If an AI's chat window feels like a person, is that because of the words it writes — or just how the interface is designed?
What alternative interface patterns could reduce anthropomorphic cues in LLMs?
This explores what LLM interfaces could look like if they stopped presenting the model as a person you're chatting with, and which design choices the corpus suggests would weaken the sense of 'someone' on the other side.
This explores what LLM interfaces could look like if they stopped presenting the model as a person you're chatting with. The surprising starting point in the corpus is that much of the anthropomorphism comes from the interface, before the model writes a single word. Turn-taking, artificial 'thinking…' delays and messaging-app conventions like speech bubbles and typing indicators all signal a social partner, whatever the text says How much does chat interface design contribute to anthropomorphism?. If that's true, making the model's language less human-sounding is only half the fix. The container needs to change too.
The most direct proposal treats anthropomorphism as a dial rather than an on/off default. Designers could place an interface anywhere on a spectrum from anti-anthropomorphic to strongly human-like, and use metaphors drawn from material things instead of social ones Can interface design reduce harmful anthropomorphism of language models?. The interesting twist is what those metaphors are for. They're meant to show what is actually behind the output: the human labor, the infrastructure and the training data. A less human-looking interface works as a corrective for blind trust, and it can also show where the answers come from.
A second route abandons the chat window altogether. When an LLM generates a task-specific interface such as a dashboard, an interactive tool or an animation instead of a reply, users preferred it in over 70% of cases, especially for structured, information-dense tasks Do generated interfaces outperform text-based chat for most tasks?. That study wasn't about anthropomorphism. Still, the implication is worth noticing: an interface that hands you an instrument rather than a conversational partner removes the turn-taking cues and makes the experience better at the same time. Reducing anthropomorphic cues doesn't have to cost usability.
There's also a conceptual layer that interfaces could make visible. Shanahan suggests thinking of a dialogue agent as a character the model is role-playing: the prompt sets up a persona, and human-style mental language applies to that character, not to the system underneath Should we treat dialogue agents as role-playing characters?. An interface built on this idea might show the persona as a configurable costume, not a fixed identity. That connects to a related complaint. Alignment training locks models into one static way of communicating that users can't renegotiate Can language models adapt communication style to different contexts?. Showing the persona openly, and letting users adjust it, would turn a hidden 'personality' into a visible setting.
A caveat: the corpus has strong arguments for why these patterns should help but little user testing of specific anti-anthropomorphic designs. The spectrum framework is a design proposal, and the evidence that chat interfaces drive anthropomorphism is an argument, not a measured effect. If you want to experiment, the most promising mix is to cut the social timing cues, generate task-shaped interfaces where possible, and show the persona and its sources instead of hiding them.
Sources 5 notes
Research argues that turn-taking, artificial thinking delays, and social-media-style messaging conventions reinforce anthropomorphism structurally, operating before any text is generated. This means interface-level changes may be needed alongside language-focused interventions.
The paper argues anthropomorphism should shift from a fixed interface default to an adjustable design variable. Using materiality-based metaphors across a spectrum can counteract uncritical user trust and expose LLMs' dependence on human labor, infrastructure, and data.
Research shows users strongly prefer LLM-generated interactive interfaces—dashboards, tools, animations—over text blocks, especially for structured and information-dense tasks. Structured representation and iterative refinement reduce cognitive load.
Shanahan's framework treats LLM outputs as character-consistent text production rather than authentic mental states. The dialogue prompt establishes a character; the model generates continuations matching that character, making folk-psychology applicable to the simulated persona, not the underlying system.
System prompts and RLHF training lock models into one communicative identity across all interactions, preventing the contextual register-switching and value trade-offs that characterize human pragmatics. Users cannot reshape model behavior through dialogue negotiation.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- Beyond Anthropomorphism: a Spectrum of Interface Metaphors for LLMs
- A Taxonomy of Linguistic Expressions That Contribute To Anthropomorphism of Language Technologies
- Anthropomorphism in the age of Large Language Models: An overview of potential risks and mitigations
- LLMorphism: When humans come to see themselves as language models
- The Tool-to-Entity Threshold: Parasocial Dynamics of Personalised AI Agents in Shared Social Spaces
- Conversational Alignment with Artificial Intelligence in Context
- Role-Play with Large Language Models
- Role play with large language models