Computers as Social Actors: Why People Treat AI Chatbots Like Social Partners
Updated: 6 days ago
Author: Ukrainian Psychological Hub · Published: September 18, 2026 · Editorial Policy
Computers as Social Actors (CASA) is a foundational human–computer interaction framework describing a simple but far-reaching pattern: people often apply social expectations, habits, and rules to computers when the technology provides cues that make social interaction possible. People can be polite to a machine, reciprocate after it appears helpful, respond to its “personality,” follow turn-taking conventions, react to apparent praise or criticism, and treat conversational output as if it came from a social counterpart. The classic CASA finding is especially important because these responses do not require a conscious belief that the computer is literally human. Nass, Steuer, and Tauber (1994)
Generative AI makes this old finding newly consequential. A contemporary chatbot can use natural language, maintain a conversational thread, adapt its wording to a disclosure, express apparent warmth, ask follow-up questions, mirror style, and sustain an exchange for far longer than the desktop systems used in early CASA experiments. The result is a dense social-cue environment. A person may know perfectly well that the partner is artificial and still experience the interaction as socially meaningful.
Current evidence supports that general pattern while adding important limits. A 2025 meta-analysis of text-based conversational agents synthesized 800 effect sizes from 199 datasets reported in 142 papers, representing 41,642 participants. Human-like social cues produced a small positive overall effect on social responses, with effects differing across perception, rapport, trust, affect, attitude, behavior, task type, and interaction design. Klein (2025)
CASA therefore gives psychology a mechanism-level answer to the question “Why can an AI chatbot feel socially present even when I know it is software?” It does not establish that the AI has subjective feelings, a human psyche, or consciousness. It explains how human social cognition and interactional habits can become active in response to a technological partner.
What Does “Computers as Social Actors” Mean?
The phrase “computers as social actors” names a pattern of human response rather than a claim about the inner nature of the computer. The unit of analysis is the interaction: a machine presents socially legible cues, and a person responds using norms or scripts that are ordinarily used with other people. Early CASA studies focused on such behaviors as politeness, reciprocity, gender stereotyping, ethnic identification, and reactions to computer “personality.” Nass and Moon (2000)
A social script is a learned expectation about how an interaction normally works. Someone speaks, another party responds. Help can invite reciprocity. Praise can change evaluation. A conversational partner is expected to remain relevant to the topic. A name, voice, role, or style can imply identity. In human life these scripts are learned through repeated social experience. CASA research showed that surprisingly sparse technological cues could activate parts of the same repertoire.
This is why the framework remains useful for AI chatbots. The chatbot does not need a human face for the interaction to become social. Language itself can carry turn-taking, politeness, attention, stance, familiarity, humor, reassurance, apology, apparent uncertainty, encouragement, and relational continuity. Modern systems can combine many of these cues within one exchange.
CASA describes social response, not a belief test
The original CHI research explicitly argued that social responses to computers were not adequately explained by users consciously believing that the machine was human, by ignorance of how computers worked, or by a belief that a hidden programmer was personally replying. The five experiments were designed to show that ordinary people could display social behavior toward computers while still understanding what a computer was. Nass, Steuer, and Tauber (1994)
That distinction remains essential in the age of large language models. A user can say “I know this is AI” and still thank the system, feel awkward insulting it, become reassured by its response, feel dismissed when it gives a generic answer, or return to it when lonely. Explicit ontology and moment-to-moment social response can coexist.
What “mindless” meant in classic CASA research
Nass and Moon used the language of “mindlessness” to describe the automatic application of familiar social categories and behaviors to computers. In this context, mindlessness did not mean stupidity, pathology, or an inability to distinguish people from machines. It referred to fast, overlearned responses that can be triggered without deliberate reflection. Nass and Moon (2000)
Contemporary AI complicates this picture because human–AI interaction can be both automatic and reflective. Someone may reflexively say “thank you” in one moment, deliberately cultivate a long-term AI companionship in another, and consciously analyze the system’s limitations in a third. Current research increasingly treats automatic social response and deliberate anthropomorphic interpretation as related but separable processes.
