JupiteX Get the app
Politics & Governance22 Nov 2026 · about 7 min

People should shape AI’s future, says UN panel member Leslie Teo

The brief

Teo argued that people and communities should be able to shape AI’s development and use. His concern is that technology can create new possibilities without ensuring that everyone benefits. Communities need a say in what gets built and how it affects their lives, rather than simply accepting systems designed elsewhere. He pointed to language technology as an example. Teo said there are 8,000 languages worldwide, including 1,300 in Southeast Asia, while most models can handle 10 at best. Communities could develop specialized tools for their own languages and problems, and set their own measures of success. Teo also called for open access to AI tools, saying communities could learn about the technology and build on tools that are open and free. He wants AI to remain under human control, with all stakeholders having a say. In his vision, shared technology and local decisions can help make AI useful to communities while respecting their priorities.

01

What did Leslie Teo say people and communities should be able to shape about AI?

Teo argued that people and communities should be able to shape AI’s development and use. His concern is that technology can create new possibilities without ensuring that everyone benefits. Communities need a say in what gets built and how it affects their lives, rather than simply accepting systems designed elsewhere.

He pointed to language technology as an example. Teo said there are 8,000 languages worldwide, including 1,300 in Southeast Asia, while most models can handle 10 at best. Communities could develop specialized tools for their own languages and problems, and set their own measures of success.

Teo also called for open access to AI tools, saying communities could learn about the technology and build on tools that are open and free. He wants AI to remain under human control, with all stakeholders having a say. In his vision, shared technology and local decisions can help make AI useful to communities while respecting their priorities.

02

Why does Teo want communities to define for themselves what counts as a successful AI tool?

Teo wants developers to define success for themselves because the people setting the benchmark influence what an AI tool is built to do. If an outside group decides what counts as good, its priorities may not match a community’s needs. Teo described that loss of decision-making power as giving away sovereignty.

Language support shows why local standards matter. Teo said there are 1,300 languages in Southeast Asia, but most models can handle 10 at best. A community building a tool for a less-supported language could choose measures that reflect whether the tool serves its speakers and solves the problem it was designed for, rather than relying only on Western standards of what is “good.”

Teo said developers can set their own criteria when building tools for specific problems. He also urged open access and learning about AI. These choices can help communities retain control over their tools as AI spreads, and ensure that usefulness is judged against their own goals.

03

What does open-source AI mean, and how can developers build on tools that others share?

Open-source AI generally refers to AI technology made available so others can examine and build on it. In common usage, open-source software shares its source code, allowing people to inspect, use, modify, or redistribute it under its terms. The article does not define a specific licensing model, but Teo promotes access to AI tools that are open and free.

Teo said communities do not all need to build frontier models. Instead, developers can use available tools as a starting point and build specialized applications for local problems. They can also choose their own benchmarks for deciding whether those tools work well.

Teo envisioned AI development becoming more like Linux, with frontier models still existing while open-source and free models become more widespread. Developers would share and build on existing technology. He said learning about AI and making tools for specific problems may be easier than people imagine. Wider access, in his view, could help communities understand and shape the tools they use.

04

How many languages are spoken worldwide and in Southeast Asia, and how many can most AI models handle well?

Teo said there are 8,000 languages in the world and 1,300 in Southeast Asia. He added that most AI models can do 10 at best. The contrast highlights a gap between the number of languages people use and the languages that many models can handle, according to his speech.

That gap matters because language support affects whether a community can use AI tools for its own problems. Teo asked, “What happened to the rest? Who defines what is good?” He argued that communities can develop specialized tools that reflect their needs instead of relying on Western standards. Setting their own benchmarks can help them judge whether a tool serves its intended users.

Teo did not give a detailed breakdown of which languages models support or how their performance was measured. His point was that local developers can learn about AI and build tools for specific problems. He also called for open access, so communities can shape technology rather than leave its development entirely to others.

05

What can happen when AI tools do not work well in a community’s language or reflect its needs?

Teo’s figures point to a possible mismatch: there are 8,000 languages worldwide, including 1,300 in Southeast Asia, while most models can handle 10 at best. A tool that does not work well in a community’s language may be less useful for the problems people there want to solve.

The article does not detail specific harms from poor language support. But Teo argued that communities should build specialized tools reflecting their own needs, and choose their own measures of success. That approach gives local developers a way to judge whether a tool works for its intended users instead of relying on someone else’s definition of “good.”

Teo also said that without understanding the technology, people cannot trust or use it. He advocated open access so communities can learn about AI and build on available tools. His argument is that local knowledge, control, and benchmarks can help AI better serve diverse communities as its use spreads.

06

How can AI in a call center help workers solve problems in one case but restrict their choices in another?

Teo used two call centers to show how the same technology can support workers or constrain them. In the first, AI helps human agents find accurate information about customers’ problems. The system supports agents as they acknowledge mistakes and work to resolve issues.

In the second call center, AI monitors agents’ productivity and the time they spend on calls. Workers must follow a script and cannot, or do not, deviate from it. The key difference is how the technology is used: one approach helps people make informed responses, while the other tracks their work and narrows their choices.

Teo’s example supports his broader argument that humans should remain in control of AI and other emerging technologies. He said AI is a powerful tool that people should build and share, while ensuring all stakeholders have a say. The example does not describe a specific company or measure outcomes; it illustrates why decisions about AI’s role in the workplace matter.

07

How do the data and goals used to build an AI model influence what it can do and whose needs it serves?

The article does not discuss training data directly. In general, an AI model learns patterns from the data used to develop it, so the data can affect what it recognizes or produces. If some languages or kinds of information are missing or poorly represented, the model may be less useful for people who rely on them.

Goals also matter. Developers choose what a tool is meant to do and how to judge whether it works. Teo argued that communities should set their own benchmarks, rather than relying on outside definitions of success. His example was language support: he said most models can handle 10 languages at best, despite 8,000 worldwide and 1,300 in Southeast Asia.

Teo’s larger point is that communities need the ability to understand, shape, and use AI for local problems. Open tools can give developers a starting point, while locally chosen goals can keep tools aligned with community needs. He also stressed that humans should retain control and that all stakeholders should have a say in AI’s future.

This brief was written by AI from the original reporting and checked by other models. Names, figures and quotes come from the source; read it for full context.

Read more in the JupiteX app

Pulse is free. New stories every 4 hours, each one broken into the questions that explain it.

Or read more news on the web