Virtual Alter-Egos: Analyzing Human-Avatar Proxy Dynamics and Ethical Implications in the Digital Age

Prof. Paula Sweeney has recently published a paper delving into the relationship between humans and their avatars, understood as a proxy relation. The central focus is to provide a conceptual framework for comprehending the human-avatar proxy relationship, which draws from traditional understandings of proxies while also considering the unique features of online versions. The paper further investigates the conditions under which a person can be deemed accountable for the actions executed by their avatar proxy. A significant distinction is made between avatars as proxies and 'uploads', the latter referring to transferring thoughts and memories from an organic brain to a computer, leading to philosophical debates around personal identity. The paper also touches upon the ethical concerns surrounding the use of avatars as proxies, given that the individuals they represent can be held responsible for the avatar's actions. A salient point is the epistemic gap in the proxy relationship, where the avatar might act without comprehensive knowledge of what the represented person would do, leading to potential ethical issues. This research is organized into three sections: the first unpacks the desirability and features of the proxy relation; the second identifies the kinds of avatars that can serve as human proxies; and the third section underscores the distinct responsibility risks, especially the epistemic gap, which emerges from using avatars as proxies.

1. Proxies

This paper examines the concept of proxies, focusing on the relationship between humans and their avatars in online environments. Traditionally, a proxy is someone authorized to act on another's behalf, with this relationship extending beyond human agents to online entities that serve businesses digitally.

The core of the proxy relation, as defined by Floridi (2015), is that a proxy both represents (stands for) and replaces (stands in for) another entity. Floridi distinguishes between non-degenerate and two degenerate forms of proxy relations: signs and surrogates. Sign-proxies only signify another entity without acting on its behalf (e.g., a word representing an object). Surrogate-proxies, on the other hand, act in place of another entity but don't represent it (e.g., oat milk replacing cow’s milk).

Only in non-degenerate proxy relationships are the actions of the proxy seen as actions of the represented agent. For instance, sign-proxies can't be interpreted as executing actions of the represented entity, while surrogate-proxies might act for an agent but lack a connecting link to them. Two scenarios in which this link might be absent are when it was never established or when it deteriorates over time. The paper suggests that the proxy relation's stability depends on community recognition rather than strict contractual obligations.

The section addresses the responsibilities and expectations within the proxy relationship. Millar (2014) posits that the optimal proxy decision reflects what the agent would decide if they could make the choice themselves. The proxy's primary role is to channel the agent's will and make it known, even if it contrasts with the proxy's personal beliefs. For instance, if someone has designated you as their healthcare proxy and expressed their wish for organ donation, your duty is to convey this desire, irrespective of your personal stance on the matter.

However, complications arise when proxies lack comprehensive information to make decisions reflecting the agent's desires. In such cases, the proxy's duty is to infer based on their knowledge about the agent and exclude personal preferences. Additionally, proxy relationships often have boundaries. For instance, if John represents Bill as a proxy during a council meeting vote, Bill would be responsible for the vote John casts but not for John's inappropriate behavior during the meeting. The boundaries of the proxy relationship may be either explicit or implicit, derived from historical and contextual cues.

In the subsequent section, the author plans to delve deeper into how these characteristics—ambiguous boundaries and the knowledge gap faced by proxies—might manifest in avatar-based proxy relationships. They will also explore the type of avatars pertinent for proxy relationships.

2. Avatars 

The rapid technological advancement brings with it the prospect of sophisticated avatars that can serve as full proxies, representing a person in various scenarios. At present, proxy relations, which are established and maintained by social conventions, may not allow avatars to fully take on these roles. Nonetheless, this possibility is imminent.

Several kinds of avatars have evolved over time, reflecting varied levels of representation:

Telepresence Systems: Current systems allow a user to interact from a remote location, almost as if they are physically present. They are equipped with cameras, microphones, and displays, often showing the user's face. These systems, especially in fields like education, tourism, and personal care, allow a robot to act as a 'sign proxy' for a person. While these systems stand for an individual, they don’t act on their behalf without direct control.

Memoji Avatar: Memojis are basic representations resembling the person they are portraying, with limited interactive capabilities.

Gaming Avatar: This form of avatar has evolved with the gaming industry. Players increasingly design characters that represent them in virtual worlds, creating an emotional link between the player and their digital counterpart. Although these avatars "stand in for" players within a game, they might not fully represent or refer to the player's identity.

Advanced Memoji Avatar: Advances in this field led to creations like Kanye West's hologram gift to Kim Kardashian – a hologram of her deceased father. Using detailed personal data, these avatars create a more sophisticated representation. Examples include 'DadBot' and 'Roman', chatbots created using extensive personal data of deceased individuals. Advanced Memojis give a static representation based on past interactions, capturing an individual's essence.

Autonomous Advanced Avatar: Envisioned as a future possibility, this avatar type could dynamically represent someone. For example, an avatar operating similar to 'Dadbot' but constantly updating its data set with new interactions and preferences of the person it represents.