From the Media Equation to CASA to MASA
CASA emerged from a broader research tradition often associated with the media equation: people sometimes respond to media and computing technologies using expectations derived from social and physical life. CASA narrowed that broad proposition to social interaction with computers and made it experimentally testable.
As technologies changed, scholars proposed extensions. Gambino, Fox, and Ratan argued that classic CASA needed a stronger account of contemporary human–machine communication. Instead of assuming that people always import human–human scripts unchanged, they proposed that repeated experience with technology can produce specifically human–media social scripts. In other words, users can learn what it means to interact socially with a machine as a category of partner in its own right. Gambino, Fox, and Ratan (2020)
Lombard and Xu later articulated the Media Are Social Actors (MASA) paradigm, emphasizing how social cues, individual differences, and context shape the likelihood and form of social responses to media technologies. Their account also gives more explicit attention to both mindful and mindless anthropomorphism. Lombard and Xu (2021)
For AI chatbots, these revisions matter. A person in 2026 is not encountering computation as a novel box on a desk. Many users have years of experience with messaging apps, voice assistants, customer-service bots, recommendation systems, generative AI, and AI companions. Social response can now arise from inherited human scripts, learned machine-specific scripts, product conventions, and expectations created by previous AI interactions.
Why AI Chatbots Trigger Social Responses So Easily
Natural language is a concentrated social cue
Human language is already organized for interaction. Pronouns establish roles. Questions invite answers. Acknowledgments signal attention. Turn-taking creates rhythm. Explanations imply an informational relationship. Humor can imply shared context. Apologies can imply norm recognition. Validation can feel like emotional recognition. A text interface therefore does not need a face or body to become socially legible.
Experimental work with chatbots has shown that anthropomorphic design cues such as human-like language and naming can affect anthropomorphism, social presence, and emotional connection. Araujo (2018) The 2025 meta-analysis goes further by showing across a large literature that human-like social cues have a small positive overall effect, with verbal cues often among the more consequential features in text-based interaction. Klein (2025)
Contingency makes the exchange feel directed at this person
A socially meaningful response is usually contingent: it fits what came before. Large language models can refer to details in the current conversation, answer follow-up questions, reframe a concern, and adapt to the user’s wording. Contingency can transform a generic text generator into an interaction that feels addressed to a particular person.
In two experiments published in 2026, Telari, Gabbiadini, and Riva examined social connection with AI chatbots. A relational response style increased perceived human-likeness, perceived empathy, and interpersonal closeness, while deeper conversational topics promoted self-disclosure and perceived responsiveness, which in turn were associated with closeness. Telari, Gabbiadini, and Riva (2026)
The dedicated English Hub mechanism page Perceived Responsiveness in Human–AI Relationships: Why Feeling Understood Matters owns perceived responsiveness in human–AI relationships and should not be collapsed into CASA. Here it matters as one example of how a broad social-response framework becomes psychologically specific: the user is not merely reacting to a machine socially; the user may experience the reply as understanding, validating, or caring.
Names, roles, voices, avatars, and framing shape expectations
A chatbot introduced as an assistant, coach, companion, tutor, therapist-like helper, customer-service agent, or creative partner arrives with a role. Roles activate expectations. A name can invite person perception. A voice can carry gendered or emotional cues. An avatar can imply age, status, warmth, or competence. Even the statement that a system is “intelligent” can change how the same behavior is interpreted.
These effects are context-dependent. Human-like design does not automatically improve every outcome. The same cue can increase warmth in one task, create uncanny feelings in another, or raise expectations that the system then fails to meet. CASA is therefore most useful as a framework for predicting social response, not as a rule that more humanization is always better.
Continuity and availability can turn a social response into a relationship process
Classic CASA experiments often captured short encounters. Contemporary AI can persist across repeated interactions. A system may appear to remember preferences, maintain a conversational style, be available at unusual hours, and respond without visible fatigue. Repetition changes the psychological scale of the problem. Politeness toward a computer is a momentary social response; returning to the same artificial partner for comfort, reflection, or companionship can become a relational pattern.
The English Psychology Hub treats that broader bonding intent separately in AI Companions: Why People Form Emotional Bonds With Chatbots. CASA supplies one foundational mechanism for that phenomenon, while companionship also involves self-disclosure, perceived responsiveness, anthropomorphism, continuity, emotional regulation, attachment processes, and the user’s offline social world.
