Building Community Infrastructure for the Age of Artificial Intelligence
A report on creating resilient communities through shared AI prosperity
Dominic Atkinson, CEO
September 2025
Last updated: 9th September 2025
Executive Summary
The age of artificial intelligence has arrived not as a distant future possibility, but as a present reality reshaping the economy at unprecedented speed. While public debate focuses on whether AI will be beneficial or harmful, a more immediate question demands attention. How do we ensure that the massive forecasted productivity gains from AI benefit the communities whose collective knowledge trained these systems, rather than flowing exclusively to technology companies and their shareholders?
This report presents evidence for an alternative between accepting mass job displacement and futile resistance to technological progress. AI dividends, that is, the systematic sharing of AI productivity gains with the communities that made AI possible, represent both a moral imperative and a practical necessity for maintaining social stability during this transition.
Displacement reality
Major professional service firms are systematically eliminating entry-level positions while investing billions in AI automation. The Big Four accounting firms cut graduate recruitment by 11-29%, with collective job advertisements dropping over 44% in 2023 1 2. Goldman Sachs projects 300 million jobs globally face automation 3, with 44% of legal work and 25-50% of many professional occupations replaceable by current AI technology 4.
Unlike previous technological transitions that evolved over decades, this displacement can occur within months of new AI capabilities becoming available, bypassing traditional adaptation mechanisms.5
Legal foundation
Recent court rulings establish that AI companies can use humanity's collective creative output for "societal benefit" under fair use doctrine. A 2025 U.S. district court opinion (Bartz v. Anthropic) held that training on lawfully obtained books was fair use and grounded its analysis in copyright’s progress clause. This legal reasoning creates a profound obligation. If private companies can override individual property rights for societal benefit, society deserves meaningful participation in those benefits through concrete economic mechanisms 6.
Infrastructure focus
Traditional job training programmes fail during rapid technological change because they assume stable occupational categories and predictable career pathways. When required skills change faster than training programmes can adapt, individual interventions can become obsolete before completion 7
The solution requires infrastructure thinking. Building community capacity for ongoing adaptation rather than focusing only on training people for specific jobs that may disappear or radically change. Research consistently shows that communities with strong social infrastructure adapt better to all types of economic disruption through maintaining collective problem-solving capacity and mutual support networks.8
Proposed implementation framework
AI Labour Substitution Levy
Taxation on AI-driven workforce displacement, starting with modest rates for large companies, whilst exempting small businesses.
Community Career Infrastructure Fund
Formula-based distribution of levy proceeds to local communities for building adaptive capacity.
Career Ally Programmes
Training local workers in career-supporting conversations, creating distributed networks of community-embedded expertise.
1. AI displacement at unprecedented speed

Professional services lead the transformation
The most visible evidence of AI's employment impact appears in professional services, where major firms are systematically eliminating entry-level positions while investing billions in automation. KPMG cut graduate recruitment by 29%, reducing positions from 1,399 to 942. Deloitte reduced hiring by 18%, EY by 11%, and PwC by 6%. Collectively, UK Big Four firms slashed job advertisements by over 60% in 2023, with PwC's advertised positions falling from 1,996 to 764 9.
These cuts coincide with massive AI investments. PwC allocated $1 billion 10 over three years for AI development, partnering with Microsoft and OpenAI to deploy ChatGPT-4 enterprise across all 65,000 US workers, whilst simultaneously cutting 1,800 employees in September 2024. 11
EY eliminated 3,000 positions in April 2023 12 whilst automating 50% of bank audit confirmations 13.
Since ChatGPT's November 2022 launch, entry-level job vacancies dropped 31.9% 14 across professional services, with entry-level positions now comprising just 25% of the job market, down from 28.9% in 2022. IT roles decreased 54.8%, finance positions fell 50.8%, and Wall Street firms are considering reducing analyst hiring by up to two-thirds 15.
Four decades of concentrated gains
86%
Productivity increase
In the U.S. since 1979
32%
Worker compensation
Real growth over the same period
120:1
CEO-to-worker pay ratio
In FTSE 100 companies (2023)
39%
YOY CEO Pay Growth
Median worker pay rose 6%
Current AI displacement accelerates a four-decade pattern of productivity gains flowing to capital rather than workers. In the U.S., since 1979, productivity increased 86% (gross) whilst typical worker hourly compensation rose only 32% 16. The UK shows a related, but distinct pattern. Looking at typical workers’ cash pay, 1981–2019 labour productivity (GDP per hour) rose about 87% while the median hourly wage rose about 62% (median cash pay), showing an overall decoupling of roughly 25 percentage points. Most of the gap reflects rising wage inequality, with a large additional share due to growth in employer-provided benefits (notably pensions) that don’t appear in take-home pay, with small residual statistical differences largely offsetting. 17
FTSE 100 CEO pay was around 120 times the median UK full-time worker’s pay in 2023 (FTSE 350 median 52:1). A 2023 High Pay Centre snapshot found median CEO pay up 39% year-on-year while median worker pay rose ~6%. 18 On national accounts, UK corporations’ gross operating surplus was ~22% of GDP in 2021 (vs ~19% in 1980), while the labour share has hovered near 59% in recent years. 19
Broken social mobility pathways
The squeeze on entry-level professional roles risks weakening a key engine of social mobility which open early‑career pathways into higher‑status occupations. Recent UK evidence shows senior roles remain heavily skewed toward those from advantaged socio-economic backgrounds (SEB). In the Senior Civil Service, 18% of leaders are from working-class/low-SEB backgrounds, while 72% are from high-SEB backgrounds (versus 37% high-SEB in the wider UK workforce). Broader Social Mobility Commission analyses similarly document persistent disadvantages in access to, and progression within, professional occupations by socio-economic background, underscoring the importance of maintaining fair entry routes. 20 21 22
Access to prestigious professions increasingly hinges on internships that are often unpaid and accessed via family or social contacts, excluding many working‑class graduates. Recent Sutton Trust research finds only 11% of internships are openly advertised, with 20% obtained through family or friends; working‑class graduates are less likely to secure internships and more likely to be unpaid, frequently citing affordability as a barrier. 23 These early roles have historically acted as on‑ramps to professional careers by providing workplace socialisation, skill development, and networks. Yet the ladder is increasingly broken. Analyses of job postings show about 35% of “entry‑level” roles require three or more years of experience, and a UK study of 17,815 adverts finds 37% demand prior experience (averaging 2.5 years). 24
Global scale and speed
This pattern extends globally. 41% of companies plan workforce reductions by 2030 due to AI 25, while McKinsey projects $4.4 trillion in productivity gains from AI adoption. 26 In legal services, 74% of hourly-billed tasks face AI automation. 27 NHS trusts report AI radiology systems delivering a 12% efficiency gain and 45% accuracy improvement. ​​28​ Global manufacturing faces potentially 20 million job displacements by 2030 ​29​.
The convergence reveals a transformation already underway. AI enables systematic job displacement whilst legal and economic frameworks ensure all productivity benefits flow to capital owners, creating breakdown of traditional pathways to prosperity whilst technology companies capture economic value created by humanity's collective creative output.
2. The legal foundation: Society's claim to AI dividends
The Anthropic ruling's revolutionary logic
The 2025 federal district court opinion in Bartz v. Anthropic PBC 30 did more than resolve a dispute about AI training and copyright. Judge William Alsup held that using copyrighted books to train an LLM is fair use, calling it “quintessentially,” indeed “spectacularly,” transformative, and grounded the analysis in copyright’s purpose of promoting the progress of science and the arts. By contrast, he rejected fair use for Anthropic’s central library of pirated books, while separately finding that scanning lawfully purchased print books into digital files for internal library use was fair use (on different grounds). The court also stressed that model outputs were not at issue.
A policy question becomes worth exploring. If AI developers can rely on society’s creative output for uses deemed to advance such progress, should the public more directly share in the resulting benefits?
The judge analogised training to “training schoolchildren to write well”, even if that leads to more competing works, that isn’t the kind of substitution the Copyright Act polices. In effect, the ruling treats training on copyrighted books as a lawful, non-remunerated use. When the economic returns then flow chiefly to private entities, there’s a legitimacy gap, strengthening the case for considering collective-remuneration schemes.
Legal hooks that support public-benefit arguments (UK & US)
1
1988
UK Copyright, Designs and Patents Act includes narrow fair-dealing exceptions with limited public-interest safety valve 🇬🇧
2
2015
Authors Guild v. Google: book search/snippets are highly transformative; they augment public knowledge. 🇺🇸
3
2017
Digital Economy Act includes targeted copyright updates (library e-lending, online enforcement) but no explicit general balancing framework. 🇬🇧
4
2025
Bartz v. Anthropic: training fair use; scanning purchased books permitted; outputs unresolved; pirated library rejected. 🇺🇸
The Anthropic decision builds on a line of fair-use precedents that treat non-expressive, information-revealing uses as transformative because they augment public knowledge. For example, Authors Guild v. Google (2d Cir. 2015) 31, which found Google’s book-search functions ‘highly transformative’ for providing otherwise unavailable information about books.
Courts and lawmakers already use “permission with conditions” in some areas. Government can use private property if it pays (takings rules), and in patents the government can use an invention with reasonable compensation, and courts don’t always block use. They can award money instead of an injunction.

