Beyond Value Elicitation: Towards Moral Profiles in Early Requirements Engineering via Role-Playing Games and Anthropologist LLMs
This study proposes a novel approach in Requirements Engineering that combines immersive role-playing games with anthropologically grounded large language models to elicit and reconstruct users' tacit moral profiles as dynamic narrative representations, thereby overcoming the limitations of traditional value elicitation methods.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Imagine you are trying to understand what a person truly cares about. You could ask them directly, "What is your most important value?" But often, people don't know the answer, or they can't explain it. Their values are like tacit knowledge—hidden habits and gut feelings that only show up when they are actually doing something, not just talking about it.
This paper proposes a new way to uncover these hidden values for software designers. Instead of filling out a boring survey, the researchers used a Role-Playing Game (RPG) and a special AI "Anthropologist" to build a "Moral Profile" for each user.
Here is how it works, broken down into simple steps:
1. The Game: A Digital "Sim City" for Ethics
The researchers created a tabletop role-playing game (like Dungeons & Dragons) set in the real world of digital privacy.
- The Setup: Ten real people played the game. They weren't acting as fantasy wizards; they were acting as themselves.
- The Plot: The players received fake emails, found a suspicious computer, met a confused elderly neighbor needing help with a smartphone, and had to decide whether to trust a shady website or call the police.
- The Twist: There were no "winning" or "losing" conditions. The goal wasn't to score points; it was to make choices.
- The Magic: Because it was a game, players felt safe to experiment. They argued with each other, asked an AI "oracle" for advice, and wrote in a diary about why they made certain choices. This created a rich tapestry of stories rather than just a list of answers.
The Analogy: Think of this like a flight simulator. You don't learn how a pilot handles a storm by asking them, "How do you handle storms?" You watch them fly the plane when the virtual weather turns bad. The game forced the players to "fly" through ethical dilemmas, revealing how they actually behave when under pressure.
2. The AI: The "Anthropologist" Detective
Once the game was over, the researchers had a mountain of messy data: audio recordings, diary entries, chat logs with the AI, and notes taken by the human researcher.
- The Problem: A human trying to read all this would take weeks.
- The Solution: They used a Large Language Model (LLM), but they didn't just ask it to summarize the text. They gave the AI a specific "job description": "Act as an Anthropologist."
- The Training: They fed the AI a "textbook" on how anthropologists study humans. This textbook taught the AI to look for:
- Stories: How did the player explain their choice?
- Context: Did they act differently because of their friends?
- Contradictions: Did they say they were careful but then take a risk? (The AI was told not to ignore these contradictions, but to treat them as part of the story).
- The Output: Instead of a checklist of "Values: Privacy, Honesty, Safety," the AI wrote a narrative profile. It described the player's "moral world" as a dynamic story. For example, it might say: "This player knows the rules but often gets swept up in the urgency of the moment, relying on their friends to make the hard calls."
The Analogy: Imagine a detective who doesn't just look at a suspect's ID card (which lists their name and age). Instead, the detective watches the suspect interact with friends, watch how they react to a crisis, and listen to how they tell their own story. The detective then writes a biography that explains who the person is, rather than just listing their stats.
3. The Test: Did the "Profile" Work?
To see if this method actually worked, the researchers played a guessing game.
- They showed the players a new set of privacy dilemmas (questions they hadn't seen before).
- They asked the AI to predict how each player would answer, based only on the "Moral Profile" it had written after the game.
- The Result: The AI guessed correctly 81% of the time.
- A standard AI (without the "Anthropologist" training) only got about 51% right.
- This proves that the "Moral Profile" (the story) captured something real about the players that a simple list of values missed.
Why This Matters for Software
Currently, when software companies want to know what users want, they often use rigid categories (like "I value security"). But people are messy and complex.
- Old Way: Trying to fit a person into a box labeled "Privacy-First."
- New Way: Understanding that a person might value privacy unless they are in a hurry, or unless their friend asks them to share.
This paper suggests that by using games to generate stories and AI to analyze those stories, we can build software that respects the real, messy, dynamic nature of human values, rather than just a simplified version of them.
Summary
The paper claims that:
- Games are better than surveys for finding out what people actually do, not just what they say they do.
- AI, when taught to think like an anthropologist, can turn messy game data into a coherent "Moral Profile" (a story about a person's values).
- These Profiles are accurate enough to predict how people will act in new situations, helping software designers build systems that fit human nature better.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.