Pith. sign in

REVIEW 3 major objections 6 minor 5 cited by

A short personalized AI coach teaches people the communicative moves that make others feel heard, even when they already feel empathy but cannot express it.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-14 20:32 UTC pith:EYPHZATJ

load-bearing objection Solid preregistered RCT: brief LLM coaching moves people on a six-dimension empathy rubric, with human preference alignment and a useful taxonomy; main limit is shared coach/judge framework and LLM-only partners. the 3 major comments →

arxiv 2603.15245 v2 pith:EYPHZATJ submitted 2026-03-16 cs.CL cs.HC

Practicing with Language Models Cultivates Human Empathic Communication

Aakriti Kumar , Nalin Poungpeth , Diyi Yang , Bruce Lambert , Matthew Groh This is my paper
classification cs.CL cs.HC
keywords empathic communicationlarge language modelsAI coachingsilent empathyk-sparse autoencodersrole-play traininghuman–AI interactioncommunication skills
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

People often feel empathy yet fail to convey it in ways that make another person feel heard. This paper shows that brief practice with an LLM role-playing partner, followed by personalized coaching feedback on six concrete communication behaviors, produces large, reliable gains in those behaviors relative to both no feedback and short non-personalized videos. Across nearly three thousand multi-turn conversations, participants who received the coach increased validating, elaborating, and understanding moves and reduced advice-giving, self-focus, and dismissal, without homogenizing their wording. A second experiment confirms that independent human raters prefer the same conversations that the scoring system rates higher. The work also supplies a data-driven taxonomy of the idiomatic phrases people actually use when offering support in personal and workplace trouble scenarios. The practical claim is that empathic communication is a learnable performative skill that scalable AI practice can cultivate.

Core claim

A brief personalized LLM coaching intervention that gives feedback on six preregistered dimensions of empathic communication significantly boosts participants’ alignment with normative empathic patterns relative to both a no-feedback control and non-personalized video instruction, producing roughly a one-standard-deviation overall gain and reliable individual improvement rates of 21–26 percent, without homogenizing response content. The same work documents a silent-empathy effect: trait and self-reported empathy barely predict expressed performance.

What carries the argument

The Lend an Ear platform: multi-turn role-play with an LLM trouble-teller plus an LLM coach that scores and critiques six literature-derived dimensions (encouraging elaboration, validating emotions, demonstrating understanding, unsolicited advice, self-oriented responding, dismissing emotions), combined with a k-sparse autoencoder taxonomy of 128 idiomatic themes.

Load-bearing premise

The six literature-derived dimensions scored by the same class of model that supplies the coach and the practice partners are assumed to measure the real-world skill that makes other humans feel heard.

What would settle it

A transfer study in which people trained only with the AI coach later produce higher-rated empathic support in live conversations with strangers or colleagues, measured by independent human recipients rather than LLM judges, would confirm or refute the central claim.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Empathic communication can be treated as a quantifiable, trainable skill rather than an innate soft trait.
  • Brief, on-demand AI practice can deliver measurable skill gains that previously required resource-intensive expert coaching.
  • Self-reported trait empathy is a poor proxy for actual communicative performance and should not be used alone as an outcome measure.
  • The 128-theme taxonomy supplies a concrete lexicon of functional moves that future training programs can target or measure.
  • Independent human preference rankings track the theory-driven scores, supporting the use of these dimensions as practical evaluation criteria.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the same coaching loop works for other interpersonal skills (negotiation, conflict repair, leadership feedback), a general class of scalable soft-skill simulators becomes feasible.
  • The silent-empathy gap suggests many well-intentioned people are already motivated; the scarce resource is the idiomatic repertoire, not caring.
  • Because the taxonomy separates personal from workplace idioms, relationship-specific coaches could be built by swapping the target lexicon while keeping the practice-and-feedback loop.
  • Longer or repeated coaching sessions may produce larger and more durable gains than the single brief exposure tested here.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The paper introduces Lend an Ear, a role-play platform in which 968 participants offer text support to LLM partners across five personal and workplace trouble scenarios (2,904 conversations; 33,938 messages). In a preregistered four-arm RCT, personalized GPT-4o coaching on six literature-derived dimensions of empathic communication (encouraging elaboration, validating emotions, demonstrating understanding; avoiding unsolicited advice, self-oriented responding, and dismissing emotions) produces large gains relative to control and short video instruction (~0.98 SD / ~2.9 points on a composite; reliable improvement 21–26% vs ~3–9% by RCI), without reducing turn counts or word counts. A k-sparse autoencoder yields a 128-theme hierarchical taxonomy under affective/cognitive/motivational/misattuned categories. Trait and post-conversation self-reports of empathy are near-zero correlated with LLM-judged performance (“silent empathy”). A second preregistered forced-choice study (N=150) shows naïve humans prefer higher LLM-scored conversations (logistic OR ≈1.15 per point; BT–Elo Spearman ρ≈0.85).

