Source-linked AI summary
Autonomy, Social Norms, and Alignment: Towards a Developmental Framework for Autonomous Artificial Agents
Marica Notte, Ludovica Marinucci, Vieri Giuliano Santucci
TL;DR
The paper addresses how autonomous embodied agents can remain aligned when pre-existing data, human feedback, and fixed rules are insufficient for dynamic environments. It proposes developmental norm learning through embodied interaction, intrinsic motivation, social cooperation, and staged autonomy, while concluding that alignment must be maintained dynamically and that responsibility cannot rest solely with designers.
Problem
Existing datasets, human feedback, and predefined rules are insufficient for aligning highly autonomous embodied agents in unstructured and unknown contexts.
Method
The paper proposes that agents learn and internalize norms through embodied interaction, intrinsic motivation, social participation, and developmentally staged learning.
Results
The paper concludes that robust alignment should be maintained dynamically through agents’ ongoing participation in normative social life rather than training alone.
Takeaways & Limitations
Appropriate autonomy is presented as a precondition for norm internalization and flexible coexistence between human and artificial agents.
Takeaways & Limitations
Artificial agents’ increasing norm-management capacities do not make them morally responsible because traditional moral responsibility requires free will and intentionality.
Abstract
from arXiv · showhide
In recent years, artificial intelligence has made extraordinary progress thanks to large-scale models capable of generalization and the generation of complex outputs. However, transferring this potential into embodied agents reveals a significant limitation: the most advanced systems rely on pre-existing datasets and human feedback strategies that are powerful but insufficient in dynamic or unknown contexts. To adapt, an agent must acquire knowledge through direct interaction with its environment. One strategy to address this challenge involves introducing higher-level mechanisms, such as intrinsic motivations, which leverage curiosity and competence, to guide exploration and learning in complex environments. While this flexibility expands autonomy, it complicates the task of ensuring agents remain aligned with human goals. Alignment, already a challenge for artificial systems in general, becomes even more complex in unstructured and dynamic contexts where predefined rules prove insufficient. To be effective and adaptable, norms must be rooted in experience through an epistemological process that starting from simple, situated principles allows for the gradual construction of more complex rules through experience, autonomous learning, and cooperation with other moral agents. Similarly to children learning social norms by exploring their environment and participating in collective practices, artificial agents must also be educated toward alignment. Following Dennett, the status of a moral agent is not innate but is attributed gradually based on the ability to responsibly manage increasing degrees of freedom. From this perspective, the regulatory sandboxes can be viewed as pedagogical environments for AI: dynamic spaces where alignment develops as a formative process, progressively shaping autonomous behaviors through interaction and cooperation in scenarios of increasing complexity.
1. Autonomy in artificial embodied agents
Embodied agents need direct interaction and intrinsically motivated learning to adapt beyond the limits of pretraining and human feedback. This flexibility makes alignment harder in unknown contexts, motivating norms learned through interaction with environments and humans.
- Pretraining and continuous human feedback are powerful but insufficient for embodied agents operating in dynamic or unknown contexts.
- Intrinsic motivations such as curiosity and competence guide exploration, goal discovery, skill learning, and adaptation to non-stationary environments.
- Alignment becomes more difficult for highly autonomous agents because unknown events cannot all be anticipated and fixed rules may constrain autonomy.
- The proposal treats alignment as an epistemic process in which agents learn increasingly complex norms through environmental interaction and human guidance.
2. Autonomy and the acquisition of social norms in children
Children develop autonomy together with social norms through exploration, imitation, and participation in social groups. This developmental pattern motivates asking how artificial agents might internalize norms for flexible alignment in novel situations.
- Children’s autonomy develops progressively through opportunities to explore, make decisions, and manage problems, supporting physical and psychological development.
- Caregiver models and imitation help children learn context-sensitive behavioural norms and integrate into social groups.
- Social norms are informal rules governing what is appropriate, required, prohibited, or permitted, and children gradually learn to apply them autonomously.
- The developmental comparison raises whether artificial agents can internalize norms rather than merely execute fixed rules, enabling flexible and robust alignment in novel situations.
3. A developmental model of norm alignment in autonomous agents
The proposed developmental model frames norm alignment as embodied, intrinsically motivated, socially interactive, and staged. It calls for gradual autonomy expansion and dynamically maintained alignment through participation in normative social life, while recognizing unresolved questions about moral responsibility.
- Norm alignment requires embodied and situated learning, intrinsic motivation, social interaction, and development through successive stages.
- The model does not make artificial agents morally responsible, because moral responsibility traditionally requires free will and intentionality.
- Because autonomous agents must interpret norms and handle conflicts, normative authority should be delegated gradually as they demonstrate reliable, flexible, context-sensitive competence.
- The goal is dynamically maintained alignment through ongoing participation in normative social life rather than perfect alignment achieved through training alone.