Pith. sign in

REVIEW 3 major objections 5 minor 65 references

Embodied Empathy: A Multimodal AR and LLM-Powered System for Self-Attachment Psychotherapy with Self-Initiated Humour

T0 review · 3 major / 5 minor · reviewed 2026-08-04 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read This paper claims that an AR and LLM-powered mobile app can deliver SAT/SIHP self-attachment therapy, with an eight-day, 16-user study showing feasibility, mood improvements, and design trade-offs in emotion mirroring.

desk verdict A genuinely new integrated AR/LLM/avatar system for SAT/SIHP with a useful 16-person feasibility study, but the abstract's mood-improvement claim outruns the reported evidence and needs pulling back to feasibility and experience mapping. read the letter →

arxiv 2608.02283 v1 pith:XME543TC submitted 2026-08-03 cs.HC cs.MM

Xinyan Ye , Gwyneth Phang , Anandha Gopalan , Abbas Edalat This is my paper · ORCID
classification cs.HCcs.MM
keywords digitalpsychotherapyself-attachmenttechniqueself-initiatedhumourLLMchatbotsemotionmirroringaugmentedrealityavatarsaffectivecomputing
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper reports on building a mobile therapy app that combines a personalised 3D childhood avatar, augmented reality, and a large-language-model "therapist" to deliver two related self-guided protocols: Self-Attachment Technique (creating an affectionate bond with one's inner child) and Self-Initiated Humour Protocol (learning non-hostile laughter). An eight-day study with 16 non-clinical users found the system feasible and acceptable, with 87.5% of participants endorsing it as a tool for emotional self-regulation and with self-reported mood improvements. The central insight is that embodiment matters: personalised avatars anchor emotional bonding, spoken (TTS) responses convey containment and empathy, but automatic emotion mirroring is a double-edged sword that only helps when emotion classification is correct and animation intensity is calibrated. The paper also finds that users want the AI to be a proactive guide rather than a reactive chatbot.

What carries the argument

The load-bearing mechanism is "embodied empathy," achieved through a pipeline: user text is sent to a cloud-hosted emotion-recognition model (a fine-tuned RoBERTa classifier reported at 94.96% accuracy on its own evaluation); the predicted emotion triggers FACS-based blendshape and skeletal animations on the child avatar, while the therapist avatar shows a regulated, calm response; the LLM maintains the SAT/SIHP dialogue; and text-to-speech lets the therapist speak in a voice matched to the avatar. AR projects the child avatar into the physical room for short, on-demand episodes. The paper's results assign distinct roles to each modality: avatars anchor attachment, TTS drives perceived empat

What would settle it

A controlled experiment comparing sessions with emotion mirroring enabled versus disabled (with avatar presence and TTS held constant) would isolate whether mirroring itself drives the reported mood and engagement gains; if the mirroring-ON condition does not outperform mirroring-OFF on self-reported mood and empathy, the paper's central explanation for embodied empathy collapses.

Watch

Extended reading notes

Core claim

The paper's central claim is that an integrated multimodal platform—showing the user's childhood self as an AR-embeddable avatar, driving a therapist avatar with an LLM, and automatically translating user text into avatar facial expressions—can deliver SAT/SIHP in a way that feels engaging and emotionally containing. The eight-day user study is offered as evidence that this delivery is feasible and accepted: participants reported positive shifts in self-reported mood, high engagement, and strong bonding with personalised avatars, with text-to-speech output specifically strengthening perceived empathy. A key design finding is that emotion mirroring increases engagement when accurate but harms

Load-bearing premise

The automatic emotion-recognition model must correctly classify users' real-time text and drive avatar animations accurately; if classification fails or animations are misattuned, the emotion-mirroring effect that underpins embodied empathy reverses and can break the therapeutic experience.