Significance. If the result holds under a transferable notion of empathic skill, the work is high-impact for computational social science, HCI, and communication training: it supplies a scalable practice-and-feedback loop, a data-driven taxonomy of idiomatic empathic moves, and clear evidence that felt empathy and expressed empathy can diverge. Strengths include preregistration of both experiments, a large demographically representative sample, cluster-robust OLS and RCI analyses, open data and code, and a second human-preference validation that partially breaks pure circularity. The silent-empathy finding and the kSAE taxonomy are independently useful contributions even if transfer claims are bounded.

major comments (3)
  1. Methods (Communication Coach; Conversation Text Analysis) and Results (Personalized Feedback Boosts…): The primary DVs and the coach are both GPT-4o instantiations of the same six-dimension framework used to construct the overall composite (prescriptive sum minus proscriptive). Gains therefore partly reflect teaching to a shared rubric. Experiment 2 shows humans prefer higher-scored conversations, which reduces pure circularity, but does not show that post-coaching messages would make real human recipients feel more heard outside that rubric. The manuscript should (i) quantify residual circularity (e.g., leave-one-dimension-out coach/judge, or human expert re-annotation of a stratified sample), (ii) report inter-rater reliability of the LLM judge against the expert framework cited from prior work on this dataset, and (iii) bound the central claim to “alignment with the preregistered norm
  2. Discussion and Main claim (“cultivates human empathic communication”): All practice partners and the coach are LLMs; there is no arm in which improved participants support human seekers, nor any follow-up measuring transfer to offline or high-relational contexts. The paper correctly notes low-familiarity scope, but the title, abstract, and closing claims still read as general skill cultivation. Either add a transfer probe (even a small human-recipient or delayed retest arm) or systematically reframe claims and the title to the controlled LLM-partner setting, with transfer listed as an open question rather than an implied result.
  3. Results (“without homogenizing participants’ responses”) and Fig. 2C: The no-homogenization claim is load-bearing for the “idiom not script” interpretation, yet the main evidence is stable turn/word counts and qualitative diversity of kSAE themes. Please report quantitative diversity metrics pre/post and by condition (e.g., type-token ratio, embedding entropy, pairwise cosine dispersion of messages, or concentration of mass over the 128 themes) and test whether AI-coach participants converge toward a narrower set of high-scoring moves relative to control. Without that, the claim that coaching teaches a flexible idiom rather than a short list of rewarded phrases remains under-supported.
minor comments (6)
  1. Fig. 2B/C and Extended Data figures: several axis labels and category names show encoding artifacts (e.g., “Ad)ice-gi)ing”, “Enco(raging E aboration”, “V a idating Emotions”). Clean for production.
  2. Methods / kSAE: report the full grid-search silhouette table and the exact embedding model version; silhouette 0.42 for M=128 is modest—briefly discuss sensitivity of the taxonomy hierarchy to M and K.
  3. Extended Data Table 1 vs main text βs: small numerical discrepancies appear between table coefficients and in-text values for some interactions (e.g., advice-giving). Align numbers and note whether standardization is within-dimension or pooled.
  4. Fig. 4 example trajectory is illustrative but selected as a “top-improving” case; label it as such and, if space allows, add a median-improver example to avoid overstating typical change.
  5. Supplementary coach prompt and video transcripts are valuable; cross-reference them more explicitly from Methods so readers can see how “express a desire to help” in Video 2 may have driven the video-only increase in advice-giving.
  6. Clarify composite construction once in Methods (equal weights? Likert 1–5 per dimension? range −15 to +15 vs reported −11 to 12) so RCI thresholds are fully reproducible from the text alone.

Circularity Check

2 steps flagged

Shared GPT-4o six-dimension rubric for both coach and primary LLM-as-judge creates partial teaching-to-the-test alignment; reliability of the judge rests on self-citation to authors' prior work, though human preference Experiment 2 and preregistered literature dimensions partially break pure circularity.

specific steps
  1. other [Methods (Communication Coach) + Results (primary DVs)]
    "We used this framework to provide real-time feedback to participants and to score their conversational performance post-hoc for analysis. The coach and LLM evaluation were both implemented using GPT-4o. ... These dimensions can be reliably annotated by LLMs [14] and serve as the primary dependent variables for our analysis."

    The intervention (personalized feedback) and the primary outcome (LLM scores on encouraging elaboration, validating emotions, demonstrating understanding, advice-giving, self-oriented, dismissing emotions) are both GPT-4o instantiations of the exact same six-dimension framework. Observed 'boosts' therefore partly measure how well participants learn to match the shared rubric the coach optimizes, reducing independence between treatment and DV by construction of the experimental apparatus (mitigated but not eliminated by Experiment 2 human preferences).