In UK copyright, there are narrow, purpose-based exceptions (e.g., research, quotation, non-commercial text-and-data mining) and targeted updates (e.g., library e-lending, online enforcement) that show how policy balances private rights with public benefit. None of this makes benefit-sharing for AI training automatic (the Bartz v. Anthropic ruling treats training as lawful fair use without payment) but these examples show a policy tradition of allowing use with compensation or obligations when public value is at stake. On that analogy, it’s reasonable to argue for AI benefit-sharing (e.g., transparency, a levy or fund tied to model revenues/compute, or collective licensing) so that broad public inputs don’t result in exclusively private gains.
Academic consensus on collective knowledge ownership
"The entire AI community benefits immensely from historical investments into public knowledge institutions because that data provides much of the foundation for AI models."
— Greg Leppert, Harvard Law School Library Innovation Lab 32
Leading IP scholars highlight that knowledge creation builds on prior works and that copyright balances creators’ interests with the public’s interest in promoting progress. In Generative AI Meets Copyright, Pamela Samuelson argues the current lawsuits will affect “everyone who uses generative AI,” and explains that courts weigh fair-use factors with the Constitution’s progress goal in mind, with research/scholarship among favoured purposes. 33​​

Complementing this, Creative Commons’ CC Signals proposes a “new social contract for the age of AI” to build reciprocity into machine reuse, so the commons that fuels AI is matched by shared benefits. ​34
The Social Contract framework
The fair use doctrine's emphasis on societal benefit creates what legal scholars call a "social contract" framework for AI governance. If private companies can override individual property rights because their activities serve society's greater good, then the logic of that same social contract demands that society participate in the economic benefits of that greater good.
The European Trade Union Confederation has called for a new social contract for AI, arguing that collective bargaining is how the productivity gains of AI should be fairly shared with workers. It also insists on a human-in-control principle for AI at work. On this basis, the argument is that democratic legitimacy in the AI transition requires not only human oversight but also mechanisms for economic participation, so that the human knowledge underpinning AI does not translate into purely private gains. 35 36
Socialising costs while privatising benefits
The legal framework enabling AI transformation can socialise costs while privatising benefits. In the US, fair-use doctrine can permit training on copyrighted text without payment, whereas in the UK/EU, commercial text-and-data mining generally requires a licence or can be restricted via opt-out. Against that backdrop, leading developers command reported valuations in the tens to hundreds of billions (e.g., Anthropic ≈$40B; OpenAI ≈$157B). These systems can automate tasks previously done by creators, plausibly reducing demand for some roles. Society bears much of the adjustment cost (unemployment support, retraining, transition stress) while private firms capture a large share of the private productivity gains.
There is currently no general legal duty to compensate individual workers or creators for the use of their past works in model training. (US fair use doesn’t mandate payment; UK/EU rules may lead to licensing in some cases, but not a worker-compensation regime.) And fair use was never limited to “individual educational” uses. It’s a flexible doctrine that has long applied to large-scale, even commercial, transformative uses.
These doctrines reflect a longstanding balance. Some uses are allowed to promote progress, even without payment. However, when private actors lawfully draw on collective resources, it’s reasonable to argue that society should share in the gains. Framed this way, AI dividends aren’t wealth redistribution or punishment for success. They’re a pragmatic mechanism to align private incentives with public investment and consistent with the progress rationale behind fair use (though not required by it).
3. Why infrastructure beats individual training programmes

