It's essential that AI systems—including agents—work to support (not undermine) human autonomy.
But aligning to autonomy is hard. Our new paper shows that there are intra– as well as inter–value trade offs.
With @leonieclaude we explore a range of options...
There's still much to explore... The full paper is available can be found here 👇🙌
https://link.springer.com/article/10.1007/s11023-026-09786-9
Rather than advocating for one approach we leave open a suite of questions:
▶️Which higher-order value should guide agent development? ▶️How could meta-autonomy be operationalised if this was the best option? ▶️ What's the relationship between individual and collective freedom?
3. The "meta-autonomy" approach prioritises choice.
It gives the user the freedom to choose the *kind* autonomy they want in a specific domain of interaction with AI.
2. The capability-boosting approach is built around the idea of flourishing.
It seeks to empower the user by increasing the capabilities needed to be autonomous over time.
This approach emphasizes the importance of knowledge and competence in safeguarding human freedom.
2. The capability-boosting approach is built around the idea of flourishing.
It seeks to empower the user by increasing the capabilities needed to be autonomous over time.
This approach emphasizes the importance of knowledge and competence in safeguarding human freedom.
1. The liberal approach is committed to autonomy-as-user-integrity.
It promotes autonomy by acting strictly on what a person says they want, without trying to change their mind.
This also means allowing the user to do things that might not be good for their own well-being.
Rather than advocating for one approach we leave open a suite of questions:
▶️Which higher-order value should guide agent development? ▶️How could meta-autonomy be operationalised if this was the best option? ▶️ What's the relationship between individual and collective freedom?
3. The "meta-autonomy" approach prioritises choice.
It gives the user the freedom to choose the *kind* autonomy they want in a specific domain of interaction with AI.
1. The liberal approach is committed to autonomy-as-user-integrity.
It promotes autonomy by acting strictly on what a person says they want, without trying to change their mind.
This also means allowing the user to do things that might not be good for their own well-being.