection of tasks
We often ask whether AI will replace programmers, lawyers, doctors or nurses. The research instead decomposes occupations into tasks.
A nurse may inspect wards, draw blood, record vital signs, write discharge instructions, plan care, order supplies and speak with patients. AI can affect these differently:
- Leave them unchanged and dependent on people.
- Assist people to work faster or better.
- Complete tasks autonomously.
- Create new human tasks, such as reviewing and correcting AI decisions.
AI generally changes an occupation's internal task structure before eliminating the whole role. Assistance increases output per person; autonomous completion of more tasks may reduce demand for people in that occupation.
Capability is not the same as replacement
The research emphasizes five variables:
| Variable | Question |
|---|---|
| AI capability | How many tasks can AI complete? |
| Adoption | In how many available settings will businesses use it? |
| Autonomy | Does AI assist people or work independently? |
| Productivity gain | How much faster does a task become? |
| Labor adjustment | How long do displaced workers need to enter new occupations? |
Capability, adoption and replacement are distinct. Rapid technical improvement has limited economic impact without adoption; adoption can amplify workers rather than eliminate roles if AI mostly assists.
The sharpest disruption comes when capability, autonomy and adoption rise together while workers cannot transition quickly enough.
Three possible futures
The scenarios are mild, significant and extreme change. They are thought experiments, not predictions with assigned probabilities.
| 2030 outcome relative to a no-AI baseline | Mild | Significant | Extreme |
|---|---|---|---|
| GDP increase | 1.6% | 8.3% | 32.4% |
| Annual GDP growth | 2.4% | 5.4% | 15.4% |
| Average wage change | +0.7% | +2.1% | +9.7% |
| Knowledge-worker wage change | +0.4% | -0.3% | -11.5% |
| Other occupations' wage change | +1.1% | +5.9% | +33.6% |
| Knowledge-worker unemployment | 2.9% | 4.5% | 17.9% |
| Overall unemployment | 3.9% | 4.6% | 11.9% |
| Labor income share | 59.4% | 56.1% | 45.2% |
| Capital income share | 40.6% | 43.9% | 54.8% |
Scenario one: gradual diffusion, like the internet
The mild scenario produces real growth within the historical range of new technologies. GDP is 1.6% above the no-AI baseline in 2030, unemployment rises only 0.1 percentage points, and wages and employment composition change little.
AI is important without restructuring the entire economy. People learn to use it, roles evolve gradually and existing adjustment mechanisms broadly cope.
Scenario two: a revolution in knowledge work
AI can perform substantial knowledge work, but firms do not automate everything possible. By 2030 it affects about 20% of knowledge-work tasks, roughly three quarters autonomously and one quarter through assistance.
GDP is 8.3% above baseline and annual growth reaches 5.4%. Yet knowledge workers do not share the gains equally: employment in those roles falls about 3.9%, wages largely stagnate, while other occupations' wages rise about 5.9%.
Society becomes richer while some knowledge workers come under pressure.
Scenario three: deep economic restructuring
AI surpasses most human knowledge workers and is adopted rapidly. By 2030 it affects about half of existing cognitive work; 90% of affected tasks are automated. Its productivity in those tasks is about 2.2 times human productivity, with almost no new human knowledge tasks created.
GDP is 32.4% above baseline and annual growth reaches 15.4%. Sustained, that pace would double the economy roughly every five years.
The other side is:
- Knowledge-work employment falls 21.5%.
- Knowledge-worker wages fall 11.5%.
- Their unemployment reaches 17.9%.
- Overall unemployment reaches 11.9%.
This is not recession. It is rapid growth and unprecedented productive capacity alongside widespread loss of established jobs and income.
Growth does not mean everyone benefits
In the extreme scenario, GDP rises 32.4% and average wages rise 9.7%. Those averages can suggest universal benefit while hiding large divergence.
Knowledge-worker wages fall 11.5%, while other occupations gain 33.6%. Some incomes rise sharply while others lose employment. Society can become richer overall while many people become worse off.
Cheaper knowledge may make physical execution scarce
The model's counterintuitive result is that knowledge workers may be affected before physical workers. AI can rapidly design, code, analyze, approve and coordinate but cannot yet independently do much physical work.
Cheaper planning and design may expand construction and services, raising demand for nurses, electricians, builders and other in-person workers.
As knowledge production becomes cheaper, physical execution may become the scarce resource.
This helps explain rising non-knowledge wages in the significant and extreme scenarios. It depends on robotics not advancing equally rapidly. If physical work is also widely automated, that transition may not hold.
Distribution may become more difficult than production
Today, most people participate in economic distribution by selling labor for wages. Firms employ workers, who produce, and wages distribute some output to ordinary people.
Autonomous AI production weakens that mechanism. In the model's current baseline, roughly 60 cents of each output dollar goes to labor and 40 to capital. In the extreme scenario, labor's share falls to 45.2%, while capital's rises to 54.8%.
Despite GDP growing by about a third, total labor income barely increases; new wealth largely goes to capital owners.
The central question becomes:
How should people share wealth when human labor is no longer its main source?
Occupational transitions are difficult
Automation need not create lasting unemployment if displaced workers quickly find new occupations. But experience may lose value, skills need relearning, and transitions can change location, income and identity.
A software engineer might theoretically become an electrician, but that does not make the change achievable in a few months.
Unemployment therefore depends on society's ability to help workers adjust as well as the number of tasks automated. Training, education, matching, unemployment protection and income support all affect outcomes.
Which future will we reach?
Anthropic surveyed 10,980 US adults. Median assumptions broadly resemble significant change: substantial cognitive capability by 2030, partial adoption, a mixture of autonomous and assisted work, and roughly eight months for displaced workers to enter another occupation.
Those assumptions imply GDP about 8.6% above the no-AI baseline and overall unemployment around 4.6%. Opinions vary widely; roughly 10% of respondents give assumptions close to the extreme scenario.
No one knows which scenario is closest to reality. Current labor-market effects remain limited. That may imply mild eventual change or that we are near the beginning of a much larger transition.
A final thought
We used to ask whether society could produce enough goods and services. AI may create a different problem: strong productive capacity, but fewer opportunities for ordinary people to share wealth by selling labor.
The question is not merely whether jobs will remain:
If work is no longer necessary for wealth creation, can it remain most people's main route to income?
Technology's future is not determined by technical capability alone. Capability, adoption, human adjustment and distribution jointly shape outcomes.
The future is not written. We need to discuss not just stronger AI, but what kind of society we want when that power arrives.
---
References:
- Anthropic: What will our economic future look like?
- Economic Scenarios for Transformative AI: technical report
Thanks for reading.