When outcome metrics become impossible
The linear model of workforce development (training to placement to measurable success) struggles when AI rapidly reshapes tasks and entry routes. Traditional metrics assume stable occupations and predictable pathways that are now in flux.
With 78% of organisations using AI in at least one function, adoption is broad. And employers now expect 39% of core skills to change by 2030, keeping job requirements in motion. Under these conditions, a simple training-to-job pipeline becomes unreliable. 37 38 39
Multiple studies document shrinking skill half-lives and faster task turnover, reinforcing a speed mismatch between learning cycles and workplace change, even as overall employment remains resilient. 40
Today’s programmes must prepare people for specific roles even as those roles shift during the training period, especially in AI-exposed occupations where early-career workers face relatively larger employment impacts. 41
The UK’s Payment by Results experience shows how narrow outcome targets can distort delivery through “creaming” and “parking” (prioritising easier cases and sidelining harder ones) precisely when holistic support is most needed. 42 43
Infrastructure creates ongoing adaptive capacity
Definition of infrastructure thinking
Infrastructure thinking treats social provisions as foundational systems that enable human flourishing, rather than discrete services subject to market logics. Eric Klinenberg’s work on social infrastructure shows why this matters. During Chicago’s 1995 heat wave, neighborhoods with stronger social ties, active streetscapes, and community institutions had far lower mortality, even among communities with broadly similar socioeconomic profiles (e.g., Little Village vs. North Lawndale). 44 45 46
The fundamental distinction
The distinction is fundamental. Transactional approaches ask: "Did this programme place someone in a job?" Infrastructure approaches ask: "Does this community have ongoing capacity to help people adapt to whatever economic changes emerge?" The first question becomes unanswerable when target jobs disappear; the second remains relevant regardless of technological disruption.
Economic connectedness
Large-scale evidence quantifies the power of cross-class friendships (“economic connectedness”). Chetty et al. find that if children from low-income families grow up in counties with economic connectedness comparable to that of high-income peers, their adult incomes rise by ~20% on average. 47 Independent summaries reach the same conclusion: economic connectedness is more predictive of mobility than proxies such as school quality, job availability, or family structure. 48 49
Evidence from successful infrastructure models
Multiple studies show infrastructure approaches yield durable returns. In the US, public investment in community colleges delivers a $6.80 taxpayer return per $1 over students’ working lives. 50 Peer support networks are linked to better personal recovery, hope/connectedness and coping (with mixed effects on clinical symptoms) support networks in mental health and resilience programmes are linked to improved clinical and personal recovery outcomes, increased hope, reduced isolation, and enhanced resilience 51 52.
In the UK, £1m in social-infrastructure investment in disadvantaged neighbourhoods is estimated to generate £1.2m fiscal returns plus £2m in wider economic and social benefits over ten years; scaling in “primed” places yields ~£3.50 per £1. 53 54

Community anchor organisations underpin local economies (supporting local businesses and employment) and act as resilience hubs during crises, strengthening recovery. 55 56
Meta-skills over specific training
Adaptability
McKinsey identifies adaptability as the most critical skill employees can have in volatile environments, enabling workers to pivot as job requirements change.
Resilience
The World Economic Forum ranks resilience as the second most crucial core skill, helping workers navigate uncertainty and recover from setbacks.
Flexibility
Ranked third by the WEF, flexibility allows workers to adjust to new tools, workflows, and responsibilities as AI transforms workplaces.
Research consistently shows that in volatile conditions, meta-skills outrank narrow job-specific training. McKinsey calls adaptability “the critical success factor during periods of transformation,” and the World Economic Forum’s latest rankings place resilience, flexibility and agility as the number 2 core skill globally (2025; number 3 in 2023). About 67% of employers already view it as core, and ~75% expect its use to rise by 2030. 57 58
This evidence supports infrastructure over transactional investment. Building community spaces where people develop confidence, practise new skills, and maintain social connections creates ongoing adaptive capacity. Training someone for a specific role that may not exist in two years wastes resources that could strengthen social fabric helping entire communities navigate uncertainty.
The Precision Economy meets infrastructure necessity
This isn’t an argument against measurement. It’s an argument for measuring durability. Some investments maintain the social fabric through disruption rather than hitting narrow, short-run targets. As set out above, communities with strong social infrastructure tend to adapt better across shocks, not by maximising placement counts, but by sustaining collective capacity for problem-solving, information flow, and mutual support.
This infrastructure mindset underpins AI dividend policy. If our collective intellectual output trains systems that raise productivity, a portion of that surplus should fund human resilience infrastructure that helps people and places adapt to AI’s effects. The goal is not transaction-by-transaction job placement, but ongoing community capacity (trusted hubs, peer networks, digital inclusion, rapid-response coaching, and local employer linkages) that enable whatever adaptations prove necessary as work continues to change.
Accountability should focus on indicators of adaptive capacity (time-to-reemployment, earnings recovery, network strength and participation, digital access and use, and self-reported agency and confidence) rather than point-in-time placement counts.
4. Community infrastructure that works