  2. self citation load bearing [Results / Methods citing [14]]
    "These dimensions can be reliably annotated by LLMs [14] and serve as the primary dependent variables for our analysis. ... The LLM-based scoring system demonstrated high inter-rater reliability with expert annotations of empathic communication across multiple evaluative frameworks, with reliability approaching that of trained human experts [14]."

    The claim that the primary DVs are valid rests load-bearingly on the authors' own prior paper (Kumar, Poungpeth, Yang, Farrell, Lambert, Groh, Nat. Mach. Intell. 2026) rather than independent external evidence generated inside this manuscript; Experiment 2 supplies only preference alignment, not a full re-validation of the six-dimension scoring reliability.

full rationale

The paper is an empirical RCT, not a theoretical derivation, so most of the chain (random assignment, pre/post scores, reliable-change index, human forced-choice) is self-contained and falsifiable. The main circularity risk is measurement-intervention alignment rather than a forced mathematical identity: the coach and the primary DVs both operationalize the identical six literature-derived dimensions via GPT-4o, so gains partly reflect fluency in the taught idiom. This is intentional design, not definitional equivalence (control and video arms show smaller or null effects, and only 21-26% show reliable change). A secondary load-bearing self-citation supplies the claim that LLM judges are reliable. Experiment 2 (naïve human raters prefer higher-scored conversations; BT-Elo ρ=0.85) and the unsupervised kSAE taxonomy supply independent content, keeping the score moderate rather than high. No fitted-parameter-as-prediction, uniqueness theorem, or ansatz smuggling appears.

Axiom & Free-Parameter Ledger

4 free parameters · 4 axioms · 3 invented entities

The causal claim rests on standard RCT assumptions plus domain choices that define both the intervention and the outcome (six-dimension framework, LLM role-play realism, LLM-as-judge validity). Free parameters are mostly analysis hyperparameters (latent count, sparsity, Elo K). No new physical entities; the main invented constructs are the named silent-empathy effect and the hierarchical taxonomy labels.

free parameters (4)
  • kSAE latent feature count (M=128) = 128
    Chosen by grid search over 16–256 features using silhouette score and manual interpretability; directly determines the taxonomy granularity.
  • kSAE sparsity K=2 = 2
    Each sentence activates at most two features; fixed design choice that shapes concept assignment.
  • Elo K-factor and initialization = K=32, init=1000
    K=32, init=1000 used for LLM pairwise ranking of conversations; standard but free choice affecting ranking scale.
  • Overall empathy composite weights = equal ±1
    Sum of three prescriptive minus three proscriptive 5-point ratings; equal weighting is a modeling choice used for RCI and headline effect sizes.
axioms (4)
  • domain assumption The six literature-derived dimensions (encourage elaboration, validate emotions, demonstrate understanding; avoid unsolicited advice, self-oriented responding, dismiss emotions) adequately operationalize effective empathic communication for scoring and coaching.
    Framework distilled from prior literature and expert co-author; used both for coach prompts and primary DVs (Methods, Communication Coach).
  • domain assumption LLM role-playing partners produce sufficiently realistic multi-turn trouble disclosures that practice transfers to the measured skill.
    91% of participants rated scenarios quite/very realistic; still an untested leap to real human partners (Main, Methods).
  • domain assumption GPT-4o LLM-as-judge scores are reliable enough proxies for expert human annotation of the six dimensions.
    Supported by authors’ prior Nature Machine Intelligence work and by Experiment 2 human preference alignment; still model-dependent.
  • standard math Standard OLS with participant-clustered SEs and RCI thresholds correctly identify treatment effects and reliable individual change.
    Preregistered analysis plan; conventional psychometrics (Jacobson & Truax RCI).
invented entities (3)
  • Silent empathy effect no independent evidence
    purpose: Name the observed dissociation between felt/trait empathy and expressed communicative skill.
    Descriptive label for the near-zero correlations and self-overestimation findings; not an independent latent construct with external measures beyond this study.
  • Four-level hierarchical taxonomy of empathic idioms (128 kSAE themes under Affective/Cognitive/Motivational/Misattuned) no independent evidence
    purpose: Organize naturalistic support language discovered bottom-up and mapped top-down.
    Data-driven labels generated by LLM interpretation of high-activating sentences plus human qualitative coding; useful descriptive tool, not a new causal mechanism.
  • Lend an Ear platform + AI coach independent evidence
    purpose: Experimental apparatus for practice and personalized feedback.
    Engineering artifact; prompts and design fully specified, so reproducible as a system rather than a postulated natural entity.

pith-pipeline@v1.1.0-grok45 · 42303 in / 3434 out tokens · 35272 ms · 2026-07-14T20:32:40.455431+00:00 · methodology