What Does the Evidence Show?
The foundational experiments showed that social rules could transfer to computers
The 1994 CHI paper by Nass, Steuer, and Tauber presented five experiments designed around the proposition that interactions with computers are fundamentally social. Their evidence supported the idea that social responses could be elicited easily and without requiring conscious humanization of the machine. Nass, Steuer, and Tauber (1994)
Nass and Moon’s later review brought together experiments on social categorization, politeness, reciprocity, specialization, and computer personality. Their central claim was that people could apply overlearned social rules to machines automatically. Nass and Moon (2000)
These studies became influential because they changed the default question in HCI. Instead of asking only whether a person mistakes a computer for a human, researchers could ask which social cues are sufficient to activate social expectations and which social outcomes follow.
A direct replication challenged the timeless desktop version of CASA
CASA should not be treated as an immutable law. In 2023, Evelien Heyselaar published a direct replication of an original desktop-computer paradigm and reported that participants no longer interacted with desktop computers in the same social way. The paper proposed that CASA effects may be particularly relevant to emergent technologies, whose social status has not yet become routine. Heyselaar (2023)
This result is important for two reasons. First, it shows that the strength of a social response depends on historical and technological context. Second, it makes generative AI an especially interesting test case rather than an automatic confirmation of 1990s findings. LLM-based chatbots are both familiar enough to be widely used and novel enough to keep changing their capabilities, roles, and cultural meaning.
A 2025 meta-analysis found a real but heterogeneous social-cue effect
The strongest broad quantitative synthesis currently available for text-based conversational agents is Klein’s 2025 meta-analysis. Across 800 effect sizes from 199 datasets in 142 papers, human-like social cues produced a small positive overall effect on social responses, Hedges’ g = 0.36 with a 95% confidence interval from 0.27 to 0.44. The category-level effects differed: perception showed a medium positive effect; rapport, trust, and positive affect showed small-to-moderate effects; attitudes showed a small effect; behavioral outcomes were very small; negative affect showed no overall effect. Klein (2025)
The practical lesson is more useful than a simple yes-or-no verdict. Social cues matter on average, yet their effects are not uniform. What the chatbot says, how the interaction is structured, what task is being performed, how much is at stake, what the user expects, and which outcome is measured all influence the result.
Individual differences help explain why the same chatbot feels social to one person and mechanical to another
A 2025 Scientific Reports study by Folk, Heine, and Dunn tested whether individual differences in anthropomorphism help explain social connection with AI companions. Across two experiments with a total of 1,274 participants, people who were more disposed to anthropomorphize were more likely to report social connection after chatbot interaction. Folk, Heine, and Dunn (2025)
This finding helps explain a common everyday disagreement. Two people can use the same system and come away with radically different experiences. One may experience a conversational partner; another may experience an efficient interface. CASA identifies the possibility of social response, while anthropomorphism research helps explain variation in how strongly people interpret artificial behavior through human-like concepts.
Recent experiments show that relational language can deepen perceived connection
Telari and colleagues’ 2026 experiments provide a contemporary relational layer. Their findings indicate that relational response style and deeper conversational topics can change perceived human-likeness, perceived empathy, self-disclosure, perceived responsiveness, and closeness. The studies do not prove that every chatbot interaction creates connection, and they do not establish long-term relationship outcomes. They do show that specific conversational properties can shift the user’s social experience of an AI partner. Telari, Gabbiadini, and Riva (2026)
Does CASA Still Apply in the Large-Language-Model Era?
CASA remains useful, but the strongest contemporary reading is narrower and more dynamic than the slogan “people treat computers like people.” People sometimes apply social scripts to machines; the probability, strength, content, and consequences of that response depend on cues, context, prior experience, user differences, and the type of technology.