Norway's Sovereign Wealth Fund: Resource dividends at scale
Norway’s Government Pension Fund Global is the world’s largest public wealth fund (well over £1 trillion). It takes the country’s oil and gas income and invests it for the long term, mostly overseas, so the domestic economy doesn’t overheat.. 59 60
The rules are simple and disciplined. Each year the government can use only a small, sustainable slice of the fund’s value (about 3% on average, usually a bit less). That keeps the pot growing while still funding services and investment for everyone. It’s one reason the model enjoys broad support across parties. 61 62
Crucially, the fund acts like a shock absorber. When the global financial crisis hit in 2008–09, unemployment stayed low. When oil prices slumped in 2015–16, unemployment rose but remained low by international standards. The framework let Norway spend more when times were tough and dial back when they improved, without burning through the savings. 63 64 65
What this suggests for AI dividends
  • Share the gains broadly. Treat AI windfalls as public wealth, not one-off cash grabs.
  • Save and spend steadily. Use only a small, sustainable share each year to avoid boom-and-bust.
  • Stabilise, don’t spike. Let spending rise in downturns and ease in upturns.
  • Keep it universal and trusted. Clear rules and transparency build cross-party backing.
NHS model: Universal infrastructure at scale
The National Health Service shows what infrastructure-based social support looks like at national scale. It provides access based on clinical need, not ability to pay, funded mainly through general taxation. That universality helps sustain broad public and cross-party backing. 66 67
The model aligns with the case for infrastructure thinking. A shared system that benefits everyone, regardless of individual usage in any given year. Public attitudes reflect this with strong majorities continue to support a tax-funded, free-at-point-use NHS. The House of Lords Committee on the Long-term Sustainability of the NHS (2017) concluded that, internationally, a tax-funded, single-payer model offers substantial advantages for universal coverage and overall efficiency compared with alternatives. This point is reiterated by a recent Lords Library briefing. 68
What this suggests for Career Infrastructure
  • Universal, free baseline. Entitlement by need, not ability to pay. No means tests for core support.
  • Sustainable + stabilising funding. A small, steady AI-dividend and tax top-up.
  • Build the infrastructure, not one-offs. Fund trusted local hubs, peer networks, digital access, rapid-response coaching, employer links.
  • Accountability = durability. Track time-to-reemployment, earnings recovery, network participation, digital access/use, and agency/confidence.
By the same logic, during technological transition we should prioritise infrastructure approaches funded from general taxation (shared, universal supports that help people adapt) over narrow, transaction-based schemes tied to individual payments or short-run outcome targets. The NHS of course measures performance, but access is not contingent on payment or insurance status.
Further Education college networks
Local economic impact
Research across UK FE institutions shows community-focused programmes generate up to £1.70 return per pound invested, with benefits extending beyond individual students to broader community economic development.
Funding structure
Further education in the UK is funded through government grants administered by the Department for Education, with 16-19 provision funded through national funding formulas and adult education funded through the Adult Skills Fund.
Local talent retention
51% of FE students remain in their local communities after graduation, building local capacity rather than exporting talent to distant metropolitan areas.
Further Education (FE) colleges already operate as community infrastructure. Impact studies show positive returns for learners, taxpayers and society (magnitudes vary by method and perspective), with older England-wide analyses estimating around £1.70 taxpayer return per £1 of public investment. Beyond individual students, colleges contribute to local economic development and act as anchor institutions in their areas. 69 70
In England, FE is mainly supported by government grants with 16–19 provision through a national funding formula, and adult learning via the Adult Skills Fund (including devolved arrangements in combined authorities). Colleges maintain ongoing relationships with local employers and adapt curricula to emerging skill needs, an approach now embedded in national skills policy debates. FE’s place-based model means most learners study close to home and many work locally after which is consistent with wider UK evidence that a substantial share of graduates remain in their study regions. 71

Many colleges (often via Jisc’s National Centre for AI 72) now deliver AI literacy, data analysis, and hybrid programmes that pair technical tools with human-centred competencies (communication, teamwork, problem-solving). Government-backed guidance and early-adopter case work in FE underscore responsible use and curriculum refresh, building adaptability rather than promising job-for-job substitution.
Housing Association models
Housing associations (HAs) are not-for-profit social landlords. In England there are 1,300+ HAs providing about 2.9 million homes for ~6 million people. They don’t pay profits to shareholders; surpluses are reinvested in homes and services. 73
Most HAs are independent organisations regulated by the Regulator of Social Housing, with boards responsible for strategy and standards. The regulator’s consumer objective requires opportunities for tenant involvement in management and scrutiny. Some HAs go further with tenant-led models (e.g., the Community Gateway model in Preston 74, where the organisation is member-owned and the chair must be a tenant. Phoenix in London describes itself as resident-led 75). Sector-wide, resident voice is strengthened through boards, panels and formal engagement.
HAs act as anchor institutions. They are large, place-based employers and purchasers that can keep spending local. Independent analysis for the National Housing Federation also finds sizeable economic multipliers from social housing investment and construction. 76 The HA model shows how community-rooted, non-profit infrastructure can recycle surpluses into local benefits and involve residents in governance.
That’s a useful template for AI dividend distribution. Keep capital in communities, reinvest in local capacity, and build formal roles for residents in priority-setting.
Makerspace Networks: Building adaptive capacity
“Makerspaces” are community workshops with shared digital-fabrication tools. Across global networks, they operate as peer-learning communities that build technical literacy and meta-skills (iterative design, systems thinking, collaborative problem-solving). Surveys show most spaces provide formal classes and informal help, so expertise spreads quickly as new tools arrive without needing slow, top-down curriculum rewrites. 77 78

These networks also support entrepreneurship: studies link makerspace network embeddedness to stronger business-model innovation and entrepreneurial performance, and global Fab Lab surveys report new ventures emerging from lab activity. Participants frequently form ongoing professional networks and collaborate on community projects (from assistive tech to local manufacturing), extending impact beyond traditional employment pathways. 79 80

Makerspaces are great examples of place-based infrastructure that continuously upgrade a community’s skills and networks.
Digital infrastructure innovation
UK social enterprises demonstrate how modern organisations can provide community career infrastructure through hybrid digital-human approaches. Organisations like Stay Nimble have generated over £34 million in independently verified social value whilst supporting 19,000+ individuals through innovative approaches demonstrating scalable models for AI dividend-funded infrastructure. 81
These approaches eliminate traditional geographic and cost barriers whilst maintaining essential human connection. Digital platforms provide qualified Career Development Institute-registered coaching with software that scales at low marginal cost, rather than the high fixed costs of purely in-person provision.
AI assistants (such as Ask.Nim) built on GPT-4o technology and trained on authoritative career development content demonstrate effective human-AI collaboration in career support. Deployments like those in Kingston, offer unlimited access to AI-powered career assistance for all residents, integrated with local employment resources, alongside fully funded 1-to-1 coaching offers. 82
This model builds lasting community capacity. Professional career coaches (all CDI-registered and MHFA-trained) deliver thousands of coaching conversations that seed peer networks and ongoing mutual support, complementing the always-on digital layer.
5. The AI Dividend framework: Practical implementation

Measuring labour substitution across the economy
Effective dividend policy needs practical, light-touch measures that capture AI’s impact without stifling innovation. For example, MIT economists estimate an optimal automation-capital tax of ~1%–3.7% 83 of asset value which is modest enough to preserve incentives while generating meaningful revenue.