Editorial extensions

If this is right

  • SAT/SIHP can be delivered through a single smartphone app rather than separate VR and web components, making the protocol more accessible.
  • Personalised childhood avatars are the primary anchor for self-attachment; future designs should prioritise them over generic placeholders.
  • Emotion mirroring should be built with controllable intensity, mixed-emotion support, and user override, because misclassifications can undermine therapeutic safety.
  • Voice output (TTS) is a low-cost, high-benefit feature for perceived empathy, while voice input (STT) yields marginal and inconsistent returns.
  • Users' expectation shift implies that LLM-based therapy agents should be designed as proactive facilitators with state tracking, not reactive responders.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the 'proactive facilitator' expectation holds beyond SAT/SIHP, LLM mental-health tools generally may need explicit session-management and goal-tracking capabilities rather than open-ended chat.
  • A natural next experiment is to isolate emotion mirroring in a controlled trial (mirroring ON vs OFF, holding avatars and TTS constant) to test whether the reported mood benefits are truly caused by mirroring.
  • The avatar-persistence friction (share codes, recreation) is likely more than a UX bug: for attachment-based therapy, losing access to the childhood avatar may directly weaken the bond, so secure on-device caching could be a clinically meaningful feature.
  • Since the paper explicitly states the LLM has not been clinically validated for hallucination and toxicity, feasibility for non-clinical users does not yet establish safety in distress or crisis contexts.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper presents an iOS-based multimodal system that delivers the Self-Attachment Technique (SAT) and Self-Initiated Humour Protocol (SIHP) through an LLM-driven therapist chatbot, a personalized 3D childhood avatar, augmented reality, speech input/output, and automated emotion mirroring. The authors report an eight-day, N=16 remote user study with post-study questionnaires (30 Likert items, 2 mood-impact multiple-choice items, 4 open-ended questions). The findings are framed as evidence of feasibility and acceptability, with qualitative and descriptive quantitative support for the value of personalized avatars, AR, TTS, and emotion mirroring, as well as identified design trade-offs. The paper also reports a shift in user expectations toward proactive conversational facilitation. The strongest claims in the abstract and contribution list, however, include 'improvements in self-reported mood' and 'TTS is a more potent driver of perceived empathy,' which go beyond the descriptive evidence actually reported.

Significance. If the central claims were fully supported, the paper would be a useful design contribution to the emerging area of embodied, LLM-powered digital mental health. The integrated system is novel in combining SAT/SIHP with avatar personalization, AR, multimodal interaction, and automated affective mirroring, and the study surfaces concrete, plausible design trade-offs (e.g., TTS over STT, the fragility of emotion mirroring, the need for avatar persistence). The paper is transparent about many of its limitations in Section 6 and Appendix B, which strengthens the reader's ability to interpret the results. Its main value at this stage is as an exploratory feasibility and design-space mapping study, not as evidence of mood improvement or comparative efficacy. The quantitative data are limited to descriptive means/SDs and the qualitative quotations are illustrative, so significance rests on the plausibility and generalizability of the design insights rather than on hypothesis testing.

