There is a need for policies that redefine the social covenant, and they are needed at multiple levels: national and international.
Some concrete ideas.
1. Understand the type of shock: it is not only that “jobs are missing”, but that tasks are changing
Recent economic research shows that AI tends to:
- automate tasks (displacement effect),
- but also create new tasks and new occupations (reinstatement effect) – and the net outcome is ambiguous.
Evidence from the ILO and the OECD indicates that:
- today we do not yet see a clear net collapse in jobs due to AI,
- but a non-trivial share of occupations (around a quarter in OECD countries) is at high risk of automation; the adjustment could be rapid and painful.
In the scenario you propose, let us assume that the pendulum has swung decisively towards displacement. In that case the state has three major levers:
- how the income produced by AI is distributed,
- who bears the risk of the transition (the individual worker or society as a whole),
- how working time is organised and how new “useful” but currently unpaid forms of work are created.
2. National labour policies: mitigating the trauma, not slowing progress
a) Working time reduction and job sharing
If productivity soars while labour demand falls, an obvious response is:
fewer hours for each person, with the same (or almost the same) overall pay.
Possible instruments:
- shorter working week (32–34 hours) with:
- collective bargaining that redistributes existing jobs,
- public support (reductions in social security contributions) to compensate firms during the transition;
- work-sharing schemes in the style of Kurzarbeit, but designed for the long term, not only as anti-crisis tools.
The OECD stresses measures that facilitate adaptation and resilience in labour markets (more flexible working hours, income protection, active labour market policies).
This does not “create” jobs out of thin air, but spreads labour demand over more people, easing mass unemployment.
b) Active labour policies, training and large-scale retraining
Here the risk is the rhetoric of “reskilling” as a magic panacea. However, some measures are serious:
- mandatory continuous training co-financed by the state and firms, with an individual right of workers to a certain number of training hours per year;
- redeployment towards sectors that are less automatable (care, education, environmental maintenance, personal services, green infrastructure). ILO and OECD data show that jobs with a strong relational, physical and care component are less exposed to full automation;
- strengthened public employment and career guidance services, with intelligent (and controlled) use of AI for job matching.
Here the problem is one of scale: if unemployment really explodes, it is not a question of “retraining a few thousand people”, but of entire segments of the labour force.
c) Strengthening (not dismantling) labour protections
There is a familiar temptation: making the labour market more “flexible” to absorb the shock. But with AI there is a risk of getting the worst of both worlds: a few stable, highly skilled jobs, a mass of precarious work, and many people pushed out altogether.
Possible policies:
- extending social protection to those working on platforms, in on-demand work, and in the “micro-jobs” created by the AI ecosystem;
- regulating the use of AI in the workplace (worker evaluation, automated dismissals, surveillance), as suggested by the ILO debate on job quality under AI.
3. Income and welfare: from traditional benefits to forms of basic income
If, despite everything, there are fewer jobs for a long period, we need to answer the question: “how do people live?”.
a) Basic income / guaranteed minimum income
The debate on universal basic income (UBI) is now explicitly intertwined with the idea of technological unemployment from AI.
Possible forms:
- pure UBI: the same amount for all residents, financed by:
- taxation of profits and rents linked to AI and automation,
- taxes on capital and wealth,
- carbon taxes and digital taxes, etc.;
- selective basic income: similar to the pilot programmes of guaranteed minimum income that several countries and cities are experimenting with (such as the guaranteed minimum income programme in Cook County, Greater Chicago).
The pros:
- it puts a floor of security in a context of high employment uncertainty;
- it reduces the disciplining power of mass unemployment.
The cons (as recent critical literature shows):
- risk of becoming a “band-aid” that legitimises the economic power of those who control AI, shifting attention away from structural reforms (taxation, antitrust, workers’ rights);
- very high fiscal costs if it is genuinely universal and substantial.
In practice, states are likely to move towards hybrid forms:
- a robust minimum income,
- income top-ups for low-wage workers,
- free or almost free public services (healthcare, education, transport, digital connectivity).