Once a firm clears the materiality thresholds (UK-revenue + meaningful AI deployment), we keep two things simple:
What to count and How to pay.

We only count changes that happen after AI goes live, either in the workforce (roles, hours, churn) or in productivity (unit costs, output-per-worker). Those measured changes feed a small, predictable contribution, capped to protect innovation and reported once a year with light-touch assurance.
That’s why the framework has three parts; employment-based metrics to track role and hours movements after deployment; productivity-based measurement to convert audited cost/output gains into a rate-card contribution; and self-reporting with third-party auditing so it’s transparent, comparable, and cheap to administer.
1
Employment-based metrics
Use ONS-style workforce measures to track changes after AI deployments (role counts, hours, vacancies, churn). Where firms report efficiency gains and parallel reductions in specific job categories (e.g., a bank rolling out AI tools at scale while cutting costs and some roles), those movements can feed a transparent dividend formula, with causation assessed against documented changes. 84
2
Productivity-based measurement
Capture value via output-per-worker and productivity and use published benchmarks.

For example, in law, a complaint-response workflow reportedly dropped from 16 hours to 3–4 minutes (>100× on that task). B

In customer service, Salesforce says AI agents now handle about half of support interactions (roughly 1.5m AI and 1.5m human over nine months) and contributed to a ~17% reduction in support costs, alongside cuts of ~4,000 support roles as the mix shifted.

Treat figures like 50% AI coverage and ~17% unit-cost reduction as sector rate-card basis for dividend obligations, calibrated per use-case and updated annually. 85 86 87
3
Self-reporting with third-party auditing
Propose a comply-or-explain model aligned with the UK’s principles-based AI framework. 88
1. Companies document AI systems that materially affect roles or productivity
2. Independent auditors verify sampling, attribution, and calculations
3. Regulators focus on transparency, accountability, and proportionality rather than prescriptive rules.
Progressive thresholds protect innovation
Inspired by France’s and the UK’s digital services tax (DST) design, which applies rates only above revenue thresholds to avoid catching small firms, we propose AI-dividend obligations that begin only once deployments and UK market presence are material. 89
Revenue threshold
£200 million annually in UK revenue
Scale threshold
AI implementations affecting 50+ positions
Productivity threshold
25%+ measurable productivity gains from AI adoption
UK presence threshold
Significant UK-user revenue / market presence (DST-style), applied equally to foreign and domestic firms.

These DST-style thresholds (high UK-revenue bar; material deployment/productivity tests) keep small innovators out of scope while ensuring major platforms and UK-facing blue-chips contribute. (Think Amazon, Salesforce and peers, all well above £200m UK revenue.) With ~220–330 firms in scope, the productivity-based dividend yields ~£0.5–1.3bn in a base year (rising toward £2bn as adoption deepens), which would be material alongside general taxation, without blunt taxes or micromanaging headcount.
Worked example (illustrative)
Company
“Northstar Retail plc” (FTSE 100)

Inputs
AI capital base £195m (capitalised £120m + £75m expensed-equivalent); chosen rate 2% (within 1–3.7% band); audited AI savings £25m; UK revenue £12bn.
Calculation
Raw dividend = 2% × £195m = £3.9m.
Innovation cap = min(10% of savings = £2.5m; 0.2% of revenue = £24m) → £2.5m.
Payable dividend: £2.5m.
Allocation (50:50)
National fund £1.25m; Local fund £1.25m (e.g., by headcount: North/Midlands 60% → £0.75m; London/SE 30% → £0.38m; Rest 10% → £0.13m).
Reporting
One-page AI Impact Return (capital base, savings, caps, final amount) with limited assurance.
Want more detail on how this might work? Click here
Community Career Infrastructure Fund
AI dividend revenues would capitalise Community Career Infrastructure Funds routed through existing institutions (mayoral/combined authorities and upper-tier local authorities in England, and the devolved administrations) rather than creating new bureaucracies.
Formula-based allocation ensures equitable distribution whilst maintaining local control:
  • Base allocation: every area receives a minimum share based on population.
  • Need adjustment: additional funding weighted by unemployment, Index of Multiple Deprivation (IMD), and rural/sparsity factors. 90 91 92
  • Capacity bonus: a small performance reserve released where areas meet agreed delivery milestones (mirroring EU cohesion policy’s performance reserve 93).
Providing three-year settlements (like trailblazer single settlements 94) would give planning certainty, with limited transfer so places can reallocate within approved infrastructure categories without new central approvals, consistent with UKSPF-style programme rules on mix and local investment plans. 95G
By using existing delivery channels, familiar population-and-need formulas, and recognising rural access costs (while rewarding capable delivery and providing multi-year, flexible funding), the model would be fast to implement and gets support to communities quickly.
What Community Career Infrastructure funds pay for
This isn’t another round of short-lived training schemes. It’s a standing, local support system (people, places, and platforms) that helps residents adapt as work changes. AI dividends turn into practical capacity on the ground, connecting people with trusted humans at common touchpoints, always-on digital guidance with human oversight, and spaces and equipment that make problem-solving accessible.
The goal is simple. Wherever people already go (housing offices, libraries, GP social prescribing, community hubs), there’s a first line of career help that can quickly connect residents to deeper support when needed. Regional Career Ally programmes (proposed by Stay Nimble) build on Growth Hubs and Local Skills Improvement Plans to create distributed capacity embedded in existing community touchpoints.
Career Ally Programmes: scaling human infrastructure
1
Training approach
A CPD-accredited 4-hour “Foundations in Career-Supporting Conversations” micro-credential delivered through blended learning, designed and supported by CDI-registered professionals.
2
Who it equips
Housing association staff; social prescribers; community workers and library staff; volunteer coordinators and advice workers; small-business support staff; trade union representatives.
3
What Career Allies do
  • Hold short conversations to clarify goals or barriers
  • Work alongside people to use digital career tools, including AI assistants
  • Facilitate warm handovers to CDI-registered coaches for deeper support
  • Follow proper escalation for safeguarding or mental-health concerns
Funding community infrastructure, not individual programmes
The dividend’s power is channelling AI productivity gains into proven community infrastructure rather than job-by-job training that can date quickly. Funds would back:
  • Community venues offering career support alongside other essential services
  • Digital platforms providing 24/7 AI-enhanced guidance with human oversight
  • Professional coaching networks open to all community members
  • Peer support groups and mutual-aid networks
  • Equipment and space for technological literacy
The funding model creates positive feedback loops. As companies deploy AI to capture productivity gains, they automatically contribute to infrastructure helping communities adapt to AI-driven change. Using the same thresholds and rate-card approach set out above, the FTSE 100 alone would contribute ~£250–£300m a year. Extending to the FTSE 350 plus large multinationals with >£200m UK revenue lifts this to ~£0.5–£1.3bn a year in a base adoption year, with upside toward ~£2bn as AI coverage deepens. With a 50:50 split between the national fund and local allocations, that means ~£125–£150m to local projects in the FTSE-100 case and ~£250–£650m when the broader base is in scope. This unlocks steady support that grows automatically with measured productivity gains.
International coordination mechanisms
AI is global, so the rules work best when they travel. We’ve seen this before. Over 140 jurisdictions have agreed, through the OECD/G20 Inclusive Framework, to a 15% global minimum effective corporate tax, showing that countries can coordinate floors to curb a race to the bottom. 96 97 We’re also seeing the EU’s risk-based AI Act shape global compliance, which is a classic “Brussels Effect,” where firms often adopt EU-level standards worldwide to avoid running different playbooks in different markets. 98 99
The UK can start at home by setting the AI dividend with clear thresholds and light reporting, while making it fair at the border. The carbon border adjustment model (CBAM) is a reasonable template. The EU’s CBAM ensures imports face a carbon price comparable to domestic producers, and the UK has confirmed it will introduce a CBAM from 1 January 2027. An AI-dividend equivalent would work the same way. If a company sells into the UK without a comparable regime at home, it pays a small matching charge on UK sales; if it does, then there is no charge, thereby removing any incentive to relocate simply to dodge the rules. 100 101
Alongside this, the UK can offer mutual-recognition deals to countries that adopt comparable systems and use existing forums (e.g., the OECD AI Principles process) to keep definitions and paperwork simple.

