About the Project
Nævnenes Hus is exploring how generative AI can support legal casework without weakening the professional judgment behind administrative decisions.
The central challenge is to use AI as support while preserving the legal expertise needed to recognize when its output is wrong.
The Context
Nævnenes Hus handles cases across 16 different boards and represents the final stage of administrative decision-making before cases potentially reach the courts.
The organization already has access to an OpenWebUI-based AI platform, and some employees are experimenting with AI-supported legal work. The main challenge is therefore not access to AI, but how it should be integrated without weakening professional competence.
This is particularly important because Nævnenes Hus employs many students and early-career legal professionals. The organization explicitly wants AI adoption and competence development to go hand in hand: employees should become better legal professionals through using AI, rather than gradually becoming dependent on it.
The HI Element
A concrete situation at Nævnenes Hus captures this challenge.
A caseworker testing an AI system noticed that it referred to previous decisions based on an outdated version of the law. The legislation had changed only a few months earlier, but the AI had not captured the change.
The immediate technical response could be to improve the system so that it automatically detects this type of error. However, this led to a different question:
Should the AI catch the mistake, or is recognizing the change in legislation something the caseworker should still be able to do?
This is where the Hybrid Intelligence element becomes visible. The AI can support the caseworker by retrieving, processing, and suggesting potentially relevant information. But the legal professional must use their knowledge to judge whether that information is current, relevant, and legally applicable.
This reflects a deliberate division between AI prediction/support and human judgment:
| Legal professional – Judgment | AI – Prediction and support |
| Defines the legal question and relevant context | Retrieves and processes potentially relevant information |
| Applies knowledge of current legislation | Generates suggestions, summaries, or possible interpretations |
| Detects outdated or inappropriate material | Identifies patterns and possible relevant material |
| Judges whether the output applies to the specific case | Supports comparison and exploration |
| Interprets the information and owns the final decision | Does not determine the legal outcome |
he important distinction is that human judgment does not appear only as a final approval step. The legal professional remains active throughout the process by questioning the output, recognizing inconsistencies, interpreting their significance, and deciding what should be used.
In this example, the caseworker’s ability to recognize outdated legislation was exactly what prevented the AI output from being treated as authoritative. Improving the system could make the task easier, but it could also remove moments where legal expertise is exercised.
This matters especially for younger legal professionals. René Lind emphasized that AI projects and competence development should go hand in hand, so employees become better through using AI rather than losing the expertise required to assess it.
The approach also fits with the logic of legal method already used in the profession: information must be critically examined, interpreted, and assessed rather than simply accepted.
The Envisioned Outcomes
The approach is still developing, so these should be understood as intended outcomes rather than evaluated effects:
- More critical use of AI, where employees actively identify outdated, incomplete, or inappropriate outputs.
- Preservation and development of legal expertise, particularly among students and early-career professionals.
- Clearer division between AI support and human judgment, rather than treating AI output as an answer to be approved.
- AI-supported workflows that maintain professional responsibility, with legal professionals remaining responsible for interpretation and final decisions.
A possible next step discussed by Nævnenes Hus and CHI is to translate these principles into more structured workflows and training, potentially using FERC. The purpose would not simply be to improve prompting, but to make the points where professional judgment is required more explicit throughout AI-supported legal work.