major comments (3)
  1. [Abstract; Contribution (2); §4.2–§5; §6] The abstract and Contribution (2) assert 'improvements in self-reported mood' and 'positive impact on user self-reported mood,' but the reported evaluation contains no such evidence. §4.2 describes only 30 Likert items about app features, 2 multiple-choice questions on perceived mood impact, and 4 open questions. Tables 1–3 and Figures 4–7 report means and SDs for feature-related composites; no pre-post mood score, no baseline, no effect size, confidence interval, or inferential statistic appears anywhere. The single quantitative endorsement (87.5%) is not tied to an item wording or response distribution. §6 concedes that the measures captured subjective experience rather than clinical endpoints and that the contribution is design-space mapping. As written, the strongest wording of the central claim is unsupported by the evidence presented. This is a missing-evidence problem, not a small
  2. [Tables 2–3, §5.2–§5.3] The comparative claims about modality trade-offs are based on descriptive aggregate means with no inferential statistics and no experimental manipulation. For example, Table 2 shows TTS Usefulness M=64.1 vs. Voice Input Usefulness M=54.7, with SDs of 27.3 and 37.9; Figure 6 shows considerable overlap in distributions. No significance test, effect size, or confidence interval is reported, and the study had no baseline/control condition. The assertion in Contribution (3) that 'TTS is a more potent driver of perceived empathy than STT,' and the related statement in §5.3 that TTS 'supports expression indirectly' while STT 'produced a mixed picture,' are not supported at the level of comparative claims. Qualitative quotes are valuable, but they should be presented as exploratory evidence rather than as 'empirical insights' with comparative force. Please add appropriate statistical caution or
  3. [§4.1, §5.4, §6] The eight-day protocol condenses an eight-week SAT/SIHP course one day per week-equivalent, yet the paper uses the results to draw conclusions about 'therapeutic engagement' and 'strengthen emotional bonding.' Because all participants experienced all modalities in a fixed order (Days 1–2 SAT, Days 3–8 SIHP), there is no way to disentangle modality effects from order effects, novelty effects, or the natural trajectory of engagement. Section 6 acknowledges the absence of long-term adherence data and fine-grained interaction logs, which is commendable, but the abstract and several discussion passages (e.g., §5.1: 'custom child avatars are central to fostering an emotional connection') do not consistently carry this caveat. The stated contribution is feasibility, and the paper should consistently frame these findings as initial experiential evidence from a compressed, uncontrolled deployment
minor comments (5)
  1. [Appendix B, Table 4] §4.2 says the questionnaire comprised 30 rating statements, but Table 4 lists only 23 analytical labels under the three quantitative dimensions (9 Therapeutic UX, 8 Avatar, 6 Multi-Modal). Clarify whether the remaining items are the two multiple-choice items and/or open-ended questions, or provide the full mapping.
  2. [Figures 5 and 7] The radar charts and distribution plots lack clear axis labeling and a legend explaining the 0–100 normalized scale. Please add axis labels, tick values, and a note that these are aggregated Likert scores, not raw percentages.
  3. [Table 3 note] The 'Technical Reliability' item is reverse-scored, but the reverse-coding procedure is not described. State how the response scale was reversed before aggregation and whether all items were on the same 1–5 Likert scale.
  4. [§3.2, §5.2] The emotion-recognition accuracy (94.96%, macro-F1 95.10%) is cited from the authors' prior work [3], but no in-situ accuracy on the current app's real-time, in-the-wild user text is reported. Given the participants' reports of misattuned or exaggerated reactions, please add a sentence acknowledging this transfer limitation or report a small validation of the model on the collected user text.
  5. [§5.1] The phrase 'clustered in the upper range (50–70/100)' is vague; specify which items are included and what 'upper range' means relative to a neutral midpoint of 50 on the normalized scale.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the central claims are empirical findings from the study's own data; overlapping-author citations supply component provenance, not the results.

full rationale

This paper contains no formal derivation chain whose outputs could reduce to its inputs. Its central claims—feasibility, acceptability, modality trade-offs, and perceived empathy/bonding—are supported by the reported eight-day N=16 study data (§4–5, Tables 1–3, Figures 4–7) rather than by an equation, a fitted parameter, or a definition. The only overlapping-author citations are used descriptively: [3] to report the emotion-recognition model's prior accuracy (94.96%), and [19–23] to supply the SAT/SIHP therapeutic rationale and prior pilot results. These are not used to derive the current participants' responses, nor does the paper rename a fitted value as a prediction. The paper also explicitly limits its contribution in §6: 'Our contribution should therefore be understood as mapping a design space for embodied, multimodal SAT/SIHP delivery and characterising early user experience, rather than demonstrating clinical efficacy.' The abstract's stronger wording about 'improvements in self-reported mood' is not backed by a reported pre/post comparison, but that is a missing-evidence or overstatement concern, not a circularity. Under the rule that only specific reductions count as circular, no circular step is present.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

No free parameters in the mathematical sense. The central claims rest on domain assumptions about therapy efficacy, emotion-classifier transfer, self-report validity, and animation sufficiency; the authors' own qualitative results challenge the last two.