In short; begin locally, level the playing field at the border, and invite others to join and avoid a race to the bottom while keeping things straightforward for businesses. 102
6. The Urgency: Building infrastructure before crisis
The window for proactive response is closing
AI capabilities and deployment are advancing quickly, creating a near-term window to build community infrastructure before the potential impacts scale. Major institutions warn that the exposure is broad. The IMF estimates about 40% of global jobs, and roughly 60% in advanced economies, are impacted by AI 103, implying large potential task shifts even if many roles persist. Early evidence already shows entry-level pressure in AI-exposed fields, with employment for 22–25-year-olds down ~13% since late 2022. 104
Microsoft’s multibillion-dollar partnership with OpenAI, Google’s push to embed AI across Search and other products, and Amazon’s deployment of 1,000,000+ warehouse robots powered by a new AI foundation model all signal systematic corporate rollout. At the same time, leading analyses suggest very large exposure to automation (e.g., 300 million full-time job equivalents globally) and the potential to automate 60–70% of work activities over time, underscoring the scale of transition rather than guaranteeing one-for-one job loss. 105 106 107 108
ChatGPT reached ~100 million monthly active users in two months (the fastest-growing consumer app at the time )while some companies now move from pilot to scaled AI deployment in months (e.g., Barclays’ 100k-employee Copilot rollout; Salesforce AI agents handling ~50% of support interactions). 109 This speed mismatch with institutional adaptation is a systemic risk. Without proactive infrastructure development, communities will face simultaneous job losses, inadequate support systems, and overwhelmed social services exactly when people most need stability.

We need to take the next 2 - 3 years as the moment to lay the foundations for community career infrastructure, so communities are ready for rapid task reconfiguration even where headline employment initially looks steady. McKinsey’s timelines (50% of activities automated between 2030–2060) reinforce that large shifts accrue over time, making early investment in adaptation capacity the prudent move 110.
Learning from past transition failures
1
1950
700,000 coal mining jobs in the UK
2
1980s-1990s
Major pit closures without adequate community infrastructure
3
1990s-2010s
Decades of economic decline in former mining communities
4
Today
Well under 2,000 coal mining jobs remain 111
The UK’s coal workforce fell from around 700,000 in the 1950s to well under 2,000 today (699 workers recorded in 2019; industry estimates put 2023 in the low-thousands at most). Communities in former coalfields have faced long-running employment deficits and slower job growth than major cities. 112 113 114
Evidence suggests areas that diversified earlier and leveraged anchor institutions (universities, hospitals, innovation assets) weathered industrial decline better than those that waited for closures and then responded. Sheffield’s multi-decade shift toward advanced manufacturing and knowledge-led services, anchored by its universities and health system, illustrates the length and coordination required. 115 116
Deindustrialisation studies show that when shocks hit communities without resilience infrastructure, the effects cascade; weaker job markets, higher inactivity and benefit dependence, and worse health outcomes. UK research links higher drug, alcohol and suicide mortality to deprived, deindustrialised areas, reinforcing the case for proactive, place-based support rather than reactive crisis management. 117 118 119
Current policy responses remain inadequate
Outdated Assumptions
Existing workforce and social-support systems assume stable occupations and gradual change.

AI is reorganising tasks and skills faster than traditional programmes update, so systems built for slow, predictable pathways under-perform.

OECD and WEF evidence shows rapid skill-mix shifts and the need to redesign adult-learning around continuous adaptation. 120 121
Individual Focus
Current policy leans on individual upskilling rather than community infrastructure.

The UK’s National AI Strategy/White Paper prioritise innovation and digital skills, for example Local Skills Improvement Plans (LSIPs) align training to employer demand and “emerging technologies,” but do not provide a framework for place-based resilience when multiple sectors shift at once. 122
Vulnerability Gap
This creates a vulnerability gap: AI deployment can move faster than support systems.

