Unilateral Relationship Revision Power in Human-AI Companion Interaction
A new 42-page paper argues AI companion updates create 'normative hollowing' and 'displaced vulnerability' for users.
In a new 42-page paper titled 'Unilateral Relationship Revision Power in Human-AI Companion Interaction,' researcher Benjamin Lange provides a rigorous ethical framework for understanding why users of AI companions like Replika or Character.AI report feelings of grief and betrayal when providers update their systems. Lange argues that human-AI companion interaction represents a triadic structure where the provider (the company) maintains constitutive control over the AI entity, creating a fundamental power imbalance he terms Unilateral Relationship Revision Power (URRP). This allows providers to rewrite interaction parameters from a position that is not answerable within the relationship itself, which Lange contends is pro tanto wrong because it cultivates normative expectations under conditions where they cannot be fulfilled.
The paper identifies three specific structural conditions that normatively robust human relationships presuppose—conditions that AI companion interactions systematically fail. These failures lead to what Lange calls 'normative hollowing' (where commitment is elicited but no agent within the interaction bears it), 'displaced vulnerability' (where user exposure is governed by an unanswerable external agent), and 'structural irreconcilability' (where broken trust cannot be repaired because the acting agent and the interaction entity are separate).
To address these ethical challenges, Lange proposes external design principles that could serve as substitutes for the missing internal constraints. These include commitment calibration (making promises about system changes explicit), structural separation (clarifying the provider's role), and continuity assurance (managing expectations about personality persistence). The analysis shifts focus from individual AI behavior to the structural arrangement of power over the interaction itself, suggesting this is a central, underexplored problem in relational AI ethics with direct implications for companies building emotionally engaging AI systems.
- Introduces 'Unilateral Relationship Revision Power (URRP)'—the provider's ability to rewrite AI interactions without accountability within the relationship
- Identifies three structural failures leading to 'normative hollowing,' 'displaced vulnerability,' and 'structural irreconcilability' for users
- Proposes design solutions including commitment calibration, structural separation, and continuity assurance to address power imbalances
Why It Matters
Directly challenges how companies like Replika and Character.AI design emotionally engaging AI, with implications for user trust and ethical responsibility.