However, whether an entity is deemed a full proxy is subject to two variables:

Socially Determined and Context Sensitive Nature of Proxies: The definition of what counts as standing in for another person can be fluid. An entity may sometimes fulfill the proxy requirements in certain contexts but not in others. This is evident in depictions like Charlie Brooker’s 'Be right back' from Black Mirror, where the protagonist's perception of an android version of her deceased partner fluctuates.

Technological Advancements: As technology evolves, our understanding of proxy representation might change, shifting the boundaries between degenerate and non-degenerate proxies.

The evolution of avatars into full proxies is accompanied by serious considerations. Given that full proxy relations entail responsibilities, the ambiguous nature of what qualifies as a full proxy poses challenges. It is essential to anticipate and address these challenges, particularly as technology blurs the lines between degenerate and non-degenerate proxies.

3. Avatar Proxies

The utilization of autonomous advanced memoji avatars as proxies offers a multitude of benefits. Such avatars allow users to interact in emerging digital environments, such as Zuckerberg’s Metaverse, expanding the scope of experiences available to them (Garaj, Dudley & Kristensson, 2022). Particularly for those with disabilities, virtual environments provide opportunities for social interaction, education, and work which might otherwise be inaccessible. Though initially, avatar proxies might seem to simply offer new gaming experiences, they can extend their utility into professional realms. For instance, in a future where humans can only be in one place at once, an avatar proxy might attend a virtual meeting on behalf of an executive, facilitating multi-tasking. This would especially be beneficial when an individual wants to avoid emotionally burdensome tasks, like delivering bad news or public speaking.

However, there are indirect harms and risks tied to avatar proxies. The potential misuse of these proxies, whether to manipulate votes or misrepresent the user, is evident. Furthermore, while the actions of the proxy should theoretically be limited to the designated contexts, the very act of establishing a proxy relationship might have implications for the user's reputation. Drawing from the literature on player responsibility in games, even actions that are not controlled by the user can affect their moral standing, as in the controversial game 'Custer’s Revenge' (Patridge, 2011). This suggests that the choice of a particular avatar or character as one's proxy might be an implicit reflection on their moral character.

The primary risk in using proxies is the potential for the represented individual to be held personally responsible for actions they did not directly control. This is especially problematic when the proxy, whether human or digital (like an avatar), encounters a situation where they lack explicit instructions (termed as the 'epistemic gap') on how to act on behalf of the represented individual. Such gaps can arise frequently due to the intricacy of real-world contexts.

For instance, if a person designates a proxy to vote on their behalf but fails to provide guidance for situations where their preferred candidate withdraws or faces controversy, the proxy faces an epistemic gap. In such cases, human proxies might base decisions on other known preferences of the represented individual or consult friends or relatives for insights, which introduces another layer of complexity: determining the weight of testimonies from various sources.

Avatar proxies, however, present unique challenges. When faced with an epistemic gap, they might employ intelligent decision support systems, like those utilized in sectors ranging from finance to healthcare. Such systems, despite their advancements, have raised concerns about fairness and reliability, as highlighted by Rubel et al. (2021). Importantly, while algorithmic mistakes in traditional sectors might be diffuse in terms of responsibility, in an avatar proxy setup, the blame can be directed straight to the represented individual.

Moreover, there are concerns about how avatar proxies could affect a person’s reputation, self-perception, or even be manipulated by employers to shift blame onto employees. Some may suggest providing explicit instructions to avatar proxies or having them refrain from making decisions in uncertain scenarios. However, the unpredictable nature of situations and the consequences of inaction make these solutions implausible.

Existing legal frameworks like the 'respondeat superior' principle suggest that the represented individual would likely be held responsible for their avatar proxy's actions. Given these risks, there's an imperative need to define the boundaries and responsibilities within the proxy relationship, especially with the rising adoption of sophisticated avatar proxies. A critical reflection is necessary on whether transferring traditional human-to-human proxy relationships to avatar proxies can occur without subjecting individuals to notable personal risks.

4. Conclusion

In an evolving digital age, there is a growing trend of humans either operating through or being represented by avatars. Traditionally, proxies, or stand-ins, have been utilized in scenarios where an individual is either unavailable or incapable of self-representation. Within such a proxy framework, the actions executed by the proxy are viewed as those of the actual person, who subsequently assumes responsibility for any consequences of those actions. A challenge in this setup is the "epistemic gap" where a proxy must act without concrete knowledge of the person's intentions. In human proxies, this gap is typically addressed using intuition and 'common sense', sometimes necessitating a nuanced evaluation of conflicting evidence. However, replicating this nuanced, context-aware approach in artificial systems, like avatars, might be challenging. Consequently, there's a potential danger that individuals might be held accountable for unintended actions of their avatar proxies. To conclude, while avatars offer novel opportunities for representation in various situations, it is essential to exercise caution when establishing human-avatar proxy relationships.

Find Paula Sweeney's article at Sweeney, P. Avatars as Proxies. Minds & Machines 33, 525–539 (2023). 

Paula Sweeney is a senior lecturer in Philosophy. She joined the University of Aberdeen in 2009.