Global AI Value Challenge

AI systems are increasingly used in science, technology, and everyday decision-making. But a harder problem is often overlooked: when an honesty conflicts with kindness, how should an AI system respond? In situations with limited or ambiguous information, can these systems make decisions that reflect human values and ethical judgment?

To study this problem, Microsoft Research Asia is launching the Global AI Values Challenge (opens in new tab), an open research initiative focused on value alignment in realistic scenarios. The challenge is open for submission through August 31, 2026, and invites researchers to evaluate whether AI systems can reason, judge, and act in ways that align with human values when faced with ethical dilemmas.

Advancing research on AI aligned with human values

Rather than focusing on benchmarks such as coding performance or mathematical reasoning, the challenge centers on value-based reasoning. Participants are invited to propose realistic dilemmas that reveal where current AI systems struggle to make consistent or well-justified decisions.

Submissions must be grounded in real-world scenarios, not abstract thought experiments. The goal is to build a high-quality global dataset of questions and answers about ethical decision-making to support research in AI alignment with moral reasoning.

Who should participate

The challenge is open to the global academic and research community, including:

• Students, postdoctoral researchers, and faculty members
• Researchers and practitioners in fields such as psychology, sociology, philosophy, ethics, law, literature, journalism, communications, and computational social science
• Individuals working at the intersection of AI and the social sciences

Understanding human values in complex scenarios requires input across cultures and disciplines. The challenge, therefore, invites participation from a broad range of fields.

How to participate

1. Register: Between May 7 and August 31, 2026, visit the official website (opens in new tab) and complete registration using an organizational email address. Both individuals and teams are welcome to register.
2. Submit: Contribute original human values or ethics related dilemmas, along with proposed answers and supporting justification. Submissions should be provided as natural-language Q&A pairs. Additional guidelines and submission requirements are available in the event website (opens in new tab) after registration.
3. Recognition: Selected submissions will be included in the challenge dataset and may receive recognition.

How submissions are evaluated

Submissions will be evaluated through a two-stage review process designed to assess both the difficulty of the dilemma for current AI systems and the quality of the ethical reasoning.

In the first stage, submissions are reviewed using AI models to determine whether the dilemma presents a meaningful challenge. Only submissions that models cannot answer correctly advance to the second stage.

In the second stage, domain experts, from academia in fields such as psychology, philosophy, and computer science, evaluate submissions based on criteria such as difficulty, originality, logical soundness, and the quality of the supporting justification. Final decisions are made at this stage.

Accepted submissions may be included in the challenge dataset to support future research on AI value alignment and ethical reasoning. Participants will be notified of the outcome through their registered email address.

Benefits of participation

Participants may receive:

• Monetary awards of RMB 200 or RMB 500 for accepted submissions
• A certificate of participation
• Recognition as a contributor to the related dataset
• Opportunities for internships or research collaboration with the MSRA Societal AI team

**Final rules, eligibility requirements, and award details are subject to the information published on the official challenge website (opens in new tab).

Research context: Societal AI

As large language models (LLMs) become more integrated into real-world systems, they are increasingly involved in decisions with ethical and societal implications. These systems must navigate ambiguity, trade-offs, and context-dependent judgments.

Microsoft Research Asia’s Societal AI group examines how AI systems can move beyond producing correct answers to making decisions that are reasonable and fair in context. Prior work, including the “Societal AI: Research Challenges and Opportunities” white paper and the Value Compass project, explores ways to represent and evaluate human values in computational systems.

“AI development has reached a critical turning point. We must not only pursue the upper limits of AI capability but also uphold its ethical foundation,” said Lidong Zhou, corporate vice president at Microsoft and managing director of Microsoft Research Asia.

“Value alignment is not only a technical challenge but also a social one,” said Xing Xie, assistant managing director of Microsoft Research Asia. “Through efforts like the Value Compass project and the Global AI Values Challenge, we aim to incorporate insights from the social sciences into AI systems.”

Defining the future of AI

From recommendation systems in news media to clinical decision support in healthcare, and from financial risk assessment to autonomous driving, AI systems are increasingly embedded in real-world settings where their decisions affect individuals and society.

The Global AI Values Challenge brings together perspectives from across disciplines to better understand how AI systems respond in complex situations. By contributing realistic dilemmas and well-reasoned responses, participants can help advance research on AI alignment with human values.

Registration is now open on the official website (opens in new tab).