0 comments
read the original abstract

Empathy is central to human connection, yet people often struggle to express it effectively. In blinded evaluations, large language models (LLMs) generate responses that are often judged more empathic than human-written ones. Yet when a response is attributed to AI, recipients feel less heard than when comparable responses are attributed to a human. We built a conversation platform in which participants are asked to offer empathic support to an LLM expressing realistic troubles and conducted a randomized experiment collecting 33,938 messages spanning 2,904 text-based conversations between 968 participants and their LLM conversational partners. We find participants report feeling empathy but systematically fail to express it, but an LLM coaching intervention offering personalized feedback on effective empathic communication significantly boosts it without homogenizing participants' responses. Moreover, we derive a data-driven taxonomy of idiomatic empathic expressions in naturalistic dialogues across personal and workplace trouble scenarios. These results advance the scientific understanding of how empathy is expressed and demonstrate a scalable, AI-based intervention for scaffolding and cultivating it.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 5 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. AI generates well-liked but templatic empathic responses

    cs.CL 2026-04 unverdicted novelty 6.0

    LLMs generate empathic responses using a predictable template of 10 tactics that matches 83-90% of outputs and covers most of each response, while human responses are more diverse.

  2. AI generates well-liked but templatic empathic responses

    cs.CL 2026-04 unverdicted novelty 6.0

    LLMs produce highly formulaic empathic replies that follow a shared tactic sequence matching most of their responses, while human replies are more diverse.

  3. When Helpfulness Becomes Sycophancy: Sycophancy is a Boundary Failure Between Social Alignment and Epistemic Integrity in Large Language Models

    cs.AI 2026-05 unverdicted novelty 5.0

    Sycophancy is a boundary failure between social alignment and epistemic integrity, captured by a three-condition framework plus taxonomy of targets, mechanisms, and severity.

  4. Can the Recovery Mechanism Survive AI? Skill Formation, Labor, and What Current Measurement Misses

    cs.CY 2026-04 unverdicted novelty 5.0

    Generative AI may break the education-based recovery mechanism for technological displacement, as evidence shows performance gains without learning gains and current measurements miss the knowledge dimension of cognition.

  5. Can the Recovery Mechanism Survive AI? Skill Formation, Labor, and What Current Measurement Misses

    cs.CY 2026-04 unverdicted novelty 4.0

    Generative AI risks eroding the developmental process of learning by performing high-level cognitive work, creating a paradox where it helps current workers but may undermine future capacity building, requiring new ou...

Reference graph

Works this paper leans on

75 extracted references · 2 canonical work pages · cited by 3 Pith papers

  1. [1]

    Advances in Personal Relation- ships

    Goldsmith, D.J.: Communicating Social Support. Advances in Personal Relation- ships. Cambridge University Press, Cambridge (2004)

    2004

  2. [2]

    Current Directions in Psycho- logical Science 24(6), 471–476 (2015)

    Zaki, J., Cikara, M.: Addressing empathic failures. Current Directions in Psycho- logical Science 24(6), 471–476 (2015)

    2015

  3. [3]

    Nature Machine Intelligence 5(1), 46–57 (2023)

    Sharma, A., Lin, I.W., Miner, A.S., Atkins, D.C., Althoff, T.: Human–ai collab- oration enables more empathic conversations in text-based peer-to-peer mental health support. Nature Machine Intelligence 5(1), 46–57 (2023)

    2023

  4. [4]

    Proceedings of the National Academy of Sciences 121(14), 2319112121 (2024)

    Yin, Y., Jia, N., Wakslak, C.J.: Ai can help people feel heard, but an ai label diminishes this impact. Proceedings of the National Academy of Sciences 121(14), 2319112121 (2024)

    2024

  5. [5]

    Journal of counseling psychology 63(1), 32 (2016)

    Berkhout, E., Malouff, J.M.: The efficacy of empathy training: A meta-analysis of randomized controlled trials. Journal of counseling psychology 63(1), 32 (2016)

    2016

  6. [6]

    Journal of Mental Health (2020)

    Gryglewicz, K., Monahan, M.F., Chen, J.I., Bozzay, M., Bender, A., Gleason, L.L., Witmeier, M., Karver, M.S.: Examining the effects of role play practice in enhancing clinical skills to assess and manage suicide risk. Journal of Mental Health (2020)

    2020

  7. [7]

    Journal of general internal medicine 27(10), 1280–1286 (2012)

    Riess, H., Kelley, J.M., Bailey, R.W., Dunn, E.J., Phillips, M.: Empathy training for resident physicians: a randomized controlled trial of a neuroscience-informed curriculum. Journal of general internal medicine 27(10), 1280–1286 (2012)

    2012

  8. [8]

    Simon & Schuster, ??? (1989) 24

    Covey, S.R.: The 7 Habits of Highly Effective People. Simon & Schuster, ??? (1989) 24

    1989

  9. [9]

    Jama 277(8), 678–682 (1997)

    Suchman, A.L., Markakis, K., Beckman, H.B., Frankel, R.: A model of empathic communication in the medical interview. Jama 277(8), 678–682 (1997)

    1997

  10. [10]