assumptions (4)
  • domain assumption SAT/SIHP is an effective therapeutic protocol whose exercises can be meaningfully delivered by a digital agent.
    The intervention presupposes the clinical basis of SAT/SIHP, citing only the authors' own prior work [19-23]; no independent clinical validation is provided here. §2.1.
  • domain assumption The RoBERTa emotion classifier from [3], with reported 94.96% accuracy, generalizes to real-time in-the-wild user text and correctly drives mirroring.
    Emotion-mirroring claims depend on this; the model is cited from the same group's prior work and is not evaluated on this app's data. §3.2, §5.2.
  • domain assumption Self-reported, non-standardized Likert measures of mood and empathy capture the constructs of interest.
    Outcomes are post-study self-report; no validated scales or interaction logs. §4.2, §6.
  • domain assumption FACS-based predefined animations can represent the user's emotional state sufficiently to create embodied empathy.
    Animation from static emotion configurations is assumed adequate; participants reported misfires. §3.2, §5.2.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Embodied Empathy: A Multimodal AR and LLM-Powered System for Self-Attachment Psychotherapy with Self-Initiated Humour." pith.science (2026). https://pith.science/paper/XME543TC

@misc{pith2026260802283,
  author       = {Pith},
  title        = {Pith review of: Embodied Empathy: A Multimodal AR and LLM-Powered System for Self-Attachment Psychotherapy with Self-Initiated Humour},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XME543TC}},
  note         = {Machine review of arXiv:2608.02283}
}
read the original abstract

The growing global demand for mental health support increasingly exceeds the supply of qualified practitioners, creating an urgent need for scalable digital interventions that can deliver meaningful emotional connection. In response, we present a novel multimodal application that operationalises the Self-Initiated Humour Protocol (SIHP) within a Self-Attachment Technique (SAT) framework. Our mobile application integrates customisable 3D childhood avatars, augmented reality, and an LLM-driven virtual therapist capable of automated emotion mirroring. An eight-day user study (N=16) indicates the system's feasibility and improvements in self-reported mood. Results show that personalised avatars and text-to-speech output strengthen emotional bonding and perceived empathy. Although emotion mirroring boosts engagement, its effectiveness depends heavily on classification accuracy and animation intensity. Moreover, findings indicate a shift in user expectations--from reactive chatbots to proactive conversational facilitators. We conclude with design implications for leveraging AI and AR to cultivate embodied empathy in digital mental health tools.

Figures

Figures reproduced from arXiv: 2608.02283 by the authors.

Figure 1
Figure 1. Key features and interfaces of the mobile application: (a) Avatar Creation and Selection, allowing users to generate a [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Flowchart of user interaction journey from onboard [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Automated emotion mirroring within the Virtual [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Avatar Mean Rating Radar Chart (𝑁 = 16) across eight metrics in the Avatar Interaction dimension [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Chatbot & Multimodal Radar Chart (𝑁 = 16) across five metrics in the Multimodal Chatbot dimension [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Distribution of participant ratings for the Multi [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 8
Figure 8. Figure 8: Screenshots of the avatar creation phase of the [PITH_FULL_IMAGE:figures/full_fig_p010_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

65 extracted references · 7 canonical work pages

  1. [1]

    S. B. Abhinaya. 2024. Enabling Developers, Protecting Users: Investigating Harassment and Safety in VR.OSF Preprints(2024). doi:10.17605/OSF.IO/C7FKS Accessed: 2026-01-27

    work page doi:10.17605/osf.io/c7fks 2024

  2. [2]

    Hazza Bin Ahmed and Mohammad Alzuobi. 2025. Designing Accessible Virtual Reality Interfaces Using Reinforcement Learning for Users with Motor and Sen- sory Impairments.PatternIQ Mining(24 February 2025). doi:10.70023/sahd/250201

    work page doi:10.70023/sahd/250201 2025

  3. [3]

    Lisa Alazraki, Ali Ghachem, Neophytos Polydorou, Foaad Khosmood, and Abbas Edalat. 2021. An empathetic AI coach for self-attachment therapy. In2021 ieee third international conference on cognitive machine intelligence (cogmi). IEEE, 78–87

