011 min
Where anthropomorphism shows up
Picture a product designer in a review meeting for a customer-support chatbot. The team has built two versions on the same language model, with the same error rate. Version A is labelled Support assistant and answers in plain, impersonal sentences. Version B has a human first name, a small smiling avatar, and apologises in the first person after a wrong answer.
Testers use both for a week and write notes. About Version A, the notes say wrong answer and did not find the page. About Version B, the notes say it got confused, it kept ignoring me and it was trying its best. The software is the same. Only three things changed: a name, a face and the word I.
Testers now describe Version B as an agent with a mind. They forgive it, blame it and expect it to learn. That reaction is anthropomorphism. It appeared before anyone told the testers what the chatbot is. Psychologists measured the same reaction in the 1940s, with no software involved.
021 min
Where research on anthropomorphism started
In 1944 the psychologists Fritz Heider and Marianne Simmel at Smith College made a film about two and a half minutes long. It shows a large triangle, a small triangle and a circle moving around a rectangle that has a door. The door opens and closes. The shapes have no faces and no voices. They only change speed and direction.
The film was shown to three groups of undergraduate women. In the first group, which was asked only to write down what happened, all but one student described the shapes as living beings, mostly people. Nineteen of the 34 students in that group told one connected story. Typical accounts described a fight and a chase.
Heider and Simmel studied perception, and their point was that movement alone was enough to make people see persons. The film does not say why. The next question is which features of an object trigger the reaction.

032 min
What triggers anthropomorphism
Heider and Simmel's shapes offered only movement, so movement is the first trigger to examine. In 2000 Patrice Tremoulet and Jacob Feldman reported that a single moving object, with no other cue, can look alive to viewers because of how it moves. The viewers rated the object as more alive when its speed and its direction changed at the same time. The authors read this as consistent with a simple rule: people class an object as alive when its path would be unlikely without a living cause.
Faces are the second trigger. People see faces in the front of a car, in a wheel rim or in a cloud. This is called pareidolia, the perception of a face in something that is not a face. Voices and written sentences are the third trigger. The effects section below shows what they do.
Two kinds of mind that people attribute
Not every attribution is the same. In 2007 Heather Gray, Kurt Gray and Daniel Wegner found that people judge minds on two separate dimensions, experience, such as feeling hunger, and agency, such as self-control. The finding came from online surveys in which people compared the mental abilities of many human and nonhuman characters.
A designer can read this as two separate choices. A chatbot that says I decided invites an attribution of agency. A chatbot that says I am sorry invites an attribution of experience. A product team can choose one without the other.

042 min
Why anthropomorphism happens
A trigger does not explain why one person anthropomorphizes more than another. Nicholas Epley, Adam Waytz and John Cacioppo proposed in 2007 that people anthropomorphize more when knowledge about humans is easy to apply, when they want to understand an agent, and when they lack social contact. The theory starts from a plain fact: anthropomorphism is common, but it is not constant. Its three factors say when it rises.
- Elicited agent knowledge. People hold a large store of knowledge about other people. A thing with a human feature brings that knowledge to mind, and it is easy to apply.
- Effectance motivation. People want to explain and predict what something will do. A complex or unpredictable system makes the wish stronger, and a mind is the explanation people know best.
- Sociality motivation. People want social contact. In 2008 Epley, Akalis, Waytz and Cacioppo reported that people without enough social ties to other humans may make up for it by seeing nonhuman agents as humanlike. Their three studies included people who were lonely for a long time and people who were made to feel lonely in the laboratory.
Back in the review meeting, the three factors predict who will treat Version B as a person. It is a tester with no one to talk to, using a system they cannot predict, that has a name and a first-person voice. Version A offers less to work with on the first factor.
One proposal for why the tendency exists is that treating an unclear shape as a possible agent costs little, while missing a real agent could cost a lot. This is a proposal, not a settled finding.
052 min
What anthropomorphism changes
Once a person decides that a machine has a mind, they act toward it differently. Two cases show how.
A car named Iris
In 2014 Adam Waytz, Joy Heafner and Nicholas Epley ran a driving-simulator experiment. Participants drove one of three cars: a normal car, a car that controlled its own speed and steering, or that self-driving car with a name (Iris), a gender (female) and a human voice that gave instructions. In the simulator, drivers of the self-driving car that had been given humanlike features reported more trust in it, showed a calmer heart rate during a simulated accident, and blamed the car and related parties less.
The two self-driving versions differed only in the humanlike features. The study measured trust, not whether the trust was deserved. And because the name, the gender and the voice came as a bundle, it cannot say which of them mattered.
ELIZA and the secretary
In 1966 Joseph Weizenbaum at MIT published ELIZA, a program that found key words in what a user typed and answered with stored reply patterns. Its best-known script imitated a Rogerian psychotherapist, who often repeats the patient's words back as questions. Weizenbaum's secretary reportedly asked him to leave the room so that she could talk with the program in private.
Weizenbaum was surprised. He later wrote that he had not expected so short an exposure to such a simple program to produce strong false beliefs in ordinary people. The reaction is now called the ELIZA effect.
Both cases explain what the designer saw in Version B. A name and a voice raised trust, and a conversation, even a scripted one, led people to treat the program as a listener.
061 min
How anthropomorphism shows up in product, design, and AI
The designer's two versions now have a clear trade-off. Version B is likely to earn more trust, as the Iris result suggests, and more forgiveness after mistakes. Both can help a product. Both can also hide a fault, because a user who forgives a chatbot may never report that it keeps failing.
For people who build products, the cues are easy to list: a name, an avatar, a voice, first-person sentences, apologies and a typing delay before an answer. Some are chosen on purpose, as with brand mascots and other forms of humanization. Others arrive by accident. A chatbot replies in sentences by default, so it carries conversational cues before anyone designs a persona.
For people who evaluate models, the same habit appears in the words of reviews and postmortems. Phrases such as the model wanted or the model knew assign an intention that the data does not show. A reviewer who writes the input, the output and what changed is describing the evidence.
When more humanlike stops helping
Masahiro Mori proposed in 1970 that people's affinity for a robot grows as it looks more human, until it reaches a dip that he named the uncanny valley. The uncanny valley is a proposal about appearance, and it is not a law of nature. It still warns that adding human cues does not raise comfort without limit.
071 min
How to guard against anthropomorphism
A team that wants Version B's trust without its risks has three jobs.
List every humanlike cue and decide each one on purpose.
Each cue makes knowledge about people easier to apply, which is the first factor in the theory. For every name, avatar, pronoun, apology and voice, write what the user will believe because of it, and whether that belief is true.
Test beliefs, not liking.
Ask testers, before and after a week of use, what the system can do, what it knows and what it does when it is wrong. A gap between their answers and the real behaviour is a wrong mental model. Remove the cue that causes the gap, or state the limit where the user can see it.
Describe behaviour in reviews.
The Iris participants blamed the car less. An agent frame moves blame from the system to a someone. In an incident review, write the cause as an input, a rule and an output.
Do not count on a label that says this is a bot. A label does not remove the other cues, so measure what people believe after they see it.
082 min
Where anthropomorphism misleads
Anthropomorphism does more than raise trust in machines. It can lead an observer to read a mind into behaviour that has a simpler cause, and it can make a weak finding look strong.
Clever Hans
A horse called Clever Hans appeared to do arithmetic and other tasks at exhibitions in Germany in the early 1900s. His owner, Wilhelm von Osten, was a mathematics teacher. In 1907 the psychologist Oskar Pfungst showed that Hans was not doing arithmetic and was reacting to signals from his trainer. The trainer gave involuntary cues and did not know it.
The audience had inferred a mind for arithmetic. The simpler account was that the horse followed signals from the people around him. When an animal or a program seems to understand, check for a plain cause first.
A contested case: hurricane names
In 2014 Kiju Jung and colleagues reported in a US journal that hurricanes with feminine names had killed more people than hurricanes with masculine names. They argued that the names changed how severe people expected the storms to be. Two statisticians reanalysed the same death records and concluded that the result was based on biased presentation and invalid statistics. Other researchers also published objections in the same journal. The finding is unsettled.
A practitioner should not use it as evidence when naming a product or a storm warning. It shows how easily a story about human traits can attach itself to a statistic.

