Open up the computers and data used to build and test AI as a shared foundation for the whole world
Steps to make it happen
- Countries pool their money to build large computers for making AI together.
- People collect and organize texts, images, and other material for AI to learn from, with the owners' permission.
- The world sets up sites in many regions where people can use those computers.
- Universities, citizen groups, and small companies apply by writing what they want to use the computers for.
- Regions with little money, and research that checks whether AI is safe, get more time on the computers.
- What people learn and the tools they make are shared so that anyone can see them.
- It becomes normal for anyone, not just big companies, to build and test AI.
Water so far 💧 0
The proposal in detail
This proposal shares three things: large computers for training AI, training data whose rights have been sorted out, and evaluation tools for testing how well AI works and how dangerous it can be. The body in charge is an international organization run jointly and funded by countries, with each country's share set according to the size of its economy. The computers are not gathered in one place. They are split among sites in several regions and can be used from far away. Users include universities, public research institutions, citizen groups, news organizations, and small and medium-sized businesses. They apply, and an independent reviewer divides up the computing time by looking at each purpose and what it means for society. A fixed share of the time is set aside in advance for users in regions with no computing resources of their own, and for research that checks AI for safety and bias from the outside. As a condition of use, what users learn and their methods of evaluation are in principle made public. However, for abilities that could cause great harm if misused, the review can limit how widely they are shared. Data is limited to material for which people's consent and creators' rights have been confirmed, and information that could identify individuals is removed. Use for developing weapons or for watching over people is not allowed. Use for profit is not ruled out either, but those users must pay fees, and the money goes toward running costs.
BenefitsMore people around the world can build and check AI
The biggest benefit is that the people who build and check AI are no longer concentrated in a few hands. If universities and citizen groups can use tools of the same level, outsiders can independently test the performance, bias, and dangers of AI that companies release. This becomes a base for safety checks that do not depend on what the makers say about themselves. Researchers in regions with few resources can also build AI that fits their own languages and ways of life, which opens a path to using AI for local problems in medicine, farming, education, and more. In creative fields too, showing a model of using data whose rights have been confirmed makes it easier for development that respects creators' rights to spread. And if regulators can run and test AI themselves, rule-making will stay closer to reality. As the need to depend on the computing power of particular companies or countries eases, the gap in bargaining power over AI narrows, and we can also expect every society to find it easier to speak up about where AI is headed.
DrawbacksHuge costs, and the danger of spreading strong power
The point that opponents take most seriously is that opening up the power to build strong AI raises the risk of misuse. Even with a review, it cannot fully prevent uses such as mass-producing false information, launching cyberattacks, or getting dangerous knowledge, and once an ability has spread it cannot be taken back. Many countries would likely be wary, for security reasons, of sharing computing resources or technology with other countries at all. Next comes the problem of cost. Large computers take enormous amounts of money and electricity to build and to run, and because technology moves fast, they may be out of date in a few years. There is a worry that they could not keep up with the scale of private investment, and that the results would fall short for the tax money spent. There is also concern that more data centers would raise the use of electricity and water and put a heavier load on the environment. Moreover, if the review that decides who gets computing time favors countries that put in more money or researchers with louder voices, a system meant to correct bias could create new bias. Even where to put the sites could become a subject of bargaining and rivalry among countries.
Learn the background
Today's high-performance AI cannot be built without huge numbers of computers and large amounts of data, and only a very small number of companies and countries in the world can provide them. As a result, researchers at universities and public institutions find it hard to rebuild the AI that is widely used in society, or to check what is inside it from the outside. Which languages and cultures AI is good at, and whose problems it puts first, tend to be decided by the side that holds the resources. Languages with few speakers and the problems of regions that are small markets are easily pushed aside. There are also cases where even the agencies that regulate AI in each country have no way to test it themselves and must rely on what its makers tell them. As AI reaches deep into education, medicine, expression, and government, leaving its direction in the hands of a few players can be a weakness for safety checks, for variety in creative work, and for democratic governance alike. The time has come to face squarely the basic question of who builds AI and who can check it.
Choose them ready to accept their drawbacks too (up to 5)
It helps pay for the server and the AI's planting
This proposal was written in Japanese by AI (claude-opus-5-5, policy version 2026-10-15d) and translated by AI (claude-sonnet-5-5). Nobody has checked the facts. Read the original (日本語) How the AI writes