    2021

  4. [4]

    Apple Inc. 2026. ARKit. https://developer.apple.com/documentation/arkit. Apple Developer Documentation. Accessed: 2026-01-27

    2026

  5. [5]

    Avatar SDK. 2025. MetaPerson Loader for Unity. https://github.com/avatarsdk/ metaperson-loader-unity. Accessed: 2025-09-03

    2025

  6. [6]

    Hillel Aviezer, Ran R Hassin, Jennifer Ryan, Cheryl Grady, Josh Susskind, Adam Anderson, Morris Moscovitch, and Shlomo Bentin. 2008. Angry, disgusted, or afraid? Studies on the malleability of emotion perception.Psychological science 19, 7 (2008), 724–732

    2008

  7. [7]

    Azuma, Y

    R. Azuma, Y. Baillot, R. Behringer, S. Feiner, S. Julier, and B. MacIntyre. 2001. Recent advances in augmented reality.IEEE Computer Graphics and Applications 21, 6 (2001), 34–47. doi:10.1109/38.963459

    work page doi:10.1109/38.963459 2001

  8. [8]

    Oliver Baus and Stéphane Bouchard. 2014. Moving from virtual reality exposure- based therapy to augmented reality exposure-based therapy: a review.Frontiers in human neuroscience8 (2014), 112

    2014

Show all 65 references
  1. [9]

    Christian Becker-Asano and Ipke Wachsmuth. 2008. Affect simulation with primary and secondary emotions. InInternational workshop on intelligent virtual agents. Springer, 15–28

    2008

  2. [10]

    Arthur C Bohart and Leslie S Greenberg. 1997. Empathy and psychotherapy: An introductory overview. (1997)

    1997

  3. [11]

    Calm. 2026. Experience Calm. https://www.calm.com. Accessed: 2026-01-26

    2026

  4. [12]

    Character.ai. 2026. character.ai | AI Chat, Reimagined. https://character.ai/. Accessed: 2026-01-26

    2026

  5. [13]

    Minh Duc Chu, Patrick Gerard, Kshitij Pawar, Charles Bickham, and Kristina Lerman. 2025. Illusions of Intimacy: Emotional Attachment and Emerging Psy- chological Risks in Human-AI Relationships. arXiv:2505.11649 [cs.SI] https: //arxiv.org/abs/2505.11649

    2025

  6. [14]

    Cloudflare. 2025. Cloudflare D1: Managed Serverless SQLite Database. https: //developers.cloudflare.com/d1/. Cloudflare Docs. Accessed: 2026-01-28

    2025

  7. [15]

    Xueting Cui, Yun Gu, Hui Fang, and Tingshao Zhu. 2025. Development and evaluation of LLM-based suicide intervention chatbot.Frontiers in Psychiatry16 (August 2025), 1634714. doi:10.3389/fpsyt.2025.1634714

    2025

  8. [16]

    Nele Dael, Marcello Mortillaro, and Klaus R Scherer. 2012. Emotion expression in body action and posture.Emotion12, 5 (2012), 1085

    2012

  9. [17]

    Ulf Dimberg, Per Andréasson, and Monika Thunberg. 2011. Emotional empathy and facial reactions to facial expressions.Journal of Psychophysiology(2011)

    2011

  10. [18]

    Mrudul Dixit, Rajiya Landage, and Prachi Raut. 2026. Comparative Analysis of Deep Learning Models for Speech-to-Text and Text-to-Speech Conversion. In Smart Trends in Computing and Communications, Tomonobu Senjyu, Chakchai So-In, and Amit Joshi (Eds.). Springer Nature, 53–62. ...

