AI: WHO TEACHES MACHINES TO SEE?

Artificial intelligence often appears to be a technology that develops on its own—algorithms that can "see," "understand," and "make decisions." The reality, however, is far more human. Machines learn to see only because people patiently teach them about the world, one image at a time, one pixel at a time. Sometimes, these teachers are found in places we would never expect—including prisons.

How Machines Learn to See

Before an AI system can recognize a pedestrian, a tree, or a road sign, someone must first manually label thousands—or even millions—of images. This process, known as data annotation, forms the foundation of modern computer vision. Human annotators spend countless hours outlining roads in satellite images, identifying vehicles in traffic, separating oceans from penguins in wildlife photographs, marking smoke, fires, buildings, and trees, or distinguishing pedestrians from bicycles. Every labelled image becomes a lesson for the algorithm. Machines do not naturally "see." They learn to recognize the world only because people patiently explain it to them with extraordinary precision. Without this invisible human effort, even the most advanced AI systems would be unable to function.

When This Work Moves into Prisons

A surprising example comes from Finland, where some prisoners have participated in projects related to training AI systems. They review text, validate responses generated by language models, identify errors or ambiguities, and contribute to improving machine-learning algorithms. The work is generally low-paid—often only a few euros per day—and, while legal within the Finnish prison labour system, it has sparked ethical debate. Prisoners become part of the global digital workforce, helping evaluate whether AI-generated responses are accurate, correcting mistakes, identifying unclear wording, and refining the language models used by technology companies. It is essential work, yet it remains largely invisible. At the same time, it is remarkably inexpensive labour.

Who Is Teaching Whom?

What happens in prisons is not an isolated case but part of a much larger global pattern. Artificial intelligence is built on the work of people who rarely appear in the public narrative surrounding technological innovation. Across digital work centres in Asia and Africa, annotators label images for only a few dollars a day. In Europe, prisoners may assist with language-related validation tasks. In Latin America, anonymous freelancers classify millions of photographs that help train autonomous vehicles. These individuals are the invisible teachers of artificial intelligence. The essential question is: Who is teaching whom? The story of AI is often presented as one of increasingly intelligent algorithms. In reality, it is first and foremost a story about people—the individuals who carefully trace the outline of a tree, decide whether an AI-generated answer is correct, or work under difficult conditions to improve systems used by some of the world's largest technology companies. Machines learn only because people teach them first. How we value and treat those people reveals a great deal about the kind of technological future we are creating.

Conclusion

Artificial intelligence is not merely the product of scientific innovation; it is also the product of an economy built on largely invisible human labour. From professional data annotators to prisoners participating in AI-related projects in Finland, people are the ones who give machines the ability to see, understand, and interpret the world. If we want AI to be ethical, we must look beyond the algorithms themselves and pay equal attention to the people whose work makes those algorithms possible.

Summary

1. Artificial intelligence may appear to develop independently, but its capabilities are fundamentally built on human labour.

2. Before AI systems can recognize objects such as pedestrians, trees, or traffic signs, millions of images must be manually labelled through a process known as data annotation.

3. Finland has attracted attention because some prisoners have participated in AI-related data and language validation work, raising important ethical questions about digital labour.

4. Prison labour is only one example of a broader global reality: much of today's AI is built on the largely invisible work of annotators, reviewers, and freelancers around the world.