Skip to content

The human impact

This is ultimately about people.

Each question below sets out competing perspectives with their evidence. We label the shape of each argument, not who is right. Where the evidence runs out, we say so.

01

Will people work fewer hours?

Keynes expected a 15-hour week by 2030. Productivity has risen enormously since 1930, but working hours have fallen far less than he predicted. Shorter-week trials show the idea is workable in some organisations.

Optimistic case

Productivity can buy time

Keynes argued that compounding productivity would solve 'the economic problem' within a century. In a six-month UK pilot, most of the 61 participating organisations kept a four-day week afterwards.

Structural view

Gains have mostly become income, not leisure

Rising living standards since 1930 have mostly been taken as more consumption — better housing, health care, transport — rather than less work. Whether AI changes this depends on bargaining power and policy, not technology alone.

Cautious reading

Pilot evidence is limited

Four-day-week pilots are self-selected organisations, mostly in office work; results may not generalise to shift-based, care or manufacturing work.

Evidence gap: No robust evidence yet on whether AI productivity gains will be taken as shorter hours.

02

Will traditional employment disappear?

Every previous wave of automation destroyed some occupations and created others. Around 60% of US employment in 2018 was in job titles that did not exist in 1940. The open question is whether AI breaks this pattern.

Optimistic case

New work keeps appearing

Autor and colleagues show most of today's work is in occupations created since 1940, and that new work emerges where technology augments people.

Pessimistic case

Automation can outrun new tasks

Acemoglu and Restrepo find that since the 1980s automation has displaced tasks faster than new tasks were created, contributing to wage inequality.

Cautious reading

Employers expect churn, not collapse

Employers surveyed by the World Economic Forum expect 170 million jobs created and 92 million displaced from 2025 to 2030 — large churn, positive net.

03

Will AI create widespread unemployment?

Estimates of 'jobs at risk' vary from about 9% to 47% depending on method. Exposure is not the same as job loss: so far, the OECD finds little evidence of significant negative employment effects from AI.

Pessimistic case

High exposure estimates

Frey and Osborne put 47% of US employment at high risk of computerisation; Goldman Sachs estimated generative AI could expose the equivalent of 300 million full-time jobs globally.

Cautious reading

Task-based estimates are lower

Accounting for variation in tasks within occupations, OECD researchers found around 9–14% of jobs at high risk, with a further 32% facing significant change.

Optimistic case

Augmentation dominates so far

The ILO finds augmentation is the more likely effect for most occupations; field studies show AI raising the productivity of less-experienced workers most.

Evidence gap: Generative AI is too recent for long-run employment effects to be measured.

04

Will new professions emerge?

History strongly suggests yes, though which ones is hard to predict. In 1926 nobody could have named software developer, air traffic controller or social media manager.

Optimistic case

History says yes

New job titles have been created continuously; much of the growth came in occupations augmented by technology.

Structural view

Not automatically, and not for the same people

New occupations often require different skills and locations from the jobs displaced. The workers who lose out are not necessarily those who gain.

05

Will universal basic income become necessary?

Guaranteed income is proposed as a response to automation. Large randomised trials now exist; they show modest reductions in work and mixed effects on other outcomes.

Optimistic case

A floor for a volatile labour market

Advocates argue a guaranteed income would cushion transitions and share productivity gains if automation reduces demand for labour.

Cautious reading

Trial evidence: small labour supply reduction

A three-year US trial of $1,000/month found recipients worked about 1.3–1.4 fewer hours per week, with no significant improvement in measured job quality.

Structural view

Cost and design trade-offs

A universal payment large enough to live on is very expensive; many economists favour targeted social protection, wage insurance or retraining. The World Bank emphasises social protection reform in developing economies.

06

Who owns the productivity gains created by AI?

Whether AI raises wages, profits or both depends on whether it complements or replaces workers, and on market power and policy.

Optimistic case

Large gains possible

Goldman Sachs estimated generative AI could eventually raise global GDP by around 7%.

Pessimistic case

Gains may be modest and unequally shared

Acemoglu estimates modest total productivity gains over the next decade and warns of a shift from labour to capital.

Optimistic case

Novices may gain most

In customer support, AI assistance raised productivity most for novice and low-skilled workers, compressing skill differences.

07

Will education systems become obsolete?