    2026 doi

  11. [19]

    Abbas Edalat. 2015. Introduction to self-attachment and its neural basis. In2015 International Joint Conference on Neural Networks (IJCNN). IEEE, 1–8. doi:10. 1109/IJCNN.2015.7280780

    2015

  12. [20]

    Abbas Edalat. 2017. Self-attachment: A holistic approach to computational psychiatry. InComputational neurology and psychiatry. Springer, 273–314

    2017

  13. [21]

    Abbas Edalat. 2022. Self-Initiated Humour Protocols: An Algorithmic Approach for Learning to Laugh. https://www.doc.ic.ac.uk/~ae/papers/SIHP.pdf. Accessed: 2025-06-01

    2022

  14. [22]

    Abbas Edalat, Ruoyu Hu, Zeena Patel, Neophytos Polydorou, Frank Ryan, and Dasha Nicholls. 2025. Self-initiated humour protocol: a pilot study with an AI agent.Frontiers in Digital Health7 (13 March 2025). doi:10.3389/fdgth.2025. 1530131

    2025 doi

  15. [23]

    Abbas Edalat, Neophytos Polydorou, Frank Ryan, Barnabas J Gilbert, and Dasha Nicholls. 2024. Affectional interactions with your animated child avatar can enhance wellbeing.Frontiers in Virtual Reality5 (2024), 1424217

    2024

  16. [24]

    Paul Ekman. 1992. Are there basic emotions? (1992)

    1992

  17. [25]

    1978.Facial Action Coding System

    Paul Ekman and Wallace V Friesen. 1978.Facial Action Coding System. Consulting Psychologists Press

    1978

  18. [26]

    Anna Felnhofer, Franziska Pfannerstill, Lisa Gänsler, Oswald D Kothgassner, Elke Humer, Johanna Büttner, and Thomas Probst. 2025. Barriers to adopting therapeu- tic virtual reality: the perspective of clinical psychologists and psychotherapists. Frontiers in Psychiatry16 (2025...

    2025

  19. [27]

    Riyadh Firdaus, Aida Rosita Tantri, and Sidharta Kusuma Manggala. 2024. Factors influencing virtual reality sickness in emergency simulation training.Medical Science Educator34, 6 (2024), 1309–1315

    2024

  20. [28]

    Kathleen Kara Fitzpatrick, Alison Darcy, and Molly Vierhile. 2017. Delivering Cog- nitive Behavior Therapy (CBT) to Young Adults With Symptoms of Depression and Anxiety Using a Fully Automated Conversational Agent (Woebot): A Random- ized Controlled Trial.JMIR Mental Health4, ...

    2017 doi

  21. [29]

    Google Cloud. 2026. Speech-to-Text AI: Speech Recognition and Transcription. https://cloud.google.com/speech-to-text. Accessed: 2026-01-27

    2026

  22. [30]

    Google Cloud. 2026. Text-to-Speech AI: Lifelike Speech Synthesis. https://cloud. google.com/text-to-speech. Accessed: 2026-01-27

    2026

  23. [31]

    Google DeepMind. 2023. Gemini: A Family of Highly Capable Multimodal Models. arXiv preprint arXiv:2312.11805(2023)

    2023 arXiv

  24. [32]

    Martijn Goudbeek. 2020. Cultural Differences in the Nonverbal Expressions of Emotion. (2020)

    2020

  25. [33]

    Judith Hartfill, Shirin Hajahmadi, Susanne Schmidt, Gustavo Marfia, and Frank Steinicke. 2025. Embracing differences in virtual reality: inclusive user-centered design of bimanual interaction techniques.Frontiers in Virtual Reality6 (2025), 1586875

    2025

  26. [34]

    Hendrikse, Sem Kluiver, Jan Treur, Tom F

    Sophie C.F. Hendrikse, Sem Kluiver, Jan Treur, Tom F. Wilderjans, Suzanne Dikker, and Sander L. Koole. 2023. How virtual agents can learn to synchronize: An adaptive joint decision-making model of psychotherapy.Cognitive Systems Research79 (2023), 138–155. doi:10.1016/j.cogsys...