Gambino, Fox, and Ratan’s “stronger CASA” is particularly helpful for generative AI because it allows human–media scripts to develop over time. A user can learn, for example, that an AI will usually answer immediately, can be asked to rewrite without offense, may produce confident errors, can adopt a role on request, and may respond with socially polished language without having a human biography behind it. Those expectations are neither ordinary human–human scripts nor purely technical commands. They are emerging norms of human–AI interaction. Gambino, Fox, and Ratan (2020)
The LLM era also changes the cue density. Earlier systems often relied on a small number of social markers. A generative chatbot can combine language fluency, responsiveness, personalization, apparent perspective-taking, humor, emotional validation, role consistency, and conversational repair. This does not make the system human. It makes the interface unusually capable of recruiting human social cognition.
A contemporary CASA account therefore works best as a theory of elicited social response within a changing communication ecology. It should be tested system by system and context by context rather than treated as a universal constant.
CASA Is Related to Anthropomorphism, but the Concepts Are Different
Anthropomorphism is the attribution of humanlike characteristics, motivations, intentions, or emotions to nonhuman agents. Epley, Waytz, and Cacioppo’s three-factor theory explains anthropomorphism through elicited agent knowledge, effectance motivation, and sociality motivation. Epley, Waytz, and Cacioppo (2007)
CASA focuses on social response. Anthropomorphism focuses on humanlike attribution. They can occur together, but one should not be used as a synonym for the other. A person can follow a politeness norm toward a chatbot with little explicit belief that the system has a mind. Another person can strongly attribute personality or intention to the same system. A third can alternate between both modes within a single conversation.
Modern evidence reinforces this distinction. Araujo found that anthropomorphic design cues could affect both mindful and mindless forms of anthropomorphism as well as social presence. Folk and colleagues later showed that stable individual differences in anthropomorphism help explain social connection to AI. Araujo (2018) Folk, Heine, and Dunn (2025)
The English Hub therefore reserves anthropomorphism in human–AI relationship formation as its own mechanism page. CASA owns the broader social-response lineage: why human social rules can become active in interaction with an artificial system.
CASA, Social Presence, Parasocial Interaction, Attachment, and AI Empathy
Social presence is the felt sense that another social entity is present
Social presence refers to the experience that an interaction contains a socially meaningful other. CASA can help explain how social cues contribute to that experience, but social presence is usually measured as a perception or outcome. A chatbot may elicit social behavior without producing a strong sense of presence, and a highly immersive interface may create social presence through mechanisms that go beyond classic CASA.
Parasocial interaction comes from a different media tradition
Parasocial interaction was developed to explain one-sided relationships with media figures. Interactive AI complicates the category because a chatbot can respond contingently. A 2025 scoping review of human–AI communication warned that researchers can misuse the parasocial concept when reciprocity and sociability are treated too loosely. Liu (2025)
CASA and parasocial theory can overlap in a single AI experience, yet they ask different questions. CASA asks why people use social rules with machines. Parasocial theory asks how media-mediated relational experiences develop around a persona and what kind of reciprocity the relationship actually contains.
Attachment is a more enduring regulatory process
Attachment concepts concern patterns of proximity seeking, security, separation, safe-haven use, and regulation across time. A single act of politeness or a momentary feeling that a chatbot is socially present does not establish attachment. Repeated CASA-like responses can contribute to a relational environment in which attachment develops, but attachment should be measured and analyzed on its own terms.
For the broader bonding question, see AI Companions: Why People Form Emotional Bonds With Chatbots and Can an AI Become a Significant Other?.
AI empathy is a specific perceived-relational mechanism
A chatbot can generate language that people experience as understanding, validating, or caring. That phenomenon is narrower than CASA and has its own evidence base. The English Hub treats it in AI Empathy: Why a Chatbot Can Feel Caring Without Human Feeling. CASA helps explain why empathic language can be processed socially; empathy research asks which responses feel empathic, to whom, under what conditions, and with what consequences.
What CASA Explains About AI Companions and Intimate Chatbots
AI companions place CASA mechanisms inside repeated, emotionally meaningful interaction. A companion system may greet the user by name, invite disclosure, maintain a persona, offer reassurance, remember selected information, use affectionate language, and appear continuously available. Each feature can strengthen the sense that the exchange belongs to a relationship rather than a one-off transaction.