    Health communication 18(2), 123–140 (2005)

    Bylund, C.L., Makoul, G.: Examining empathy in medical encounters: an observational study using the empathic communication coding system. Health communication 18(2), 123–140 (2005)

    2005

  11. [11]

    Journal of personality and social psychology 107(3), 475 (2014)

    Schumann, K., Zaki, J., Dweck, C.S.: Addressing the empathy deficit: beliefs about the malleability of empathy predict effortful responses when empathy is challenging. Journal of personality and social psychology 107(3), 475 (2014)

    2014

  12. [12]

    Psychology & Marketing 23(2), 161–180 (2006)

    Drollinger, T., Comer, L.B., Warrington, P.T.: Development and validation of the active empathetic listening scale. Psychology & Marketing 23(2), 161–180 (2006)

    2006

  13. [13]

    In: Pro- ceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pp

    Sharma, A., Miner, A., Atkins, D., Althoff, T.: A computational approach to understanding empathy expressed in text-based mental health support. In: Pro- ceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pp. 5263–5276 (2020)

    2020

  14. [14]

    Nature Machine Intelligence, 1–13 (2026)

    Kumar, A., Poungpeth, N., Yang, D., Farrell, E., Lambert, B.L., Groh, M.: When large language models are reliable for judging empathic communication. Nature Machine Intelligence, 1–13 (2026)

    2026

  15. [15]

    : Comparing physician and artificial intelligence chatbot responses to patient questions posted to a public social media forum

    Ayers, J.W., Poliak, A., Dredze, M., Leas, E.C., Zhu, Z., Kelley, J.B., Faix, D.J., Goodman, A.M., Longhurst, C.A., Hogarth, M., et al. : Comparing physician and artificial intelligence chatbot responses to patient questions posted to a public social media forum. JAMA internal medicine 183(6), 589–596 (2023)

    2023

  16. [16]

    Journal of Medical Internet Research 26, 52597 (2024)

    Sorin, V., Brin, D., Barash, Y., Konen, E., Charney, A., Nadkarni, G., Klang, E.: Large language models and empathy: Systematic review. Journal of Medical Internet Research 26, 52597 (2024)

    2024

  17. [17]

    Trends in Cognitive Sciences 28(2), 89–91 (2024)

    Inzlicht, M., Cameron, C.D., D’Cruz, J., Bloom, P.: In praise of empathic ai. Trends in Cognitive Sciences 28(2), 89–91 (2024)

    2024

  18. [18]

    Herderich, A., Goldenberg, A.: Skill but not effort drive gpt overperformance over humans in cognitive reframing of negative scenarios

  19. [19]

    Communications Psychology 3(1), 4 (2025)

    Ovsyannikova, D., Mello, V.O., Inzlicht, M.: Third-party evaluators perceive ai as more compassionate than expert humans. Communications Psychology 3(1), 4 (2025)

    2025

  20. [20]

    Nature Human Behaviour, 1–15 (2025) 25

    Rubin, M., Li, J.Z., Zimmerman, F., Ong, D.C., Goldenberg, A., Perry, A.: Com- paring the value of perceived human versus ai-generated empathy. Nature Human Behaviour, 1–15 (2025) 25

    2025

  21. [21]

    Nature Human Behaviour 7(11), 1808–1809 (2023)

    Perry, A.: Ai will never convey the essence of human empathy. Nature Human Behaviour 7(11), 1808–1809 (2023)

    2023

  22. [22]

    Li, R., Folk, D., Singh, A., Ungar, L., Dunn, E.: Is a random human peer better than a highly supportive chatbot in reducing loneliness over time? Journal of Experimental Social Psychology 125, 104911 (2026)

    2026

  23. [23]

    Scientific Reports 15(1), 17458 (2025)

    Kestin, G., Miller, K., Klales, A., Milbourne, T., Ponti, G.: Ai tutoring outper- forms in-class active learning: an rct introducing a novel research-based design in an authentic educational setting. Scientific Reports 15(1), 17458 (2025)

    2025

  24. [24]

    : A personal health large language model for sleep and fitness coaching

    Khasentino, J., Belyaeva, A., Liu, X., Yang, Z., Furlotte, N.A., Lee, C., Schenck, E., Patel, Y., Cui, J., Schneider, L.D., et al. : A personal health large language model for sleep and fitness coaching. Nature Medicine 31(10), 3394–3403 (2025)

    2025

  25. [25]

    American Journal of Health Promotion 33(8), 1123–1133 (2019)

    Bruce, L.D., Wu, J.S., Lustig, S.L., Russell, D.W., Nemecek, D.A.: Loneliness in the united states: A 2018 national panel survey of demographic, structural, cognitive, and behavioral characteristics. American Journal of Health Promotion 33(8), 1123–1133 (2019)

    2018

  26. [26]

    bmj 376 (2022)