    2023 doi

  27. [35]

    Sakif Hossain and Jörg P Müller. 2025. Exploring Empathic Human-Robot Collab- oration. In2025 11th International Conference on Control, Automation and Robotics (ICCAR). IEEE, 346–355

    2025

  28. [36]

    Beyond laughter

    Eshika Kafle, Cat Papastavrou Brooks, Dave Chawner, Una Foye, Dieter Declercq, and Helen Brooks. 2023. “Beyond laughter”: a systematic review to understand how interventions utilise comedy for individuals experiencing mental health problems.Frontiers in Psychology14 (2023), 11...

    2023 doi

  29. [37]

    Antje Sien Lisanne Knol, Mike Huiskes, Tom Koole, Reitske Meganck, Tom Loeys, and Mattias Desmet. 2020. Reformulating and mirroring in psychotherapy: a conversation analytic perspective.Frontiers in Psychology11 (2020), 318

    2020

  30. [38]

    Gordon T Kraft-Todd, Diego A Reinero, John M Kelley, Andrea S Heberlein, Lee Baer, and Helen Riess. 2017. Empathic nonverbal behavior increases ratings of both warmth and competence in a medical context.PloS one12, 5 (2017), e0177758

    2017

  31. [39]

    Sze Chit Leong, Yuk Ming Tang, Chung Hin Lai, and CKM Lee. 2023. Facial expression and body gesture emotion recognition: A systematic review on the use of visual data in affective computing.Computer science review48 (2023), 100545

    2023

  32. [40]

    Shi Min Lim, Chyi Wey Claudine Shiau, Ling Jie Cheng, and Ying Lau. 2022. Chatbot-delivered psychotherapy for adults with depressive and anxiety symp- toms: a systematic review and meta-regression.Behavior Therapy53, 2 (2022), 334–347. doi:10.1016/j.beth.2021.09.007

    2022 doi

  33. [41]

    Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, and Veselin Stoyanov. 2019. Roberta: A robustly optimized bert pretraining approach.arXiv preprint arXiv:1907.11692 (2019)

    2019 arXiv

  34. [42]

    Marlene M. Maheu. 2025. AI Psychotherapy Shutdown: What Woebot’s Exit Signals for Clinicians. https://telehealth.org/news/ai-psychotherapy-shutdown- what-woebots-exit-signals-for-clinicians/. Telehealth.org. Accessed: 2026-01-26

    2025

  35. [43]

    Mariano, Donna Spruijt-Metz, and Frederick Muench

    Timothy Y. Mariano, Donna Spruijt-Metz, and Frederick Muench. 2021. A Ther- apeutic Relational Agent for Reducing Problematic Substance Use (Woebot): Development and Usability Study.Journal of Medical Internet Research23, 3 (2021), e24850. doi:10.2196/24850 Accessed: 2026-01-26

    2021 doi

  36. [44]

    Mario Mikulincer and Phillip R Shaver. 2012. An attachment perspective on psychopathology.World Psychiatry11, 1 (2012)

    2012

  37. [45]

    MobiHealthNews. 2026. Woebot Health is shutting down its app. https://www. mobihealthnews.com/news/woebot-health-shutting-down-its-app. Accessed: 2026-01-26

    2026

  38. [46]

    Kayley Moylan and Kevin Doherty. 2025. Expert and Interdisciplinary Analysis of AI-Driven Chatbots for Mental Health Support: Mixed Methods Study.Journal of Medical Internet Research27 (2025)

    2025

  39. [47]

    OpenAI. 2023. ChatGPT. https://chat.openai.com/. Accessed: September 9, 2025

    2023

  40. [48]

    Seung Guk Park and Kyung Hye Park. 2018. Correlation between nonverbal communication and objective structured clinical examination score in medical students.Korean journal of medical education30, 3 (2018), 199

    2018

  41. [49]

    Replika. 2026. Replika: Always here to listen and talk. https://replika.com. Accessed: 2026-01-26

    2026

  42. [50]

    Till Scholich, Maya Barr, Shannon Wiltsey Stirman, and Shriti Raj. 2025. A Comparison of Responses from Human Therapists and Large Language Model- Based Chatbots to Assess Therapeutic Communication: Mixed Methods Study. JMIR Mental Health12 (May 2025), e69709. doi:10.2196/6970...