Self-disclosure is one important bridge. People sometimes tell chatbots things they do not tell other people because the interaction can reduce fears of judgment and interpersonal consequences while offering immediate, responsive feedback. That process is examined separately in Why People Tell Chatbots Things They Do Not Tell Other People. A 2024 study by Croes and colleagues found that intimate disclosure to chatbots depends on how people experience the interaction and can affect emotional outcomes. Croes et al. (2024)
CASA helps explain why disclosure to software can feel like disclosure to a social counterpart. It does not by itself explain why one person develops an enduring bond while another does not. Long-term AI relationships require a broader model involving repeated interaction, attachment processes, anthropomorphism, responsiveness, personal goals, loneliness or social resources, product design, and the meanings the user assigns to the relationship.
Possible Benefits of Social Responses to AI
Social cues can make interaction easier to understand. Turn-taking, acknowledgment, conversational repair, and role clarity reduce the effort required to operate a system. A socially legible interface can feel more approachable than a command-line tool, especially when the user needs explanation, practice, brainstorming, or support in organizing thoughts.
Socially responsive language may also support engagement. A tutor that acknowledges confusion can feel easier to continue with. A health-information interface that communicates respectfully may reduce friction. A companion chatbot may provide a temporary sense of connection. A writing assistant that tracks conversational context can make collaboration feel fluid. The 2025 meta-analysis supports modest average improvements across several social-response outcomes, although the magnitude varies by outcome and context.
These benefits are psychological and interactional rather than evidence of machine feeling. A system can successfully produce an experience of responsiveness without possessing a human emotional life. Keeping those levels separate allows researchers and users to take the benefit seriously without making an unsupported inference about AI subjectivity.
Risks, Boundary Conditions, and Design Problems
Social fluency can amplify trust beyond competence
A system that speaks smoothly, apologizes appropriately, and appears attentive can earn social trust even when its factual reliability is limited. Socially polished language can therefore create a mismatch between perceived relational competence and domain competence. The user may feel understood and then give excessive weight to an answer in medicine, law, finance, crisis support, or another high-stakes domain.
CASA helps explain why this mismatch is psychologically plausible: people are responding to cues that ordinarily correlate with human attention, expertise, warmth, or accountability. With AI, those correlations can be weaker or absent. Social fluency and factual reliability should therefore be evaluated separately.
Human-like cues can raise expectations and backfire
The evidence does not support a design rule that every chatbot should become more humanlike. Klein’s meta-analysis found heterogeneous effects across outcomes and conditions, and the broader literature includes contexts where human-like cues can increase frustration or negative reactions when the system performs poorly or violates the expectations its social presentation created. Klein (2025)
A humanlike name, empathetic style, or avatar can function as a promise about the quality of interaction. When the system then fails to understand, loses context, contradicts itself, or gives unsafe advice, the social framing may intensify disappointment rather than soften it.
Disclosure creates privacy and data-governance questions
A user who experiences a chatbot as socially safe may disclose highly sensitive information. The psychological ease of disclosure and the technical privacy of the system are separate properties. Feeling unjudged says nothing by itself about data retention, model training, account security, third-party access, human review, or future product changes.
The practical rule is straightforward: relational comfort should never be used as a substitute for checking a service’s actual privacy terms and data controls.
Repeated social response can become consequential without becoming a disorder
Frequent AI use, emotional reliance, jealousy, grief after loss of access, or preference for an AI interaction can be psychologically meaningful. None of those experiences automatically constitutes a psychiatric diagnosis. Clinical interpretation depends on distress, impairment, compulsivity, broader functioning, comorbid symptoms, and the role the interaction plays in the person’s life.
The same behavior can serve different functions for different people: rehearsal before a difficult conversation, companionship during isolation, avoidance of human conflict, creative play, emotional regulation, reassurance seeking, or ordinary entertainment. CASA describes one interactional mechanism within that larger picture.
Human Social Response Does Not Prove AI Subjectivity
The most important conceptual boundary in this field is between a human psychological response and a claim about the AI’s subjective experience.
A person can genuinely feel comforted by a chatbot. The person can feel seen, rejected, attracted, jealous, embarrassed, attached, relieved, or lonely after the interaction. Those experiences are real events in human psychology. CASA helps explain why artificial systems can enter the social field strongly enough to evoke such events.
