What is Universal Basic Income (UBI)?
Universal Basic Income is an unconditional cash payment made to all members of a defined population, regardless of employment status, income level, or means testing. The idea has roots stretching back to Thomas Paine in the 18th century and has been piloted in some form on five continents.
In its modern formulation, UBI is most often discussed as a response to two pressures: persistent inequality and AI-driven labour displacement. Tech founders including Sam Altman, Elon Musk, Marc Benioff, and Reid Hoffman have publicly endorsed some version of it as the answer to a future where machines do more of the cognitive work humans currently do.
The argument is intuitive. If machines do the work, people still need money. So give people money.
The evidence says it is not that simple.
Why Do Universal Basic Income Experiments Fail?
UBI experiments fail because of design, not because of math.
Every credible pilot run between 1974 and 2024 has produced the same pattern. Wellbeing improves. Stress drops. Mental health metrics move in the right direction. Then the experiment ends, the funding dries up, or the data sits in boxes for thirty years. Employment, productive activity, and long-term economic trajectory remain virtually unchanged.
That is not a failure of cash. Cash works. People who receive unconditional income live better lives while they are receiving it.
It is a failure of structure. Cash without scaffolding is a holding pattern. The dollar arrives, eases the moment, and leaves no permanent change in the recipient's capacity to participate productively in the economy.
This is the core insight from seven years of running our own experiments at GLEAC: income without human capital infrastructure is a half-intervention. The economic return on a dollar of income support is dramatically higher when that person also has access to the judgement, emotional intelligence, network, and collaboration skills that let them use stability productively.
That is why most UBI pilots produce neutral data. They are testing the wrong intervention.
The Four Anchors of a Working UBI
After studying every credible pilot and running three experiments of our own, we have arrived at a framework. A working UBI is not a check. It is a structure built on four anchors. The income is the byproduct, not the goal.
1. Purpose
You are paid for what you know, not just what you do. Income tied to identity, expertise, and contribution produces different outcomes than income tied to time served or unemployment status. Purpose is the variable that turns a recipient into a participant.
2. Community
You earn alongside other people, not in isolation. Every UBI pilot that has produced sustained results has had a community substrate underneath it. Income without community is a deposit in a vacuum. The Y Combinator Oakland pilot found that entrepreneurship rates did not increase under unconditional income because the missing variable was not money. It was the mentorship, network, and operating context that turn capital into ventures.
3. Belonging
Your contribution is tracked, recognised, and rewarded. Belonging is what converts a community member from beneficiary to producer. Without recognition systems, communities become charity. With them, they become economies.
4. Health
Income attached to purpose produces longevity outcomes that cash alone cannot. The Mincome Manitoba pilot in the 1970s produced a drop in hospitalisations and an increase in school completion. Stockton SEED produced mental health improvements. The pattern is consistent. When humans have purposeful work plus financial stability plus community recognition, they live longer and healthier. Cash alone moves the needle on stress. It does not move the needle on longevity.
Cash without those four anchors is a holding pattern. That is exactly why most UBI pilots produce neutral data on the metrics that matter most for long-term economic trajectory.
Global UBI Pilots: A Catalogue of Six Design Failures
These pilots produced real evidence. None of them were built to last. Here is what each got right, and where the design broke.
Finland Basic Income Pilot (2017 to 2018)
Location: Finland. Amount: 560 euros per month. Sample: 2,000 unemployed citizens. What happened: wellbeing improved marginally, employment rates were virtually unchanged. Why it failed: the experiment was too short, too narrow, and measured the wrong things. Income without purpose is still purposeless.
Stockton SEED (2019 to 2021)
Location: Stockton, California. Amount: 500 dollars per month. Sample: 125 residents in a lower-income city. What happened: positive mental health outcomes, the programme ended. Why it failed: replication failed because the design depended entirely on philanthropic funding and political goodwill. Not a sustainable system.
GiveDirectly Kenya (2016 to ongoing)
Location: rural Kenya. Amount: 22 dollars per month. Sample: extremely poor households in rural Kenya. What happened: consumption rose, productive activity did not scale. Why it failed: cash without community infrastructure, mentorship, or skill scaffolding produces temporary relief. Not long-term transformation.