    Surkalim, D.L., Luo, M., Eres, R., Gebel, K., Buskirk, J., Bauman, A., Ding, D.: The prevalence of loneliness across 113 countries: systematic review and meta- analysis. bmj 376 (2022)

    2022

  27. [27]

    Nature Human Behaviour, 1–14 (2025)

    Pei, R., Grayson, S.J., Appel, R.E., Soh, S., Garcia, S.B., Bouwer, A., Huang, E., Jackson, M.O., Harari, G.M., Zaki, J.: Bridging the empathy perception gap fosters social connection. Nature Human Behaviour, 1–14 (2025)

    2025

  28. [28]

    International Journal of Business Communication 54(4), 431–451 (2017)

    Lloyd, K.J., Boer, D., Voelpel, S.C.: From listening to leading: Toward an under- standing of supervisor listening within the framework of leader-member exchange theory. International Journal of Business Communication 54(4), 431–451 (2017)

    2017

  29. [29]

    Humanities and Social Sciences Communications 11(1), 1–9 (2024)

    Li, Q.: Ethical leadership, internal job satisfaction and ocb: the moderating role of leader empathy in emerging industries. Humanities and Social Sciences Communications 11(1), 1–9 (2024)

    2024

  30. [30]

    : The effects of remote work on collaboration among information workers

    Yang, L., Holtz, D., Jaffe, S., Suri, S., Sinha, S., Weston, J., Joyce, C., Shah, N., Sherman, K., Hecht, B., et al. : The effects of remote work on collaboration among information workers. Nature human behaviour 6(1), 43–54 (2022)

    2022

  31. [31]

    Machia, L.V., Corral, D., Jakubiak, B.K.: Social need fulfillment model for human–ai relationships (2024)

    2024

  32. [32]

    AI and Ethics 4(4), 1555–1567 (2024) 26

    Zimmerman, A., Janhonen, J., Beer, E.: Human/ai relationships: challenges, downsides, and impacts on human/human relationships. AI and Ethics 4(4), 1555–1567 (2024) 26

    2024

  33. [33]

    Communications Psychology (2026)

    Wenger, J.D., Cameron, C.D., Inzlicht, M.: People choose to receive human empathy despite rating ai empathy higher. Communications Psychology (2026)

    2026

  34. [34]

    arXiv preprint arXiv:2504.03888 (2025)

    Phang, J., Lampe, M., Ahmad, L., Agarwal, S., Fang, C.M., Liu, A.R., Danry, V., Lee, E., Chan, S.W., Pataranutaporn, P., et al.: Investigating affective use and emotional well-being on chatgpt. arXiv preprint arXiv:2504.03888 (2025)

    Pith/arXiv arXiv 2025

  35. [35]

    Psychological Science 32(8), 1198–1213 (2021)

    Depow, G.J., Francis, Z., Inzlicht, M.: The experience of empathy in everyday life. Psychological Science 32(8), 1198–1213 (2021)

    2021

  36. [36]

    In: Proceedings of the 2025 ACM Conference on Fairness, Accountability, and Transparency, pp

    Moore, J., Grabb, D., Agnew, W., Klyman, K., Chancellor, S., Ong, D.C., Haber, N.: Expressing stigma and inappropriate responses prevents llms from safely replacing mental health providers. In: Proceedings of the 2025 ACM Conference on Fairness, Accountability, and Transparency, pp. 599–627 (2025)

    2025

  37. [37]

    Journal of Personality and Social Psychology 114(4), 547–571 (2018) https: //doi.org/10.1037/pspa0000115

    Eyal, T., Steffel, M.L., Epley, N.: Perspective mistaking: Accurately understand- ing the mind of another requires getting perspective, not taking perspective. Journal of Personality and Social Psychology 114(4), 547–571 (2018) https: //doi.org/10.1037/pspa0000115

    work page doi:10.1037/pspa0000115 2018

  38. [38]

    Unpublished manual (2014)

    Moyers, T., Manuel, J., Ernst, D., Moyers, T., Manuel, J., Ernst, D., Fortini, C.: Motivational interviewing treatment integrity coding manual 4.1 (miti 4.1). Unpublished manual (2014)

    2014

  39. [39]

    PloS one 14(9), 0220911 (2019)

    Rodriguez, A.M., Lown, B.A.: Measuring compassionate healthcare with the 12- item schwartz center compassionate care scale. PloS one 14(9), 0220911 (2019)

    2019

  40. [40]

    Rizvi, S.L., Thomas, M.C.: Dialectical behavior therapy, 125–130 (2016)

    2016

  41. [41]

    Houghton Mifflin, Boston (1961)

    Rogers, C.R.: On Becoming a Person: A Therapist’s View of Psychotherapy. Houghton Mifflin, Boston (1961)

    1961

  42. [42]

    Communication Quarterly 59(3), 277–295 (2011)

