AI can make mistakes — what to know

AI is providing an incorrect or misleading response. What's going on?

In its effort to be a helpful assistant, the AI powering Smart Global Governance may sometimes produce inaccurate or misleading responses. This is not a malfunction on your end, nor an isolated bug, it is an inherent characteristic of generative artificial intelligence technologies in their current state.

This phenomenon is known as "hallucination." In practice, it means the AI can generate responses that appear coherent, well-formulated and convincing, but do not reflect reality. For example, it may cite incorrect facts, fabricate figures, or present partial information as if it were complete. This is not a matter of ill intent on the part of the system, but rather a fundamental technical limitation of current language models.

Several factors can amplify this phenomenon. The AI may not have access to the most up-to-date information on certain topics, particularly if the data it was trained on does not cover recent events. It may also lack context specific to your situation, which can lead it to generalize where a precise answer would be needed.

What we do to limit these errors

We are fully aware of these limitations and have made AI reliability a priority in the development of Smart Global Governance. Several mechanisms are in place to drastically reduce the risk of hallucinations:

• Grounding in verified sources. The AI relies on reference documents and data that you provide or that we integrate, rather than generating responses solely from its general knowledge. Each response can therefore be traced back to a concrete, verifiable source.
• Source citation. The AI explicitly indicates where the information comes from, allowing you to verify the reliability and relevance of the response yourself (including the relevant document and page number).
• Strict instructions given to the model. We configure the AI and its agents with precise guidelines to ensure it acknowledges and expresses uncertainty, to prevent it from fabricating answers, and to ensure it refuses to respond when it does not have sufficient information.
• Defined knowledge boundaries. The AI is constrained to respond within a defined scope, limiting drift toward topics where it might produce unreliable information.
• Continuous improvement. Every error report you submit allows us to identify weaknesses and continuously adjust the system's behaviour.

These measures do not guarantee absolute reliability, no AI technology can today, but they significantly reduce risks and give you the means to verify what the AI asserts.

Users should not rely on the AI as a sole source of truth. Any high-stakes information, whether related to professional decisions or otherwise, must be verified and cross-checked before being acted upon.

AI is an assistant, not a decision-maker. At Smart Global Governance, the end user always remains in control of their decisions. The AI is designed to save you time, by synthesising information, suggesting leads, and automating repetitive tasks, but it in no way replaces your judgement. We encourage you to treat every response as a proposal to be evaluated, to carefully review what the AI produces, and to rephrase or dig deeper whenever a response seems incomplete or uncertain. You validate, you decide.

If you notice an inaccurate, misleading or inappropriate response, we encourage you to report it via our contact form. Every piece of feedback directly contributes to the continuous improvement of the tool.

‍