PERSPECTIVE | AI, ECONOMIC DEVELOPMENT AND WORKFORCE READINESS
Perspective No. 4 · Economic Mobility
The Dividing Line Is Access, Not Exposure
Building Economic Access Policy for the Age of AI
Download this Perspective (PDF)The world's AI debate is being held over a map of exposure.
The International Monetary Fund's staff analysis of AI and the future of work estimates that almost 40 percent of global employment is exposed to AI. Exposure rises to about 60 percent of jobs in advanced economies. It falls to 40 percent in emerging markets and 26 percent in low-income countries.
The International Labour Organization's 2025 update of its generative AI research tells the same story with a finer instrument. Sorting occupations into four gradients of potential exposure, the ILO estimates that 34 percent of employment in high-income countries sits in an exposed occupation, against 11 percent in low-income countries.
So the most exposed economies on earth are also the richest ones.
Read carelessly, that looks like a warning addressed to wealthy countries and a reprieve for poorer ones. It is neither. Exposure measures how much of a job's work AI can touch. It says nothing about whether the people in those jobs can reach the tools, the training, or the time it takes to respond.
The ILO is careful on this point. Its figures describe potential exposure, not actual impact, and its researchers name the barriers that decide what exposure becomes in practice: electricity, broadband, digital skills, and the cost of the technology.
Access Is the Dividing Line
Exposure describes the job. Access describes the person's real chance to do something about it.
Access is not an announcement. A training program posted on a website is an offer, and offers are the easy part. Access is the full stack a person needs before an offer becomes usable: a working device, a connection that holds, electricity that stays on, money for the fees and for the wages given up while learning, time that an employer or a life actually allows, an entry point pitched at the level the person is starting from, and a human being to ask when the material stops making sense.
Remove any one layer, and the offer is technically open and practically closed.
We know how offers without access perform, because adult learning already runs that experiment. In the OECD's Education at a Glance 2025, drawing on the OECD Survey of Adult Skills, an average of 26 percent of adults aged 25 to 64 with low literacy proficiency take part in adult learning across OECD countries, compared with 70 percent of adults with high literacy proficiency. The OECD calls this the Matthew effect: learning flows toward the people who already have the most of it.
The same pattern is being built into the AI transition by default. In the World Economic Forum's Future of Jobs Report 2025, surveyed employers expect 59 of every 100 workers to need training by 2030. They expect to upskill 29 of those workers in their current roles and to reskill and redeploy 19 more within their organizations. They also expect 11 of the 100 to be unlikely to receive the reskilling or upskilling they need.
Eleven in every hundred, identified in advance by the employers themselves.
No exposure map shows those eleven people. An access map would start with them.
Access Looks Different at Every Level of Readiness
Access also depends on where a person is starting from, and starting points are specific to a subject.
The readiness levels this series uses, New, Capable, and Expert, describe a starting point in a subject, nothing more.
These levels describe a relationship to a subject, not a rank of a person. A hospital administrator can be an expert in healthcare operations and new to machine learning on the same morning. A data scientist can be an expert in model design and new to the regulatory process a model has to pass.
So the stack differs by level. The New learner needs the cost and the fear taken out of the first step. The Capable learner needs time, and real work to practice on. The Expert needs a route to greater depth, and standing in the system as a teacher, not only a student.
Access fails when it ignores that. A single portal pitched at the confident and the connected will serve the Expert, tolerate the Capable, and quietly turn the New away. That is the Matthew effect with better branding.
More Than One Door
Where learning is provided, format is an access decision too.
A proposed framework this series has put on the table organizes provision around four learning experiences, running from instructor-led structure to self-paced flexibility around shifts, children, and second jobs. The framework is a proposal, not an existing program or validated model, and its value would have to be established through collaborative design, testing, and evidence.
The access point stands either way. Every format quietly assumes a different stack. Solo Artist assumes a device, a connection, and the confidence to learn alone. Helping Hand assumes a schedule and a way to get to the room. When a system funds one format and calls it access, it has decided, without saying so, which people get in.
A region that wants broad access funds more than one door, and staffs the doors it funds.
What Each Institution Can Do
None of this requires a new ministry. It requires each institution to treat access as part of its existing job, and most of it can begin inside a single budget cycle.
State governments can measure access the way they already measure exposure, and publish both. A baseline that counts devices, connectivity, paid learning time, and available entry points tells a legislature more than another count of announcements. States can also aim funding at the gap employers have already named: the workers no employer plan expects to reach.