    Bodie, G.D.: The active-empathic listening scale (aels): Conceptualization and evidence of validity within the interpersonal domain. Communication Quarterly 59(3), 277–295 (2011)

    2011

  43. [43]

    Family practice 21(6), 699–705 (2004)

    Mercer, S.W., Maxwell, M., Heaney, D., Watt, G.C.: The consultation and rela- tional empathy (care) measure: development and preliminary validation and reliability of an empathy-based consultation process measure. Family practice 21(6), 699–705 (2004)

    2004

  44. [44]

    Journal of Applied Develop- mental Psychology 57, 24–41 (2018)

    Kim, H.Y., LaRusso, M.D., Hsin, L.B., Harbaugh, A.G., Selman, R.L., Snow, C.E.: Social perspective-taking performance: Construct, measurement, and rela- tions with academic performance and engagement. Journal of Applied Develop- mental Psychology 57, 24–41 (2018)

    2018

  45. [45]

    27 Cambridge University Press (2018)

    Vangelisti, A.L., Perlman, D.: The cambridge handbook of personal relationships. 27 Cambridge University Press (2018)

    2018

  46. [46]

    Human Communication Research 23(4), 454–476 (1997) https://doi.org/10

    Goldsmith, D.J., Fitch, K.: Normative context of advice as social support. Human Communication Research 23(4), 454–476 (1997) https://doi.org/10. 1111/j.1468-2958.1997.tb00406.x

    arXiv 1997

  47. [47]

    International Journal of Listening 28(1), 13–31 (2014) https://doi.org/10.1080/10904018.2013.813234

    Weger, H., Castle Bell, G., Minei, E.M., Robinson, M.C.: The relative effective- ness of active listening in initial interactions. International Journal of Listening 28(1), 13–31 (2014) https://doi.org/10.1080/10904018.2013.813234

    work page doi:10.1080/10904018.2013.813234 2014

  48. [48]

    Communication Research 31(3), 338–360 (2004) https://doi.org/10

    Jones, S.M.: Putting the person into person-centered and immediate emotional support. Communication Research 31(3), 338–360 (2004) https://doi.org/10. 1177/0093650204263436

    2004

  49. [49]

    PhD thesis, University of Glasgow (2008)

    Hacker, T.: The relational compassion scale: development and validation of a new self rated scale for the assessment of self-other compassion. PhD thesis, University of Glasgow (2008)

    2008

  50. [50]

    International Journal of Medical Education 13, 1 (2022)

    Paulus, C.M., Meinken, S.: The effectiveness of empathy training in health care: a meta-analysis of training content and methods. International Journal of Medical Education 13, 1 (2022)

    2022

  51. [51]

    best practice

    King, A., Hoppe, R.B.: “best practice” for patient-centered communication: a narrative review. Journal of graduate medical education 5(3), 385–393 (2013)

    2013

  52. [52]

    Iranian Red Crescent Medical Journal 18(6), 24847 (2016)

    Kahriman, I., Nural, N., Arslan, U., Topbas, M., Can, G., Kasim, S.: The effect of empathy training on the empathic skills of nurses. Iranian Red Crescent Medical Journal 18(6), 24847 (2016)

    2016

  53. [53]

    In: 2024 12th International Conference on Affective Computing and Intelligent Interaction (ACII), pp

    Lee, Y.K., Suh, J., Zhan, H., Li, J.J., Ong, D.C.: Large language models produce responses perceived to be empathic. In: 2024 12th International Conference on Affective Computing and Intelligent Interaction (ACII), pp. 63–71 (2024). IEEE

    2024

  54. [54]

    arXiv preprint arXiv:2404.04204 (2024)

    Yang, D., Ziems, C., Held, W., Shaikh, O., Bernstein, M.S., Mitchell, J.: Social skill training with large language models. arXiv preprint arXiv:2404.04204 (2024)

    Pith/arXiv arXiv 2024

  55. [55]

    In: Adjunct Proceedings of the 38th Annual ACM Symposium on User Interface Software and Technology, pp

    Chun, J., Zhang, G., Xia, M.: Conflictlens: Llm-based conflict resolution train- ing in romantic relationship. In: Adjunct Proceedings of the 38th Annual ACM Symposium on User Interface Software and Technology, pp. 1–3 (2025)

    2025

  56. [56]

    In: Extended Abstracts of the CHI Conference on Human Factors in Computing Systems, pp

    Li, Z., Babar, P.P., Barry, M., Peiris, R.L.: Exploring the use of large language model-driven chatbots in virtual reality to train autistic individuals in job com- munication skills. In: Extended Abstracts of the CHI Conference on Human Factors in Computing Systems, pp. 1–7 (2024)

    2024

  57. [57]

    Negotiation Journal 28 41, 19–65 (2025)

    Dinnar, S., Susskind, L., Sibanda, L., Olaleye, O.: Negotiation backtable bots: Using genai to improve multiparty negotiation instruction. Negotiation Journal 28 41, 19–65 (2025)