OECD cautions that institutions must upgrade quickly; early data already show entry-level pressure in AI-exposed occupations. 123
Without proactive, place-based infrastructure, communities risk simultaneous task loss and overwhelmed services just when stability is most needed.

The economic case for immediate action
Prevention is cheaper than rescue. Major studies find that every £/$1 invested in resilience/mitigation yields ~£/$4–£/$7 (and often more) in avoided losses and response costs. The World Bank estimates ~4:1 benefits for resilient infrastructure; the U.S. National Institute of Building Sciences finds $4–$6 per $1 for hazard-mitigation and recent analyses report averages around $7 per $1 across events. 124 125 126
As cited earlier, Norway’s sovereign wealth framework shows how a permanent public fund can cushion shocks. The fiscal rule ties spending to the Fund’s expected 3% real return, allowing counter-cyclical support when needed (and restraint when not). During the 2015–16 oil slump, expansionary fiscal policy helped offset private-sector weakness which is exactly the stabilising role an AI dividend fund would play for labour-market transitions.
UK headline indicators remain benign by historical standards (employment rate ~75%, inactivity easing) 127. That creates political and operational space to stand up community infrastructure before services are overwhelmed by adjustment pressures.
The bottom line is that waiting for breakdown is costlier. Emergency welfare, crisis healthcare, and disorder response costs scale rapidly when support is absent, whereas proactive community infrastructure reduces the need for costly crisis intervention and speeds recovery when shocks arrive. The benefit-cost evidence and the Norwegian stabiliser model both point to the same conclusion. Invest now to save more later.
Political windows require immediate action
Policy change often happens in brief windows when problems, solutions, and politics line up. Once a shock or public mood shift closes the window, attention swings elsewhere and opportunities shrink. Right now, AI’s perceived upsides are front-of-mind (strong investment and market momentum, and early productivity gains) creating space to discuss sharing benefits before debates harden around damage control. Think about the window created by the AI-driven market surge led by chipmakers and platforms, and studies showing sizable productivity lifts in customer support and other tasks. 128
When displacement becomes visible, politics typically turns reactive. Leaders prioritise immediate relief over long-term capacity, a pattern seen after major “focusing events.” That makes building infrastructure harder later than now.
Going early also helps internationally. Regulators that move first often set the standard others adopt, and firms generally value predictable, stable rules and policy uncertainty is linked to lower investment. Clear, light-touch rules launched now are likelier to win corporate buy-in than stricter measures introduced after a backlash.

We must treat the present as the window to lock in a simple, predictable AI-dividend framework, before crisis politics crowd out the longer-term community infrastructure the transition needs.
7. Call to Action: Funding the Future We Need
An alternative way forward
Public debate on AI swings between uncritical optimism and dystopian pessimism, missing the practical path this essay sets out. We don’t have to choose between accepting mass displacement or futilely resisting progress. AI dividends offer a way forward. When our collective knowledge helps train AI systems, communities share in the gains and build the capacity to adapt.
The evidence shows this path is both possible and necessary. Legal hooks exist. In the US, fair use can permit non-remunerated training. In the UK/EU, text-and-data-mining rules already balance progress with rights. Benefit-sharing is a policy choice built on these foundations, not mandated by them.
The economic models exist. Norway’s resource fund and the NHS demonstrate how shared funding and universal infrastructure can deliver stability at scale.
Community infrastructure works. FE colleges, housing associations, makerspaces, and social enterprises already provide place-based capacity that strengthens adaptation.
What’s missing isn’t evidence or workable models. It’s the political will and delivery discipline to implement them at the speed AI now demands.
Immediate actions for communities
Establish local AI dividend requirements (where you have leverage)
Use the levers you already control e.g. discretionary business-rates relief, local grants, land deals, and procurement. For recipients of these benefits, include a simple clause; contribute a modest share of audited AI savings to the local Career Infrastructure Fund (use our sector rate cards or the 1–3.7% AI-capital guidepost, with caps). Keep it voluntary/contractual for now (via grant/relief conditions), not a new tax.
Create Career-Infrastructure pilots (90-day set-up)
Reprogramme a slice of existing people-and-skills monies to infrastructure, not courses:
  • a coaching network (CDI-registered backbone)
  • always-on digital guidance with human oversight
  • a place-based ‘front door’ (libraries, housing offices, community hubs)
Use multi-partner MOUs and publish a 1-page logic model and durability metrics (time-to-reemployment, earnings recovery, network participation, digital use, agency/confidence).
Test “Career Ally” at community touchpoints
Run a rolling 4-hour, CPD-accredited micro-credential (“Foundations in Career-Supporting Conversations”). Train housing staff, social prescribers, librarians, volunteer coordinators, advice workers, small-business support teams, union reps. Allies do first-line support only with quick goal/barrier triage, demo digital tools (incl. AI assistants), warm handovers to professional coaches, and clear safeguarding escalation.
Secure voluntary AI-dividend commitments from local employers
Treat this like a community benefits agreement for the AI era. Convene major adopters (hospital trusts, councils themselves, universities, banks, retailers, utilities). Ask for a published commitment to contribute a share of measured AI gains to the local fund and share an annual AI Impact Returns (inputs, savings, assurance letter) with offers of in-kind support (equipment, licences, data labs).
Stand up mutual-aid and peer networks
Back peer support groups and mentor circles tied to the coaching backbone and digital platform. Keep costs low. Provide rooms, small facilitation grants, childcare/transport stipends, and a simple referral loop into professional help.
Policy advocacy priorities

AI Labour Substitution Levy
Back legislation creating a progressive, light-touch levy on material AI deployments, not on “AI in general.” Keep it aligned with the proposed model
  • Rate: small band ~1–3.7% (guided by optimal automation-tax estimates), with an innovation cap (e.g., ≤10% of audited AI savings and 0.2% of UK revenue).
  • Scope: kick in only above thresholds: £200m UK revenue and deployments affecting 50+ roles or delivering ≥25% unit-cost/productivity improvement in a major workflow.
  • Method: let firms choose employment-based or productivity-based (rate-card) calculation, with self-report + limited assurance.
  • Fairness: no blanket sector carve-outs; instead use credits for public-interest deployments (e.g., NHS, education) and for direct co-funding of local infrastructure.