AI changes what is worth learning and how it is taught, but evidence points to transformation rather than obsolescence.

Cautious reading

Human-centred adaptation

UNESCO's guidance recommends a human-centred approach to generative AI in education, including a minimum age and teacher training.

Structural view

Skills churn is accelerating

Employers expect 39% of workers' core skills to change by 2030, putting pressure on initial education to teach adaptability.

08

Will people need to retrain throughout their lives?

Most likely yes — and those most at risk of displacement tend to receive the least training today.

Structural view

The training paradox

OECD research finds workers in jobs at high risk of automation are less likely to participate in training than those in low-risk jobs.

Cautious reading

Displacement has lasting costs

When telephone switching was automated, incumbent operators were more likely to leave work altogether, while younger cohorts moved into other occupations.

09

Will human creativity become more valuable?

Generative AI can produce text, images and music at near-zero marginal cost. That may devalue routine creative production while raising the value of taste, originality and live human performance.

Optimistic case

AI amplifies creators

Experimental evidence shows AI assistance substantially speeds up professional writing tasks and raises quality.

Pessimistic case

Exposure is high for creative production

LLM exposure is higher for many writing- and design-intensive occupations than for manual work.

Evidence gap: Little evidence yet on long-run earnings of artists and writers in an AI-saturated market.

10

What happens to identity when work is no longer central?

Work provides income, structure, status and social contact. Losing it has well-documented psychological costs.

Cautious reading

Unemployment harms wellbeing

Meta-analyses show unemployment is associated with worse mental health, partly causally.

Optimistic case

Meaning can come from elsewhere

Keynes imagined people learning to use leisure well; whether that is possible at a societal scale is unknown.

Evidence gap: Evidence comes from involuntary unemployment, which may not resemble a world of chosen leisure.

11

Could people live in a world where machines perform most productive activity?

Technically speculative, economically unclear. This site's 2126 accelerated scenarios explore it, labelled as speculation.

Optimistic case

The accelerated view

If AI and robotics continue to improve, most routine cognitive and physical tasks could become technically automatable within decades.

Structural view

Bottlenecks and demand

Even with capable machines, adoption is limited by cost, regulation, accountability and the fact that people value human-provided services.

Evidence gap: No evidence base exists for 100-year outcomes; treat all such claims as scenarios.

12

Will wealth inequality increase or decrease?

Historically, computerisation contributed to rising wage inequality. Generative AI could cut either way.

Pessimistic case

Automation has widened gaps

Acemoglu and Restrepo attribute a large share of the rise in US wage inequality since 1980 to task displacement by automation.

Optimistic case

AI might compress skill gaps

Several experiments show the biggest gains for less-experienced workers.

Structural view

Exposure differs across countries

The IMF finds AI exposure is higher in advanced economies, which also have more capacity to benefit.

13

How will military capability and warfare evolve?

Drones and AI are reshaping reconnaissance and strike, while the infantry remains central. Rules on autonomy in weapons are contested.

Structural view

Uncrewed systems are central

Analysis of the war in Ukraine highlights pervasive UAV use for reconnaissance and targeting, alongside the continued centrality of infantry and artillery.

Cautious reading

Human control must be retained

The ICRC calls for binding rules prohibiting unpredictable autonomous weapons; US policy requires appropriate levels of human judgement over the use of force.

14

What happens to developing economies?

Factory jobs were the traditional ladder out of poverty. Automation and premature deindustrialisation may narrow that path.

Pessimistic case

The manufacturing ladder is shortening

Developing countries are reaching peak manufacturing employment earlier and at lower incomes than early industrialisers.

Cautious reading

Lower immediate exposure

The ILO finds generative AI automation exposure is much lower in low-income countries, where more work is manual and agricultural.

Optimistic case

Invest in human capital

The World Bank argues that investment in human capital and social protection can let developing countries benefit from new technology.

The future of education

What people will need to learn.

Across the 147 profiles, these are the new kinds of task the model expects to emerge. They indicate where training and qualifications will need to move — toward directing, verifying and taking responsibility for machine work, and toward the human skills machines handle worst.

Non-routine analytic

26

Judgement & accountability

6

Routine physical

6

Routine cognitive

5

Creative production

4

Non-routine physical (unstructured settings)

3

Interpersonal & care

3