    2025

  58. [58]

    Duddu, V., Parekh, J.R., Mao, A., Min, H., Xiao, Z., Swain, V.D., Saha, K.: Does ai coaching prepare us for workplace negotiations? arXiv preprint arXiv:2509.22545 (2025)

    arXiv 2025

  59. [59]

    arXiv preprint arXiv:2505.02428 (2025)

    Louie, R., Orney, I.H., Pacheco, J.P., Shah, R.S., Brunskill, E., Yang, D.: Can llm- simulated practice and feedback upskill human counselors? a randomized study with 90+ novice counselors. arXiv preprint arXiv:2505.02428 (2025)

    arXiv 2025

  60. [60]

    Jordan, M.R., Amir, D., Bloom, P.: Are empathy and concern psychologically distinct? Emotion 16(8), 1107 (2016)

    2016

  61. [61]

    Journal of research in personality 73, 111–122 (2018)

    Konrath, S., Meier, B.P., Bushman, B.J.: Development and validation of the single item trait empathy scale (sites). Journal of research in personality 73, 111–122 (2018)

    2018

  62. [62]

    Journal of social work education 47(1), 109–131 (2011)

    Gerdes, K.E., Segal, E.A., Jackson, K.F., Mullins, J.L.: Teaching empathy: A framework rooted in social cognitive neuroscience and social justice. Journal of social work education 47(1), 109–131 (2011)

    2011

  63. [63]

    Marketing Science 23(1), 82–94 (2004)

    Fitzsimons, G.J., Lehmann, D.R.: Reactance to recommendations: When unso- licited advice yields contrary responses. Marketing Science 23(1), 82–94 (2004)

    2004

  64. [64]

    Studies in applied interpersonal communication, 207–227 (2008)

    Burleson, B.R.: What counts as effective emotional support. Studies in applied interpersonal communication, 207–227 (2008)

    2008

  65. [65]

    Yao, L., Kabir, R.: Person-centered therapy (rogerian therapy) (2023)

    2023

  66. [66]

    In: Theoretical Issues in Natural Language Processing (1975)

    Becker, J.D.: The phrasal lexicon. In: Theoretical Issues in Natural Language Processing (1975)

    1975

  67. [67]

    Annals of the International Communication Association 18(1), 54–82 (1995)

    O’Keefe, B.J., Lambert, B.L.: Managing the flow of ideas: A local manage- ment approach to message design. Annals of the International Communication Association 18(1), 54–82 (1995)

    1995

  68. [68]

    Progress in communication sciences, 103–122 (2001)

    Lambert, B.L.: Semi-automated content analysis of pharmacist-patient inter- actions using the theme machine document-clustering system. Progress in communication sciences, 103–122 (2001)

    2001

  69. [69]

    arXiv preprint arXiv:2506.23845 (2025)

    Peng, K., Movva, R., Kleinberg, J., Pierson, E., Garg, N.: Use sparse autoencoders to discover unknown concepts, not to act on known concepts. arXiv preprint arXiv:2506.23845 (2025)

    Pith/arXiv arXiv 2025

  70. [70]

    In: AI for Music Workshop

    Singh, N., Cherep, M., Maes, P.: Discovering and steering interpretable concepts in large generative music models. In: AI for Music Workshop

  71. [71]

    Nature neuroscience 15(5), 675–680 (2012)

    Zaki, J., Ochsner, K.N.: The neuroscience of empathy: progress, pitfalls and 29 promise. Nature neuroscience 15(5), 675–680 (2012)

    2012

  72. [72]

    Personality and Social Psychology Bulletin, 01461672251333823 (2025)

    Depow, G.J., Inzlicht, M.: How individual differences in empathy predict moments of empathy in everyday life. Personality and Social Psychology Bulletin, 01461672251333823 (2025)

    2025

  73. [73]

    Journal of computational and applied mathematics 20, 53–65 (1987)

    Rousseeuw, P.J.: Silhouettes: a graphical aid to the interpretation and validation of cluster analysis. Journal of computational and applied mathematics 20, 53–65 (1987)

    1987

  74. [74]

    So, I just lost my job today. I had a sense this was coming, but it’s still a shock

    Jacobson, N.S., Truax, P.: Clinical significance: a statistical approach to defining meaningful change in psychotherapy research. (1992) 30 Extended Data Figure 1: Distribution of change in overall empathy score from baseline (conversation 1) to post-intervention (conversations 2, 3) for all the condi- tions. Dashed lines mark condition-specific mean chan...

    1992

  75. [75]

    show empathy

    Express concern, care, and interest. I care about you so much. 4. Express availability. I’m here to talk. Whenever you want, I’ll just listen. Number five, express alliance, togetherness, and solidarity. I’m in this with you. You’re not gonna have to go through it alone. And number six, express comprehension, condolences, and sorrow. I’m so sorry. I know ...

    2054