Mayors and local governments control the trusted ground. Libraries, community centers, and public facilities are where access becomes physical: a device that works, a connection that holds, and a person who can help. Local leaders can put learning access in the places residents already use and trust.
Economic development organizations can change the question they ask incoming and expanding employers. Not only how many jobs, but who gets trained, whose time is paid while they learn, and what happens to the workers whose roles change first.
Colleges and universities can design entry points at more than one level of readiness, recognize what adults already know, and offer guided formats alongside self-paced ones, instead of treating the portal as the product.
Industry can stop treating adaptation as a cost workers should carry alone. Paid learning time, honest notice of changing requirements, and real problems for learners to practice on are contributions no public program can substitute for.
Community organizations can do what institutions cannot. They can reach the people who do not walk into government buildings or admissions offices, in the languages those people speak, and stay with them past the first session.
Each role is modest. Together, they are the stack.
Counted in People, Not Announcements
The exposure numbers will keep improving, and regions will keep quoting them. They measure something real. They do not measure participation.
A worker in a low-exposure economy is not protected. She is more likely to be last in line for the infrastructure, the training, and the wage gains, and first in line when cheaper AI services arrive from somewhere else. A worker in a high-exposure economy is not doomed. She is standing where the tools, the employers, and the training budgets already are. What separates them is not the technology. It is access, layer by layer, person by person.
Publics have begun stating their own access terms, in more than one political system. Ireland, the Netherlands, and Singapore have each slowed or conditioned data center approvals over power, land, and sustainability. In the United States' November 2026 elections, across parties, the winning positions on AI projects paired support for the technology with conditions of the same kind: the builder pays its full share of grid costs, household ratepayers are protected, the host community keeps a say, agreements are public rather than secret, water is stewarded, and local people are trained for the work. None of that is opposition to innovation. It is publics defining access for themselves: if the future is built here, we intend to be able to reach it. Those conditions can be heard as friction. Or as the access agenda, written by the people it is meant to serve.
Innovation is only powerful when people can access it.
The measure of AI-era economic development is not how much exposure a region can map. It is how many of its people can actually reach the tools, the training, and the time. Build for that, and the exposure map becomes what it should have been all along: a planning tool, not a verdict.
TRANSLATION
| Policy priority | Practical first action | Evidence of progress |
|---|---|---|
| Access baseline | Publish a regional access baseline alongside exposure data: device ownership, broadband quality, paid learning time, and available entry points by readiness level | Published baseline, updated annually |
| The uncovered workers | Require every publicly funded AI training plan to name how it reaches the workers employer plans do not cover, and fund that provision directly | Funded seats and enrollment among workers outside employer training plans |
| Participation gap | Set a published target to narrow the adult learning participation gap between the lowest and highest skill groups, and report against it | Participation rates by skill level, with the gap narrowing year over year |
| Staffed access points | Fund libraries, colleges, and community organizations to run staffed access points with working devices, reliable connectivity, and human guidance, not portals alone | Operating access points, with usage and completion reported by neighborhood and population |
| Time and cost relief | Pair training offers with paid learning time agreements with employers and fee waivers or stipends for low-wage and independent workers | Share of participants learning on paid time or with cost support |
| More than one door | Require funded provision to offer at least two formats, guided and self-paced, with a human support channel in each | Format mix and completion rates by format and readiness level |
A region does not close an access gap by announcing opportunity. It closes it by funding the device, the connection, the time, and the human help that make opportunity usable.
Sources
- International Monetary Fund, AI Will Transform the Global Economy. Let's Make Sure It Benefits Humanity, IMF Blog, January 14, 2024 (data: IMF Staff Discussion Note SDN/2024/001, Gen-AI: Artificial Intelligence and the Future of Work).
- International Labour Organization, How might generative AI impact different occupations?, May 20, 2025 (data: ILO Working Paper 140, Generative AI and Jobs: A Refined Global Index of Occupational Exposure).
- OECD, Education at a Glance 2025, To what extent do adults participate in education and training?, September 2025 (data: OECD Survey of Adult Skills, PIAAC Cycle 2, 2023).
- World Economic Forum, The Future of Jobs Report 2025, January 7, 2025 (data: Future of Jobs Survey, 1,000+ employers, 55 economies).

