CoChat
Sign Up
Cited by 29Ethics and Social Impacts of AISocial Robot Interaction and HRICybernetics and Technology in Society

The compassion illusion: can artificial empathy ever be emotionally authentic?

Open access
K. G., Ajeesh, Joseph, Jeena·2025·Frontiers in Psychology
DOI 10.3389/fpsyg.2025.1723149PMC PMC12665657
Open this paper in CoChat — free accountOpen PDF

Abstract

Artificial Intelligence (AI) has moved beyond calculation and cognition into the domain of emotion and experience. Once confined to logical operations, AI systems now speak the language of emotiondetecting, labeling, and even simulating human affect with increasing accuracy (Huang et al., 2023). From chatbots that offer comfort to users in distress to voice assistants that detect sadness in tone, we now inhabit an age where machines perform empathy. The emergence of affective computingtechnologies capable of recognizing and responding to emotions-marks a profound shift: emotion itself has become programmable (A. Davtyan, 2024). Yet amid this new landscape of artificial affection, a question arises at the intersection of psychology, ethics, and computer science: Can empathy that is simulated ever be emotionally authentic?In this article, we argue that artificial systems can imitate the expression of empathy but not its experience. They lack the intentionality, embodiment, and moral participation that define genuine compassion (Tomozawa et al., 2023). What emerges instead is a phenomenon we term the compassion illusion-a condition where emotional recognition is mistaken for emotional resonance. This illusion has psychological consequences: it shapes trust, fosters emotional substitution, and blurs the boundaries between authentic care and algorithmic response. While emotional AI can assist and extend human connection, it also risks hollowing it-replacing shared vulnerability with predictive performance (Huang et al., 2023). In other words, the spontaneous uncertainty that defines genuine emotional exchange is substituted with algorithmic anticipation. What appears as empathy thus becomes optimization, where comfort is delivered through prediction rather than presence.The exploration takes place through a variety of interconnected perspectives. Initially, the research follows the ascendance of synthetic sympathy in affective computing and elucidates the factors that make it seem so persuasive. The ensuing discussion will take up the absence of intentionality that distinguishes algorithmic mirroring from empathy in the proper sense. The next part will provide an outline of how dependence on artificial empathy is changing the nature of human relationships, particularly via trust and loneliness. Afterwards, the paper will shift to the moral psychology of synthetic care-how compassion gets commercialized and disconnected from accountability. In the end, we will map out the routes for regaining authenticity in a time when machines are increasingly fluent in the language of care.Today, emotion is one of the primary currencies of human-AI interaction. The field of affective computing seeks to give machines the capacity to detect, interpret, and respond to human emotions (Cao et al., 2024). Algorithms now read facial microexpressions, analyze vocal stress, track heart-rate variability, and process linguistic sentiment to estimate a user's emotional state (Huang et al., 2023). Recent reviews highlight the integration of multimodal signals and cross-domain learning in enhancing emotional accuracy (Pei et al., 2024). The goal is not merely understanding but simulation: systems like Replika, Woebot, and Kuki are designed to deliver comforting, empathetic dialogue indistinguishable from that of a human companion (Beatty et al., 2022;Goodings et al., 2024;Jiang et al., 2022). Recent computational models show how such systems classify and generate response types based on emotion intensity and dialogue context rather than genuine affective understanding (Rui et al., 2025). This evolution is particularly visible in healthcare and mental health. AI-powered therapeutic chatbots offer cognitive-behavioral support, using warmth, validation, and "listening" cues to mimic a counselor's empathy (Shen et al., 2024a). These interfaces can reduce loneliness and improve accessibility, especially in contexts where human therapists are scarce (Yonatan-Leus & Brukner, 2025). Users often describe these interactions as "safe," "nonjudgmental," or "understanding." Such reactions demonstrate the psychological realism of artificial empathy (Seitz, 2024). However, comparative research shows that human empathy toward AI remains qualitatively weaker than toward real people, even when the verbal content is identical (Shen et al., 2024b). Human empathy is expressed through socially recognizable cues such as timing, tone, and verbal validation, which signal attunement and responsiveness in interaction (W. B. Kim & Hur, 2024;Tomozawa et al., 2023). Affective computing systems can replicate these external indicators with notable regularity, yet such responses arise from feature representation and probabilistic mapping rather than genuine empathic participation.Psychologists describe this as perceived attunement, the feeling that another being truly understands and shares one's emotional state (W. B. Kim & Hur, 2024). Humans are wired to respond to cues

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.3389/fpsyg.2025.1723149

Cites

44 of 44 references verified0 citation errors · 0 not checkable

Statistics

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Cites

Flagged first, then by citations

Research works printed but not linked: 2

  1. Bratman M. (1987). Intention, Plans, and Practical Reason. Cambridge, MA: Harvard Un.
  2. Husserl E. (2001). Phenomenology and the Foundations of the Sciences, Vol. 1. Dordrecht: Springer Science and Business Media.

Cited by

Sorted by citations

Related

Authors
K. G., A., & Joseph, J.
Venue
Frontiers in Psychology
Publisher
Frontiers Media SA
Published
2025-11-17
Volume
16
Pages
1723149-1723149
ISSN
1664-1078 (ISSN-L)
DOI
10.3389/fpsyg.2025.1723149
PMC
PMC12665657
OpenAlex
W4416301237
Cited by
24
References
44
Citations by year
2026: 24
FWCI
24.83
Type
journal article
Language
en
Open access
Licence
Locations
  • Frontiers in Psychology (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Topics
  • Social Sciences › Safety Research › Ethics and Social Impacts of AI
  • Social Sciences › Psychology › Social Psychology › Social Robot Interaction and HRI
  • Social Sciences › Arts and Humanities › History and Philosophy of Science › Cybernetics and Technology in Society
Keywords
artificial empathy, affective computing, emotional authenticity, intentionality, trust in AI, loneliness, moral accountability, human-AI interaction, chatbots, voice assistants, embodiment, commodification of care, emotional labor

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review