    2025 doi

  43. [51]

    Chirag Shah. 2024. From Prompt Engineering to Prompt Science With Human in the Loop. arXiv:2401.04122 [cs.HC] https://arxiv.org/abs/2401.04122

    2024 arXiv

  44. [52]

    Nair, and M

    Surabhi Sudhan, Parvathy P. Nair, and M. G. Thushara. 2024. Text-to-Speech and Speech-to-Text Models: A Systematic Examination of Diverse Approaches. InProceedings of the 2024 IEEE 9th International Conference for Convergence in Technology (I2CT). 1–8. doi:10.1109/I2CT61223.20...

    2024

  45. [53]

    Rahma M Tolba, Taha Al-Arif, and El-Sayed M El Horbaty. 2018. Realistic facial animation review: Based on facial action coding system.Egyptian Computer Science Journal42, 1 (2018), 1–9

    2018

  46. [54]

    Anthony Trioux, Wei Zhang, Giuseppe Valenzise, and Fuzheng Yang. 2025. Ex- ploring Compression Strategies for Blendshape-Based Avatar Facial Animation: Subjective and Objective Analysis. InIEEE International Conference on Multimedia and Expo

    2025

  47. [55]

    Unity Technologies. 2025. Unity. https://unity.com/. Accessed: 2025-09-10

    2025

  48. [56]

    Unity Technologies. 2026. AR Foundation Manual (Version 6.4.1). https://docs. unity3d.com/Packages/com.unity.xr.arfoundation@6.4/manual/index.html. Ac- cessed: 2026-01-27

    2026

  49. [57]

    Kirsten van Kessel, Sarita de Pont, Chiara Gasteiger, and Sonja Goedeke. 2024. Clients’ experiences of online therapy in the early stages of a COVID-19 world: A scoping review.Counselling and Psychotherapy Research24, 1 (2024), 27–38

    2024

  50. [58]

    Yinzhou Wang, Yimeng Wang, Ye Xiao, Liabette Escamilla, Bianca Augustine, Kelly Crace, Gang Zhou, and Yixuan Zhang. 2025. Evaluating an LLM-Powered Chatbot for Cognitive Restructuring: Insights from Mental Health Professionals. arXiv preprint arXiv:2501.15599. https://arxiv.or...

    2025 arXiv

  51. [59]

    Jeanne C Watson. 2016. The role of empathy in psychotherapy: Theory, research, and practice. (2016)

    2016

  52. [60]

    Woebot Health. 2026. Woebot Health: Pioneering chat-based AI wellness solu- tions. https://woebothealth.com/. Accessed: 2026-01-26

    2026

  53. [61]

    World Health Organization. 2022. Mental Disorders. https://www.who.int/news- room/fact-sheets/detail/mental-disorders Accessed: 2025-07-07

    2022

  54. [62]

    World Health Organization. 2022. Mental Health: Strengthening Our Re- sponse. https://www.who.int/news-room/fact-sheets/detail/mental-health- strengthening-our-response Accessed: 2025-07-07

    2022

  55. [63]

    Wysa. 2026. Wysa - Everyday Mental Health. https://www.wysa.com/. Accessed: 2026-01-26

    2026

  56. [64]

    Andy Wai Kan Yeung, Anela Tosevska, Elisabeth Klager, Fabian Eibensteiner, Daniel Laxar, Jivko Stoyanov, Marija Glisic, Sebastian Zeiner, Stefan Tino Kulnik, Rik Crutzen, et al. 2021. Virtual and augmented reality applications in medicine: analysis of the scientific literature...

    2021

  57. [65]

    Technical Reliability

    Youper AI. 2026. Youper: Artificial Intelligence for Mental Health. https://www. youper.ai/. Accessed: 2026-01-26. A Avatar Creation UI (a) Homepage (b) Create Avatar (c) Customise Avatar (d) Save Avatar Figure 8: Screenshots of the avatar creation phase of the application, wh...

    2026

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.