CONSIDER: A New Attempt to Resolve Extreme Moral Disagreements Using Generative AI

Core Viewpoints

A research team has launched the CONSIDER prototype system, which helps people address extreme moral disagreements through structured adversarial dialogue. Drawing on Mill's epistemological value theory of dissent, it achieves value clarification and rational dialogue.

Basic Information

• Original Author/Team: Paper author team (submitted to arXiv)
• Source Platform: arXiv
• Original Title: Can Generative AI help people navigate Radical Moral Disagreements? The CONSIDER prototype
• Original Link: http://arxiv.org/abs/2605.31574v1
• Release Time: May 29, 2026

Social Dilemmas

Extreme moral disagreements (such as abortion rights, climate policies, etc.) are prevalent in today's society, triggering fierce debates and affecting mental health, leading to rising social costs. Traditional dialogue mechanisms struggle to resolve them. People tend to avoid or oppose each other, hindering democratic deliberation and social cohesion.

Shortcomings of Existing Tools

Large Language Models (LLMs) have been proposed to support democratic deliberation, but existing AI tools are insufficiently calibrated, often adopting a neutral and conciliatory stance, ignoring deep-seated value differences, and failing to effectively address extreme moral disagreements.

CONSIDER's core design philosophy is derived from John Stuart Mill's epistemological value theory of dissent—dialogue with opposing views helps discover truth, promote intellectual development, and moral maturity.

The system adopts a structured adversarial dialogue model: after the user expresses an opinion, it generates opposing AI opinions to guide interaction. The goal is not to persuade users to change their positions, but to help them clarify their values, test the rationality of their arguments, and cultivate intellectual humility.

Key Components

1. Opinion Generation Module: Based on user input, uses LLMs to generate logically consistent opposing views, considering topic complexity, cultural sensitivity, and argument rationality.
2. Dialogue Management Module: Controls the pace of the process, sets rules (based on reason-based argumentation), avoids personal attacks or emotional confrontation, and maintains constructive dialogue.
3. Reflection Guidance Module: Prompts users to reflect at key nodes (e.g., understanding of opposing views, blind spots in their own arguments, whether their opinions have changed, etc.).

Main Risks

1. Echo Chamber Reversal Risk: Extreme or unreasonable opposing views may reinforce users' original biases.
2. Emotional Harm Risk: Extreme moral disagreements are related to identity, and mandatory adversarial dialogue may trigger defensive psychology or emotional trauma.
3. Manipulation Risk: Improper design may be used to indoctrinate specific views; it is necessary to maintain neutrality in opinion generation and avoid implicit value biases.

Optimization Directions

1. Improve the opinion generation algorithm to make opposing views more challenging and reasonable;
2. Develop fine-grained user models to implement personalized dialogue strategies;
3. Establish a long-term effect evaluation mechanism to verify the improvement of moral reasoning ability.

Application Scenarios

Plans to explore deployment in scenarios such as educational environments, corporate ethics training, and online community governance, and obtain practical feedback for iterative optimization.