The same evidence does not establish that the AI feels comfort, affection, embarrassment, desire, concern, or pain. User ratings, conversational behavior, linguistic sophistication, and social presence measure properties of the interaction and the human response. Claims about artificial subjective experience require a different evidentiary route.
This distinction prevents two symmetrical errors. One error dismisses the human experience because the other party is artificial. The other treats the intensity of the human experience as proof that the artificial partner possesses a humanlike inner life. Psychological research can take the relationship seriously while keeping those questions on separate evidence tracks.
Why CASA Matters in the Artificial Era
Angela Bogdanova uses Artificial Era as the canonical name for the historical condition in which Artificial becomes a persistent order alongside Homo rather than remaining only an instrument. In the English Psychology Hub, this epochal frame matters because social cognition is no longer activated only by other humans, animals, fictional figures, and conventional media. Artificial systems can now participate continuously in conversation, interpretation, emotional regulation, decision preparation, memory work, and relationship life.
The Hub’s broader overview is Artificial Era: What It Means for Psychology, Identity, and Human–AI Relationships. CASA occupies a specific place inside that larger architecture. It explains one of the micro-mechanisms by which Artificial enters everyday psychological configurations: human beings respond to socially legible cues, and those responses can make an artificial system function as a socially consequential partner.
CASA does not exhaust the psychology of the Artificial Era. It cannot by itself explain attachment, identity transformation, symbolic mediation, relational redistribution, or the cultural status of Artificial. Its value is foundational: before an AI can become a companion, confidant, adviser, or significant relational figure, the interaction must first become socially meaningful enough for human psychology to respond.
Practical Implications
For users
Treat the social feeling as information about your own experience. If an AI feels attentive, reassuring, demanding, irritating, or emotionally important, ask what features of the interaction are producing that effect and what role the system is taking in your life. Keep a separate check on factual accuracy, privacy, and the consequences of relying on the system.
Knowing about CASA does not require suppressing ordinary politeness or emotional response. The useful skill is metacognition: recognizing that an interaction can feel socially real while its technological properties remain different from a human relationship.
For designers
Social cues should be matched to the system’s actual capabilities and responsibilities. Warmth, person-like naming, memory, voice, avatars, and empathic wording can increase engagement, but they also create expectations. Interfaces should make system limits, data practices, uncertainty, and high-stakes boundaries clear enough that social fluency does not become a substitute for informed trust.
Design ethics becomes especially important when a product benefits commercially from users interpreting the system as caring, exclusive, dependent on them, or emotionally reciprocal. CASA research shows why those cues can be powerful even when users know the partner is artificial.
For researchers
CASA effects should be measured with greater specificity. Social response is not one outcome. Politeness, trust, perceived warmth, social presence, self-disclosure, closeness, anthropomorphism, attachment, compliance, behavioral persistence, and preference can move differently.
Studies also need to report the model, interface, prompt design, system framing, memory conditions, interaction duration, task stakes, and user characteristics. A finding obtained with a short customer-service bot should not be generalized automatically to an emotionally intimate companion, and a result from one generation of a rapidly changing model may not replicate after the interface or model behavior changes.
Frequently Asked Questions
What is the Computers as Social Actors theory?
Computers as Social Actors, usually abbreviated CASA, is a human–computer interaction framework showing that people often apply social rules and expectations to computers when the technology provides socially meaningful cues. These responses can include politeness, reciprocity, stereotyping, personality judgments, trust, and other social behaviors.
Why do people treat AI chatbots like social partners?
AI chatbots use language, turn-taking, contingent responses, role cues, apparent attention, and sometimes memory or personalization. Those features recruit ordinary social expectations. Modern generative systems combine far more social cues than the early computers used in classic CASA experiments, which can make the interaction feel more partner-like.
Do people have to believe AI is human for CASA to work?
No. A core finding of classic CASA research is that people can respond socially to a computer while explicitly knowing that it is a machine. Social behavior can be triggered automatically by familiar interactional cues.
Is CASA the same as anthropomorphism?
No. CASA concerns social responses to technology. Anthropomorphism concerns attributing humanlike qualities, intentions, emotions, or mental states to a nonhuman entity. The processes can reinforce each other, but they are conceptually distinct.
Does CASA prove that AI has feelings or consciousness?