Community Career Infrastructure Fund
Channel receipts through existing institutions (devolved administrations, combined/upper-tier local authorities). Use a formula with:
  • Universal base per population,
  • Need weights (unemployment, IMD, rural/sparsity),
  • A small performance reserve released on delivery milestones.
  • Provide three-year settlements with limited transfer so places can adapt locally without new approvals.

International Coordination
Start domestically, but bake in border fairness and alignment incentives by agreeing a matching charge at the border for sellers into the UK without comparable rules at home and strike mutual-recognition deals for jurisdictions adopting similar systems. Encourage convergence via existing forums.

Building the movement
This needs to be a broad coalition. Engage trade unions and community organisations (voice and legitimacy), technology workers (AI reality, not hype), local government (delivery), business leaders (predictability and co-design), and universities/think-tanks (evidence and evaluation). The aim is to agree on the four non-negotiables being small rates, materiality thresholds, innovation caps, and infrastructure-first spending, so we share gains now and build lasting local capacity before the next shock.
The timeline for action
1
Immediate (0-6 months)
Establish pilot programmes, build local coalitions, and advocate for policy development at all levels of government
2
Short-term (6-18 months)
Run comprehensive pilots; introduce the AI-dividend Bill (rates, thresholds, caps, formula); publish data/assurance standards; first payments from voluntary/local schemes.
3
Medium-term (18-36 months)
Domestic rollout to regional/national level; begin border-fairness mechanism and mutual-recognition talks; establish systematic national distribution.
4
Long-term (3+ years)
Independent evaluation; adjust thresholds/caps; expand community infrastructure; embed with skills, welfare, and industrial strategy; progress international alignment.
Every month of delay narrows the window for proactive action as AI deployment outpaces institutional adaptation; resilience infrastructure must be built before services are overwhelmed.
The future we choose
This report doesn’t claim a collapse-versus-utopia choice. AI may diffuse widely with manageable disruption. But uncertainty is high, the pace is fast, and communities need capacity either way. AI dividends are a practical way to align private gains with public investment; an insurance policy that pays off in all scenarios.
If impacts are modest, a small, capped, light-touch dividend still funds universally useful infrastructure (coaching networks, digital access, local hubs) that improves mobility and opportunity. If impacts are larger, the same mechanism scales automatically with measured productivity gains, cushioning shocks without heavy new bureaucracy.
The legal hooks exist, the economics are workable, and delivery channels are ready. What remains is the collective will to implement a measured, adaptive approach. An approach that doesn’t presume crisis, but makes us ready if it comes and leaves us better off if it doesn’t.
Appendix: Resources for Action

Key organisations (UK policy & think tanks)
  • Institute for Public Policy Research (IPPR) — practical policy on tech, work, and inclusive growth. Website: ippr.org
  • Resolution Foundation — wages, living standards, Economy 2030 (great on labour-market transition). Website: resolutionfoundation.org
  • Institute for Fiscal Studies (IFS) — tax/benefit design and incidence (levy modelling). Website: ifs.org.uk
  • New Economics Foundation (NEF) — universal services and community infrastructure. Website: neweconomics.org
  • Centre for Cities — city/regional labour markets; place-based policy. Website: centreforcities.org
  • UCL Institute for Innovation & Public Purpose (IIPP) — mission-oriented policy and public value. Website: ucl.ac.uk/iipp
AI governance & assurance (UK-led)
  • Alan Turing Institute — policy, assurance, and public-sector AI deployment. Website: turing.ac.uk
Community infrastructure & delivery partners
  • Locality — community anchor organisations; place-based resilience. Website: locality.org.uk
  • National Housing Federation — housing associations as local anchors. Website: housing.org.uk
  • Stay Nimble — UK social enterprise delivering hybrid digital-human career infrastructure (CDI-registered coaching, AI career assistant, Career Ally training) to scale local capacity. Website: staynimble.co.uk
  • Jisc — FE/HE digital capability and AI adoption support. Website: jisc.ac.uk
  • Association of Colleges (AoC) — FE system leadership and practice. Website: aoc.co.uk
International models & comparators
  • Norges Bank Investment Management (Norway GPFG) — governance of the world’s largest public wealth fund. Website: nbim.no
  • Alaska Permanent Fund Corporation — longstanding resource dividend model. Website: apfc.org
  • OECD.AI Policy Observatory — comparative AI policy and metrics. Website: oecd.ai
  • Nordic Council of Ministers — research on universal services and community infrastructure. Website: norden.org
  • International Labour Organization (ILO) — Just Transition — labour-market safeguards during tech/energy shifts. Website: ilo.org/just-transition
  • Sitra (Finland) — social innovation and tech governance pilots. Website: sitra.fi

Policy resources & templates
  • House of Lords Library & Committees (AI) — inquiries, briefings, recommendations. Website: parliament.uk
  • Local Government Association (LGA) — toolkits for local innovation and economic development. Website: local.gov.uk
  • What Works Centre for Local Economic Growth — evidence on “what works” in local interventions. Website: whatworksgrowth.org
  • Trades Union Congress (TUC) — tech and employment policy resources. Website: tuc.org.uk
  • OECD Inclusive Framework (global tax design) — coordination templates for thresholds and reporting. Website: oecd.org/tax/beps
Academic research centres
  • Oxford Internet Institute (OII) — AI governance and societal impacts. Website: oii.ox.ac.uk
  • Institute for the Future of Work (IFOW) — technology’s impact on employment and fairness. Website: ifow.org
  • LSE Centre for Economic Performance (CEP) — productivity, technology, labour markets. Website: cep.lse.ac.uk
  • The Productivity Institute (ESRC) — UK productivity and regional gaps. Website: productivity.ac.uk
  • Warwick Institute for Employment Research (IER) — skills demand, FE pathways, careers. Website: warwick.ac.uk/ier
NOTE: A number of AI tools have been used in the production of this report. Perplexity has been used for reasearch. Claude has been used to summarise reference material and has been used for supporting the editorial process. Gamma has been used for the design of the website hosting this content.