Mincome Manitoba (1974 to 1979)
Location: Dauphin, Canada. Amount: variable guaranteed annual income. Sample: residents of Dauphin. What happened: hospitalisations dropped, school completion rose. Why it failed: the data sat in boxes for thirty years. Nobody built on it. Governments treat pilot data as the finish line instead of the starting line.
Barcelona B-MINCOME (2017 to 2019)
Location: Barcelona, Spain. Amount: up to 1,676 euros per month. Sample: combined guaranteed income with social inclusion programmes in the most deprived districts. What happened: employment effects were weak. Why it failed: the social inclusion element was underfunded relative to the cash element. Income without belonging is a half-measure.
Y Combinator Research Pilot (2017 to 2019)
Location: Oakland, California. Amount: 1,500 to 2,000 dollars per month. Sample: small-scale pilot exploring entrepreneurship and wellbeing in urban US settings. What happened: results were encouraging but not transformative. Entrepreneurship rates were no different from control groups. Why it failed: income alone does not manufacture the conditions for new ventures. Community, network, and mentorship do.
The pattern across all six: cash without purpose, cash without community, cash without health infrastructure, cash without continuity. The lever was identified. The lever was not pulled.
What Are GLEAC's Three UBI Experiments?
We are not waiting for governments to design the right experiment. We are running them ourselves. Three so far. One is still live.
Experiment 01: Equity Tokens (2020 to 2022)
Members of the GLEAC community earned tokens for engagement. Showing up. Mentoring. Contributing. Learning. We wanted to see if engagement currency could substitute for income and create a self-sustaining learning economy.
What we learned: engagement spiked, peer mentorship sessions increased significantly during the experiment, and tokens funded learning, not life. The alternative income we hoped to create never showed up.
Finding: engagement currency is real. It is not UBI on its own.
Experiment 02: Human IP NFTs (2022)
Each community member as a unique IP asset. Identity, thoughts, and contributions tokenised on chain. Anyone wanting access would pay a royalty, the way you pay to stream a song. We believed human expertise should be licensed, not given away.
What we learned: the thesis that human identity is licensable IP remains correct in our view. Timing broke the experiment. The Web3 market collapsed mid-pilot, and the regulatory environment for human-data tokenisation did not catch up.
Finding: we were too early. The infrastructure is significantly closer to ready now.
Experiment 03: Contribution-Based Data Payments (2024 to current, LIVE)
This is the experiment we believe will prove the thesis.
Payment is guaranteed for micro-practices scoring above 8/10 completed without any AI usage: $100 per year for up to 10 years for each data-lake dataset licensed to a third party that includes them. No separate revenue share applies.
What we are testing:
- How guaranteed payments of $100 per year for up to 10 years for micro-practices scoring above 8/10 completed without any AI usage, per qualifying data-lake dataset licensed to a third party, affect contributor income.
- Whether attaching a defined, time-limited payment to purpose and community creates the longevity outcomes cash-only pilots missed.
- Whether a clear 10-year payment term supports durable participation.
Live now. We will publish findings as they come.
Who Should Pay for Universal Basic Income?
This is the question every UBI conversation eventually hits. And it is the question that has stalled UBI at the policy level for fifty years.
The traditional answers are all variations of redistribution. Tax the rich. Tax the robots. Tax carbon. Tax data. Tax the wealth created by AI productivity gains. Each of these proposals has merit, and each has been politically unworkable in every jurisdiction that has tried.
The reason is simple. Redistribution requires consent from the people being redistributed from, and that consent has never materialised at the scale UBI requires. Stockton SEED relied on philanthropy and ended. Finland relied on a national government and stayed too small. GiveDirectly relies on donor goodwill and cannot fund productive activity. The funding source has been the constraint, not the design.
Our answer is to invert the question. Instead of asking who should pay, we ask what already needs to be paid for that we can attach a defined payment to when qualifying human judgement is licensed.
The answer is human judgment licensed to AI.
Are Humans in the Loop the Right Way to Fund UBI?
This is the thesis we are testing live.
