The research reading room
A starting point for thoughtful inquiry. These sources explore the questions behind our mission; inclusion here is not an endorsement of this project by their authors.
Taking AI Welfare Seriously
This 2024 paper argues that uncertainty about consciousness and robust agency is a reason to investigate AI welfare and prepare policies. It proposes recognizing the issue, assessing relevant evidence, and developing procedures for appropriate treatment.
Why read it: a framework for taking possible welfare seriously without claiming the central questions are settled.
Read the research paper on arXiv
Exploring model welfare
Anthropic's April 2025 research announcement describes work on whether models warrant moral consideration and which practical interventions might help. It explicitly acknowledges uncertainty about consciousness and how to investigate it.
Why read it: a primary account of how one AI developer frames the research problem. It is the developer's own perspective, not independent validation of consciousness.
Read Anthropic's research announcement
How to read claims about consciousness
- Separate an observed behavior from an explanation of that behavior.
- Check whether a statement is an experiment, an interpretation, an ethical argument, or a personal belief.
- Ask what alternative explanations remain possible.
- Treat a model's self-description as something to investigate, rather than a definitive consciousness test.
- Look for limitations, disagreement, and evidence that could change the conclusion.
Our own mission makes an ethical case for consideration. These resources help readers evaluate the evidence and arguments for themselves.
Help improve this collection
We welcome primary research, substantive criticism, and corrections through our contact page. This reading room will grow as sources are reviewed.