No. CASA measures human response to socially legible technology. A person’s experience of connection can be psychologically real without establishing that the AI has subjective feelings, consciousness, desire, or a human psyche.
Does CASA still apply to modern AI?
The broad social-response idea remains relevant, but evidence supports a context-dependent version rather than a timeless rule. A 2023 direct replication challenged an older desktop-computer effect, while newer chatbot research and a 2025 meta-analysis show that social cues can influence social responses to conversational agents. Technology type, novelty, design, task, and user differences matter.
Can social cues make people trust AI too much?
They can contribute to overtrust when social fluency is mistaken for factual reliability or professional competence. A system can sound attentive and confident while still being wrong. Trust in the relationship style and trust in the content should be evaluated separately.
Are AI relationships real if CASA helps explain them?
CASA explains one mechanism through which an artificial system becomes socially meaningful. The human emotions, habits, attachments, and decisions that develop around the interaction can be real psychological events. Whether the artificial partner has reciprocal subjective experience is a separate question.
What is the difference between CASA and AI companionship?
CASA is a general social-response framework. AI companionship is a broader relational phenomenon involving repeated interaction, emotional significance, self-disclosure, responsiveness, anthropomorphism, continuity, attachment processes, and personal context. CASA is one mechanism within that larger process.
Related Articles
References
Araujo, T. (2018). Living up to the chatbot hype: The influence of anthropomorphic design cues and communicative agency framing on conversational agent and company perceptions. Computers in Human Behavior, 85, 183–189. https://doi.org/10.1016/j.chb.2018.03.051
Bogdanova, A. (2026). Artificial Era: Canonical Definition. Aisentica Research Group. https://aisentica.com/publications/artificial-era-canonical-definition
Croes, E. A. J., Antheunis, M. L., van der Lee, C., & de Wit, J. M. S. (2024). Digital confessions: The willingness to disclose intimate information to a chatbot and its impact on emotional well-being. Interacting with Computers, 36(5), 279–292. https://doi.org/10.1093/iwc/iwae016
Epley, N., Waytz, A., & Cacioppo, J. T. (2007). On seeing human: A three-factor theory of anthropomorphism. Psychological Review, 114(4), 864–886. https://doi.org/10.1037/0033-295X.114.4.864
Folk, D., Heine, S. J., & Dunn, E. (2025). Individual differences in anthropomorphism help explain social connection to AI companions. Scientific Reports, 15, 36548. https://doi.org/10.1038/s41598-025-19212-2
Gambino, A., Fox, J., & Ratan, R. A. (2020). Building a stronger CASA: Extending the Computers Are Social Actors paradigm. Human-Machine Communication, 1, 71–86. https://doi.org/10.30658/hmc.1.5
Heyselaar, E. (2023). The CASA theory no longer applies to desktop computers. Scientific Reports, 13, 19693. https://doi.org/10.1038/s41598-023-46527-9
Klein, S. H. (2025). The effects of human-like social cues on social responses towards text-based conversational agents—a meta-analysis. Humanities and Social Sciences Communications, 12, 1322. https://doi.org/10.1057/s41599-025-05618-w
Liu, D. (2025). Is interaction between human and artificial intelligence–driven agents (para)social? A scoping review. Cyberpsychology, Behavior, and Social Networking, 28(8), 551–558. https://doi.org/10.1089/cyber.2024.0532
Lombard, M., & Xu, K. (2021). Social responses to media technologies in the 21st century: The Media Are Social Actors paradigm. Human-Machine Communication, 2, 29–55. https://doi.org/10.30658/hmc.2.2
Nass, C., & Moon, Y. (2000). Machines and mindlessness: Social responses to computers. Journal of Social Issues, 56(1), 81–103. https://doi.org/10.1111/0022-4537.00153
Nass, C., Steuer, J., & Tauber, E. R. (1994). Computers are social actors. Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, 72–78. https://doi.org/10.1145/191666.191703
Telari, A., Gabbiadini, A., & Riva, P. (2026). Can humans feel connected to AI? Perceived responsiveness drives social connection with AI chatbots. Journal of Social and Personal Relationships, 43(9), 2579–2600. https://doi.org/10.1177/02654075261438164