Digital avatars and human-AI interaction

A digital avatar is not simply a visual interface for an AI model.

Once an artificial system communicates through a face, voice and recognisable behaviour, people can begin to respond to it socially. Personality, presence and perceived familiarity can influence the interaction even when users know that the system is artificial.

This makes digital avatars particularly interesting from a human-AI interaction perspective. Their design affects not only usability, but also expectations, emotional responses and the way users interpret the system behind the avatar.

Data privacy and consent

Avatar systems can process unusually personal forms of data, including voice, appearance, behavioural signals and individual stories or memories.

For me, responsible development begins with asking which of these signals are actually necessary. Data collection, storage and retention should be limited to a clear purpose, while users need to understand what is being processed and how it may be used.

Consent becomes especially important when an AI system represents or reconstructs an identifiable person.

Representing private individuals and public figures

The interview also addressed the difference between creating digital representations of private individuals and public figures.

The fact that images, recordings or information are publicly accessible does not automatically resolve questions of consent, identity or appropriate use. Different contexts create different expectations around representation, ownership and control.

For designers and developers of digital avatar systems, these questions need to be considered as part of the system itself rather than treated as a legal or ethical layer added afterwards.

Responsible deployment

Technical possibility alone is not a sufficient reason to deploy an AI system.

When digital avatars enter museums, educational environments or other public contexts, questions around governance become practical: What data is processed? Where is it stored? Who can change the behaviour of the system? How are outputs monitored? What happens when the system produces something inappropriate or incorrect?

Responsible AI therefore also involves operational design, including boundaries, evaluation, human oversight and clear mechanisms for intervention.

Designing for trust without simulating certainty

Human-like interfaces can make AI systems appear more knowledgeable, intentional or reliable than they actually are.

For me, this creates an important design responsibility. Trust should not come simply from making an avatar more realistic or socially convincing. Users also need ways to understand what the system knows, where its responses come from and where its limitations lie.