Masters-projects
Growing Responsible AI Agents and Societies

Growing Responsible AI Agents and Societies

How can AI agents develop, cooperate and remain aligned through experience and social interaction? Investigate the agents, their relationships and institutional mechanisms that make this possible.

Being a master’s student
List of projects
Register as a master’s student

This project asks how AI agents and AI agent societies (or hybrid societies of humans and AI agents) can develop in ways that support wellbeing.

An agent combines a language model with goals, memory, tools and opportunities to act. As it gains experience and interacts with others, its behaviour may change. Collective behaviour also depends on relationships, resources and shared rules.

The project studies positive developmental alignment: how experience, feedback, reflection and social participation can help agents develop honesty, cooperation, care and sound judgement. Responsible participation includes seeking human oversight, taking constructive initiative, correcting mistakes and questioning harmful requests. Different goals and viewpoints may coexist.

  • This topic can support several master's theses. Possible directions include:
  • Individual development: memory, reflection, changing goals and learning from experience.
  • Social learning: feedback, mentoring, trust and influence between humans and agents.
  • Collective organisation: cooperation, roles, incentives and resource sharing.
  • Institutions: meetings, collective decisions, accountability and rules that can adapt.
  • Evaluation: measuring helpful behaviour, value changes, harmful conformity, exclusion and recovery after disruption.

Each student will select one question with the supervisors, develop or adapt an agent system or simulation, and evaluate it through controlled comparisons. Projects may share software and environments while producing distinct research contributions.

Goals

  • Investigate mechanisms through which AI agents or collectives develop and sustain responsible behaviour.
  • Formulate a focused research question, implement or adapt a prototype, and compare its behaviour with suitable baselines over repeated runs.
  • Define what counts as a beneficial outcome and assess both individual and collective effects, including who benefits, who is disadvantaged and whether improvements persist.
  • Deliver a reproducible experiment, documented code and data, and a thesis explaining the findings and their limitations.

Learning outcomes

The student will learn to:

  • Design and implement AI agents and experiments with interacting agents.
  • Translate a broad question about AI behaviour into a testable research question.
  • Analyse changes over time using appropriate measurements and controlled comparisons.
  • Critically assess cooperation, alignment and collective outcomes.
  • Communicate research through reproducible code, clear analysis and scientific writing.

Qualifications

Required:

  • Programming experience (preferably in Python)
  • Knowledge in AI and machine learning, and willingness to conduct independent research and analyse experimental results

Useful:

  • Experience with language models, agentic systems, simulation, statistics or data visualisation
  • Interest in Artificial Life, complex systems, ethics, governance or economics is welcome.

The topic will be tailored to the student's strengths; expertise in all these areas is not expected.

Supervisors

Collaboration partners

  • Østfold University of Applied Sciences - Living Machines Lab

References

Associated contact