4. Taxing where value is generated: robot tax, AI windfall profits, capital
If AI destroys jobs but greatly increases profits and productivity, the key issue is taxation.
Possible avenues:
- taxation of AI-related windfall profits (modelled on windfall taxes in the energy sector, but applied to big tech and companies benefitting from dominant positions in AI models);
- more progressive taxes on capital and rents, to offset the shift in income from labour to capital that the literature on automation and AI considers highly likely;
- the famous (and somewhat caricatured) robot tax: in practice you do not tax the robot as a subject, but you adjust the tax system so that automation is not artificially more advantageous than employing human labour because of differences between labour taxation and capital depreciation.
Without this reform, any basic income or active labour policy risks being chronically underfunded.
5. Industrial policy: creating “new” work instead of merely patching up the old
Studies on technological transitions show that the point is not whether jobs “disappear”, but whether new sectors with high intensity of human labour emerge.
The state can:
- support sectors with high intensity of non-automatable human labour:
- personal care, community health services, education, culture, urban regeneration, ecological transition;
- promote cooperative and social enterprise models in which AI is used to increase service capacity, not to cut staff to the bone;
- make subsidies and soft loans conditional on:
- employment plans,
- ethical codes on the use of AI,
- worker participation in technical decisions (company AI committees).
6. The international level: avoiding a race to the bottom
If each state plays on its own, the classic race to the bottom problem emerges:
- cutting taxes to attract AI hubs,
- weakening labour standards to “stay competitive”,
- offering huge subsidies to big tech.
At least four fronts of international cooperation seem necessary.
a) International tax coordination on AI
Building on the OECD/G20 agreement on a minimum corporate tax, it would be possible to:
- define shared principles for taxing profits derived from AI (algorithms, models, cloud services, platforms),
- limit tax arbitrage between jurisdictions, which would otherwise prevent individual states from financing robust welfare and active labour policies.
This is not science fiction: the precedents on digital taxes and on the global minimum tax show that a basis for cooperation already exists.
b) Global standards on labour and the use of AI in the workplace
Organisations such as the ILO and OECD are already developing guidelines on AI, work, surveillance, bias and job quality.
At the international level, we could aim for:
- specific ILO conventions on the use of AI in the workplace (impact assessments, transparency, trade union participation);
- common principles on:
- the right to disconnect from AI,
- a ban on fully automated dismissals,
- limits on algorithmic surveillance.
This does not “create” jobs, but it prevents the transition from taking place mainly through extreme precarisation and the erosion of rights.
c) International transition funds and support for the most vulnerable countries
If AI reshapes global value chains, some countries (especially exporters of low-skilled labour) risk much more abrupt employment shocks.
We can imagine:
- multilateral funds (under the aegis of the World Bank, regional development banks, or new instruments) to finance:
- massive training programmes,
- the development of public digital infrastructure,
- income support during transition phases;
- trade agreements that include social clauses on AI, preventing global competition from being played out through the sacrifice of workers who can be replaced by AI.
d) Global governance of data and models
Since the value of AI depends on data and models, the distribution of employment benefits also depends on who controls these resources.
At the international level:
- rules on fair access to public data, interoperability, portability,
- safeguards against de facto monopolies over basic AI infrastructure,
- commitments to use AI also in projects with high intensity of “socially useful” work (global health, climate adaptation, education).
Without this, the countries that control platforms and cloud infrastructure capture a disproportionate share of the benefits, while others bear mainly the negative employment effects.
7. In summary
If AI really generates mass unemployment, a state – and states together – must decide at least three things:
- how much human work they want to retain as a central dimension of social life (and then solutions go in the direction of shorter working hours, “labour-friendly” industrial policies, job guarantees, etc.);
- how much income they want to detach from work, through basic income or much more generous social security systems;
- how much power they are willing to take away from the actors who currently control AI platforms and models, through taxation, antitrust and international standards.
All the technical measures – reskilling, UBI, robot tax, ILO standards, tax coordination – are ultimately variations on these underlying political choices.