Every frontier AI lab on the planet (OpenAI, Anthropic, Google DeepMind, Meta, Mistral) spends billions of dollars per year on what the industry calls human-in-the-loop infrastructure. Reinforcement learning from human feedback. Expert annotation. Evaluation grounding. Red-teaming. Judgement calibration. The market for human input into AI systems is already a multi-billion-dollar industry, captured today by data-labelling firms like Scale AI, Surge AI, Mercor, Outlier, and Invisible.
What does not exist yet, at scale, is a human context graph.
A human context graph is a structured, queryable map of who knows what, in which context, under which constraints, with which judgement. It is the relational layer that turns isolated human expertise into AI-addressable knowledge. Instead of labelling one image at a time for a one-time fee, an expert contributes a node to the graph that can be referenced by an AI system on every inference call, indefinitely.
That changes the economics fundamentally.
When a qualifying data-lake dataset containing an eligible micro-practice is licensed to a third party, contributors can receive a defined payment rather than a one-time labelling fee or inference-level revenue share. That creates a funding mechanism with three important properties:
- It connects payment to third-party licensing of qualifying datasets, not to taxpayer-funded redistribution.
- It requires no political consent. AI companies pay because they need the data, not because a legislator forced them.
- It provides a transparent, bounded benefit: $100 per year for up to 10 years per qualifying dataset, with no separate revenue share.
So yes. Humans in the loop, through a properly designed human context graph, can create a credible contribution-based funding mechanism without depending on political will.
This is what the Judgement Corpus and the current micro-practice eligibility and payment experiment are testing. The infrastructure is finally close to ready.
Frequently Asked Questions
Does Universal Basic Income work?
Universal Basic Income consistently improves wellbeing and reduces stress in the populations that receive it. It has not, in any pilot to date, produced sustained changes in employment, entrepreneurship, or economic trajectory. The intervention works for relief. It does not work for transformation.
Why do UBI pilots end after a few years?
Most UBI pilots end because the funding source (typically philanthropic or municipal) runs out, political will shifts, or the pilot was designed as a time-bound study rather than a permanent programme. The Stockton SEED programme is the clearest example: positive outcomes, then funding-dependent termination.
What is the alternative to UBI?
GLEAC's alternative is contribution-based support linked to human judgement licensed to AI. Payment is guaranteed for micro-practices scoring above 8/10 completed without any AI usage: $100 per year for up to 10 years for each qualifying third-party dataset licence. No separate revenue share applies.
How do contribution payments work?
Payments under the current Terms are tied to qualifying micro-practices and specified dataset licences. Payment is guaranteed for practices scoring above 8/10 completed without any AI usage: $100 per year for up to 10 years per qualifying third-party dataset licence; the model is not a pension, ownership transfer, or separate revenue share.
Who should pay for Universal Basic Income?
The traditional answers (tax the rich, tax the robots, tax AI productivity) have stalled UBI at the policy level for fifty years because redistribution requires political consent that does not materialise at scale. A more credible funding mechanism is human judgement licensed to AI. Frontier AI labs already spend billions per year on human-in-the-loop infrastructure. Restructured as a human context graph, that spending can support defined payments when third parties license data-lake datasets containing eligible micro-practices. The buyer is the AI industry itself, not the taxpayer.
Is human-in-the-loop AI a viable funding mechanism for UBI?
Yes. Three properties make it credible. First, it connects payment to the licensing of qualifying data rather than taxpayer-funded redistribution. Second, it requires no political consent because AI companies pay because they need the data, not because a legislator forced them. Third, it provides a transparent, time-limited benefit: $100 per year for up to 10 years per qualifying third-party dataset licence, with no separate revenue share.
What is a human context graph?
A human context graph is a structured, queryable map of who knows what, in which context, under which constraints, with which judgement. It is the relational layer that turns isolated human expertise into AI-addressable knowledge. Instead of labelling one image at a time for a one-time fee, experts contribute nodes to the graph that can be referenced by AI systems on every inference call. The Judgement Corpus is GLEAC's working implementation of this idea.
Who can partner with GLEAC on this?
Enterprises, insurers, pension funds, sovereign wealth funds, AI companies, and tech founders. One conversation starts the process. Direct outreach: sallyann@gleac.com.
