Would you trust the answer from an AI more than from a lawyer?

- We find that members challenge our lawyers' advice with findings they have made on ChatGPT, for example, says Farah Ali, head of the legal section at NITO.

NITO's lawyers experience that some members use AI to get quick answers, in addition to consulting with a lawyer. This applies to everything from simple to complicated legal issues.

- This means that we spend more time than necessary, says Ali.

AI is trained on "everything", including Twitter messages

NITO's lawyers base their advice on Norwegian legislation and relevant sources of law, while an AI that ChatGPT simply explained, uses statistics to predict the next word in a sentence," says Raymond Kristiansen, project manager at NITO's Competence Centre.

The largest AI models are trained on "everything", Kristiansen explains. Also on social media, with all that is there.

- The question then is what the statistical model says about what becomes true when it is trained on, for example, millions of Twitter/X messages? Therefore, you must always ask for sources when using ChatGPT. And comply with them.

Extra important with source criticism

- When we ask ChatGPT questions, there is always an approximation in the picture. It is important to take this into account when assessing such answers. In other words, it is not an encyclopedia or true science," says Kristiansen.

This means that even though a generative AI solution may seem both convincing and sensible in its answers, it is not always straightforward to use such models to interpret legal issues.

- You must at least use source criticism when you use the answers you get. Ask the AI solution to provide sources for the answers it provides, and check the sources. Otherwise, you risk ending up like the advisor in Tromsø Municipality who did not check the sources in an AI-generated report on school closures. They turned out to be fake, Kristiansen points out.