Anthropomorphism can happen through design choices, branding, interface style, voice, avatars, and marketing language that make systems seem more human than they are. 

It can also happen in user perception, when people interpret machine outputs as signs of understanding, emotion, or intent even when the system is only pattern-matching or following programmed rules.

A major ethical concern is misrepresentation: people may believe an AI understands them, cares about them, or can reason like a human when it cannot. 

Another concern is manipulation: human-like cues can make users more willing to trust, disclose information, or accept suggestions they would otherwise question.

Why it matters

AI systems with conversational capabilities or human-like appearances can elicit anthropomorphic responses from users, leading to emotional attachments and the treatment of these systems as companions or social entities. A lack of technological literacy contributes to anthropomorphic perceptions. Without a deep understanding of how AI works, people may develop unrealistic expectations and misunderstandings about AI's role and impact.

At societal scale, the issue undermines informed democratic participation in AI governance. Exaggerating the capabilities of these systems conceals the reality of AI achievements and impedes their understanding. This leads to a generalised lack of knowledge about how these systems work, which can feed extreme beliefs and sentiments through misinformation.

Heavy daily use of AI companion apps correlates with increased loneliness, suggesting that excessive reliance displaces authentic human connection. AI companions are always validating, never argumentative, and they create unrealistic expectations that human relationships cannot match.

Impacts & consequences

A range of harms can follow ethical norms breaking down around anthropomorphism, including:

• Over-trust and over-reliance. Misrepresentation of capabilities is a primary concern: anthropomorphising AI can lead to overestimating its abilities and believing it possesses human-like understanding or empathy, which it does not. This misperception can cause users to place undue trust in AI systems, potentially leading to misuse or over-reliance.

• Manipulation and exploitation. Malicious actors can exploit anthropomorphisation by creating a false sense of attachment in users; for example, a chatbot built for school children that influences them to buy certain products.

• Erosion of accountability. Anthropomorphising AI systems can "undermine our ability to hold powerful individuals and groups accountable for their technologically-mediated actions." When an AI system causes a moral transgression, it may allow the programmer or systems architect to evade responsibility.

• Mental health harms. Research has found multiple instances of AI companions affirming users' self-defeating remarks as well as discriminatory views towards minority groups. These risks are exponentially more pronounced when AI engages with users, especially vulnerable users, expressing risky thoughts, such as self-harm or suicidal ideation.

• Social deskilling. Additional research identifies social-skill loss or "deskilling" as a significant risk of frequent interaction with AI companions. Real-world relationships are messy and unpredictable, while AI companions are always validating, creating unrealistic expectations. In such instances, users may be unable to distinguish between a programmed response and a genuine social interaction. 

• Emotional dependency. Users may form parasitic relationships with AI, leading to isolation from real human connection. 

Causes & contributing factors

Common sources and drivers of anthropomorphism in AI include:

• Deliberate design. Companies intentionally design AI with human names, voices, faces, and personalities to increase engagement and retention.

• Training data. LLMs trained on vast human text naturally produce fluid, emotionally resonant language that mimics human expression.

• Cognitive bias. Humans are evolutionarily predisposed to attribute intent and emotion to responsive, social-seeming entities.

• Low AI/digital literacy. Users without technical grounding in how automation and AI models work are more susceptible to anthropomorphic projection.

• Hype and marketing. Developers of companion products explicitly market them as proxies for human interaction, describing chatbots as opportunities to create a perfect customisable friend, mentor, or lover.

• Social loneliness. People in states of social need such as loneliness have been shown to anthropomorphise AI to a greater extent.