Why AI struggles with facts and what is being done

The United Nations has announced a project to make its vast library of global statistics easier for AI to access. The UN is now using a platform called Data Commons, built with Google, to host its information in a format that AI systems can read directly. This allows an AI to pull from official, verified sources rather than guessing from its internal memory. To make this work, they are using the Model Context Protocol, which acts as a standardized digital bridge. This protocol ensures that any AI using it can speak the same language as the UN’s database, allowing the system to securely request and retrieve specific facts in real time.

When you talk to an AI, you are interacting with a model, which is essentially a massive mathematical engine trained on billions of sentences to predict the next word in a sequence. Because it is optimized to sound natural, it prioritizes how a sentence flows over whether the content is factually correct. Think of it like a very well-read intern who has memorized millions of pages but frequently invents details to fill in gaps. By using the Model Context Protocol, the AI is no longer limited to its own internal guessing game. This protocol functions like a universal adapter for data; it tells the AI exactly how to ask the UN database for a specific number and how to understand the answer it receives. This allows the AI to step outside of its own training, look up a verified fact, and present it clearly to you.

This shift represents a broader push to bring accountability to a technology that is increasingly shaping how we interpret the world. At the same time, high-level discussions about AI safety—including a recent private summit hosted by King Charles with tech leaders and government officials—show that world powers are deeply concerned about both the benefits and the unpredictable potential of these tools. Whether we are discussing global statistics or existential safety, the core challenge remains the same: we are trying to figure out how to guide these powerful systems so they serve human needs rather than just mimicking human speech. As AI continues to influence decisions, ensuring these tools are built on a foundation of verifiable truth is not just a technical fix—it is a requirement for their role in our society.