091 min
Anthropomorphism vs. nearby concepts
Several nearby terms cover parts of the same ground. The axis that separates them is who is attributing what, and to whom.
| Term | What it describes | How it differs from anthropomorphism |
|---|---|---|
| Personification | A writer or speaker gives human traits to a thing as a figure of speech | The speaker knows the thing is not human. Psychology uses anthropomorphism for the belief and behaviour of the observer as well |
| Anthropocentrism | Treating humans as the standard for judging everything else | It is about viewpoint and values, and it needs no mind to be seen in an object |
| Dehumanization | Treating people as less than fully human | It is the inverse process, and the 2007 theory was written to cover both |
| Animism | The belief that non-living things are alive, a term Piaget used for children's thinking | It is a belief about life, while anthropomorphism is about human traits such as intentions |
?4 questions
Questions people ask
Is anthropomorphism always a mistake?
Does giving a chatbot a name cause anthropomorphism?
Does anthropomorphism apply only to machines?
What is the ELIZA effect?
§14 sources
Sources
Show all 14 sourcesHide sources
Epley, N., Waytz, A. & Cacioppo, J. T. (2007). On seeing human: a three-factor theory of anthropomorphism. Psychological Review.
Epley, N., Akalis, S., Waytz, A. & Cacioppo, J. T. (2008). Creating social connection through inferential reproduction: loneliness and perceived agency in gadgets, gods, and greyhounds. Psychological Science.
Heider, F. & Simmel, M. (1944). An experimental study of apparent behavior. The American Journal of Psychology.
Tremoulet, P. D. & Feldman, J. (2000). Perception of animacy from the motion of a single object. Perception.
Gray, H. M., Gray, K. & Wegner, D. M. (2007). Dimensions of mind perception. Science.
Waytz, A., Heafner, J. & Epley, N. (2014). The mind in the machine: anthropomorphism increases trust in an autonomous vehicle. Journal of Experimental Social Psychology.
A summary of the Waytz, Heafner and Epley (2014) study by the Association for Psychological Science.
Weizenbaum, J. (1966). ELIZA, a computer program for the study of natural language communication between man and machine. Communications of the ACM.
An encyclopedia entry on ELIZA, including the account of Weizenbaum's secretary.
Mori, M. (1970; English translation 2012). The uncanny valley. IEEE Spectrum.
An encyclopedia entry on the horse Clever Hans and Pfungst's 1907 investigation.
Jung, K., Shavitt, S., Viswanathan, M. & Hilbe, J. M. (2014). Female hurricanes are deadlier than male hurricanes. PNAS.
Christensen, B. & Christensen, S. (2014). Are female hurricanes really deadlier than male hurricanes? PNAS.
BrainFacts.org (2020). Anthropomorphism: seeing the human in plants, pets and possessions.





