The Transition Finally Gets a Budget Line

Fourteen projects. One million dollars in cash, plus up to another million in API credits. That's the size of the policy-grant announcement OpenAI made on August 17.

Some of the projects address broader ownership of productive assets, benefits that follow people between jobs, and changes to the tax base. Others concern societal resilience. I welcome the work. I've been asking for serious preparation since this series began, and I would be a tiresome critic if my response to every attempt were to explain why it didn't count.

But let's keep the amount straight. The cash is spread across fourteen projects. API credits let researchers use software; they won't pay a research assistant's salary or a participant's rent. Useful money for developing ideas, certainly. We are still a long way from providing an income to people whose work disappears.

Anthropic has put a much larger commitment on the table. The agenda it published in July for its $200 million Economic Futures Research Fund includes work on income support and workers' participation in AI-driven growth. That could finance consequential experiments. I want to know when those experiments start, what people receive, and what happens to a program that works.

A household can't spend an announcement

Suppose your job goes away while a promising transition program is being designed. Your rent remains due on the first. The proposal may be excellent. The researchers may be doing exactly the right work. You still need to get through the period before any of it reaches you.

That is the calendar I want beside every policy announcement: how long before an eligible person can actually receive help? Include the time needed to apply, establish eligibility, correct an error, and appeal a refusal. Those steps may sound like administrative detail until you're the person stuck in them.

We also need a more honest definition of success than getting somebody into any available job. A worker can be counted as successfully placed while earning much less, traveling much farther, or losing hours they used to spend caring for someone. The new job may still be the best available choice. We should know what that choice cost them.

I'd want a pilot to follow household income and stability beyond its closing date. Did the support give people enough room to learn something useful or find a better fit? Did it merely postpone the same crisis until the payments stopped? Publish that, even when it makes the pilot look less impressive.

The replicator needs an ownership arrangement

Imagine that someone invents a working Star Trek replicator tomorrow. Food and ordinary goods become astonishingly cheap to produce. Then imagine the machines belong to a handful of companies, access requires a subscription, and losing your income means losing your account.

The machine is wonderful. Being locked out of it would be miserable.

That's why I keep returning to ownership in these essays. Having access to AI can help you do valuable work. Having a claim on the income it produces can support you when you aren't doing paid work. A society that gets very good at the first and ignores the second could become extraordinarily productive while leaving a lot of people frightened about next month.

Employee ownership, public stakes, and shared investment funds are worth testing. Each needs answers about who gets included and who bears the risk. If the arrangement only benefits people who already own assets, it won't do much for the worker entering the transition with debts and a depleted savings account.

Nor do I want people told to solve insecurity by betting the rent on an AI stock. A serious distribution mechanism has to survive disappointing returns and companies going out of business. Otherwise we've replaced dependence on an employer with dependence on a share price, and called it freedom.

Decide what happens after the pilot

OpenAI says its projects will run for six months, with results reported in 2027. That gives us a timetable for the research. The next question is who will be ready to use what it finds.

Before commissioning another study, find the institution that could act on its result. What would persuade it to expand the program? Does it have the money and authority? Can the payment system cope with ten times as many applicants? We can learn something from a small trial even if it can't scale, but the limitation should be understood from the start.

Some preparation can proceed while the evidence develops. Governments can find the gaps in eligibility and test whether payments reach the intended people. Employers can cost transition support into an automation plan. We do not have to predict the exact unemployment rate to discover whether a support system can answer the phone.

People who might need the program should help design it. They will notice problems a meeting full of researchers can miss: an application that assumes stable internet access, training scheduled when children need collecting, a payment that arrives after a fee has already been charged. Those details decide whether a well-funded idea is useful.

And researchers must be free to disappoint the sponsor. AI companies can fund valuable work, but the results need to be public, including findings that make a company's preferred policy look bad. Public decisions shouldn't depend on the funder's willingness to publish an uncomfortable conclusion.

Actionable Steps

If you're funding this work, publish a timetable alongside the award. Tell us what will be delivered, when we'll see interim findings, and who could put a successful result into practice. Keep the cash figures separate from credits and other donated services.

If you're responsible for a workforce, choose a transition measure you could deliver this year. Cost it, name the person responsible, and let the people affected help test it. A modest program that reaches people will teach you things another strategy presentation won't.

I want the abundant future. I also want the person living through its construction to be able to pay for groceries. These grants are a welcome start. The next report should tell us who they helped.

LLAP. 🖖