Bounties to Fix Our World
AI is about to free up a lot of human talent. Here’s a way to pay people to point it at the problems nobody is paying to solve.
Somewhere near you right now there’s a dumpster full of material a manufacturer down the road would happily buy, if anyone could get it clean, consistent, and delivered.
A technician knows how to do it. A lab could test it. A local shop has the equipment sitting idle. The city would love less landfill.
And still, every Tuesday, the truck comes and hauls it away.
Nobody is lazy here, and nobody is out of ideas. The problem is a boring question nobody can answer: “Who gets paid to fix it?”
That question is about to matter a lot more. As AI takes over more of the tasks people are paid for today, we’ll have more capable people with fewer obvious places to put their effort. We can use that capacity to make existing work a little cheaper. Or we can use it to finally go after the work our economy has always left on the table.
I think we should choose the second one. Bounties are a practical way to do it.
From “someone should” to “here’s the budget”
A bounty is a funded commitment to solve a specific problem. A business, government, or institution defines the problem, sets the target, commits the money, and promises to pay for verified results.
For that dumpster, it might read:
“Recover this material, meet our quality and safety specs, and show a workable cost and environmental benefit. We’ll fund development, reward verified success, and buy qualifying output on agreed terms.”
Now the problem has a sponsor, the work has a budget, and success has a customer. That last part matters most. A celebrated prototype nobody adopts is just a nice photo.
None of this is brand new. Wazoku’s Innocentive lets organizations define a problem, set criteria, and pay for solutions that meet them. HeroX lets anyone post a challenge or contribute a solution. And in its $100M Carbon Removal competition, XPRIZE awarded $1 million each to 15 milestone teams to fund their work before the final prize was decided.
So the foundation exists. The gap is in who gets paid, and how.
The trouble with winner-take-all
Most prize models ask one question: who won?
Real solutions rarely come from one person. They come from the person who lent the equipment, the facility that allowed testing, the engineer who cracked one stubborn component, the team that integrated it, and the operator who keeps it running.
Winner-take-all leaves most of those people out. It also turns earning a living into a lottery, which shuts out anyone who can’t afford to work for free and hope.
Get Scott Thornton’s stories in your inbox
Join Medium for free to get updates from this writer.
A better question:
“who helped make this work, and what did we agree they’d earn?”
One bounty, many ways to get paid
That’s the core of a variable bounty model. Different contributions, matched with different kinds of compensation, all agreed up front:
- Resource payments for equipment, materials, workspace, or computing.
- Access fees for authorized use of a site, lab, data, or test environment.
- Milestone payments for accepted components, prototypes, or trials.
- A defined share of the bounty for delivering part of the solution.
- Completion awards for independently verified results.
- Residual payments each year, for a set number of years, when a contribution keeps producing value.
- Service contracts for deployment, operation, and maintenance.
People can also choose how much risk they take. Someone lending a machine might want a guaranteed fee. An engineer might take a development payment plus a share of the completion award. A technology contributor might negotiate annual payments for five years.
The rule that holds it together: everyone knows before work starts which payments are guaranteed, which depend on success, and what rights they keep or grant.
When the benefits last, the rewards can too
Some contributions pay off once. Others keep paying off for years.
A residual can take different forms: a fixed annual amount, a percentage of qualifying revenue, a share of verified savings, or a payment tied to continued performance. These are very different promises. Revenue shares depend on sales. Savings shares need an agreed baseline. Fixed payments need real funding behind them.
Licensing offers a working precedent. Firms like yet2 help organizations find partners, negotiate terms, and manage licensing deals. A bounty framework would bring those arrangements alongside challenge funding and contributor payments, so they don’t have to be invented from scratch every time.
The guardrail: residuals need a defined term, clear reporting, and a dispute process. The aim is to reward lasting value, not create permanent claims that make good solutions too expensive to adopt.
Give AI a destination, not the steering wheel
AI can help research approaches, compare designs, write software, and analyze experiments. Institutions or providers could even let people direct computing allowances toward projects they care about.
But three things have to stay separate: choosing priorities, contributing resources, and earning money. Voting for a project shouldn’t promise a payout. Buying more computing shouldn’t buy control. Burning more AI shouldn’t count as progress unless it produces results.
And AI stays optional. Someone with practical know-how and ordinary tools should be just as welcome.
Pay for progress, verify the outcome
Two disciplines keep this honest.
First, fund the path, not just the finish line. Investigation, prototypes, testing, and implementation should earn dependable pay. Completion awards and residuals are upside, not the whole deal.
Second, verify independently. A recycling bounty shouldn’t pay for tonnes merely labeled “recovered.” It should require usable output, safe operation, a real buyer, and evidence of overall benefit. Payouts deserve the same rigor: clear calculations, auditable records, and no surprise changes after the work is done.
Key takeaways
- The bottleneck is incentives, not ideas. Neglected problems stay neglected because nobody is paid to fix them.
- Existing challenge platforms prove the concept. Innocentive, HeroX, and XPRIZE already fund problem-solving and, in some cases, milestones.
- Variable bounties widen the circle. Pay for resources, access, partial solutions, completion, and operation, not just the winning idea.
- Lasting value can earn lasting income. Time-limited residuals reward contributions that keep delivering.
- Agree first, verify always. Contribution agreements up front and independent verification at the end protect everyone.
- AI is a tool, not the price of entry. Contributing resources must never buy control or count as results on its own.
Start with one problem
You don’t need a global marketplace to begin. A local partnership could fund three bounded challenges: recover one discarded material, cut one resource loss, and extend the life of one category of equipment.
Use existing challenge platforms where they fit. Add the contribution agreements and payment tracking needed to test the variable model. Then publish everything: costs, what participants earned, what failed, and whether the fixes were still working six months later.
Bounties won’t replace stable jobs, public services, regulation, or long-term research. They can open one more road from “everyone knows this is broken” to organized, paid action.
AI may help us do more. Bounties could help us decide what’s worth doing, and give people a fair way to earn a living doing it.
