pith. sign in

arxiv: 2605.00440 · v1 · submitted 2026-05-01 · 💻 cs.AI · cs.CL· cs.HC

On the Role of Artificial Intelligence in Human-Machine Symbiosis

Ching-Chun Chang , Yuchen Guo , Hanrui Wang , Timo Spinde , Isao Echizen This is my paper

Pith reviewed 2026-05-09 19:41 UTC · model grok-4.3

classification 💻 cs.AI cs.CLcs.HC
keywords artificial intelligencehuman-machine symbiosisrole tracingnatural language generationassistive agentcreative agentAI ethicstext generation
0
0 comments X

The pith

The functional role of AI in text generation can be recovered from the output text alone.

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

This paper develops a methodology for tracing the functional role assumed by AI during natural language generation, specifically whether it acted as an assistive editor of human content or as a creative generator from a concept. The key challenge addressed is that this role, specified in the input prompt, becomes unobservable once the generated text is detached from the dialogue context. By inferring the role, embedding it into the probabilistic generation, and then recovering it from the text, the approach enables determination of how AI participated. Experimental validation in representative scenarios demonstrates effective role discrimination, robustness to perturbations, and maintenance of output quality, which matters for ethical oversight in human-AI collaboration.

Core claim

The paper claims that it is possible to infer the latent functional role of AI from the prompt, embed this role into the content during probabilistic text generation, and subsequently recover the nature of AI participation from the resulting text without access to the original prompt, as shown through experiments distinguishing assistive editing from creative generation.

What carries the argument

The proposed methodology that infers the AI's functional role from the prompt, embeds it during generation, and recovers it from detached text.

If this is right

  • The methodology can discriminate between different AI roles in content creation.
  • It remains robust against perturbations to the generated text.
  • The linguistic quality of the generated content is preserved.
  • It supports ethical assessments of whether AI has been used fairly, transparently, and appropriately.

Where Pith is reading between the lines

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

  • The tracing approach could extend to additional AI roles such as summarization or translation in collaborative writing.
  • It may guide the creation of systems that make participation levels explicit for users and auditors.
  • Applications could include auditing AI contributions in mixed human-AI creative workflows across domains.

Load-bearing premise

The functional role specified in the prompt can be reliably inferred, embedded into the probabilistic generation process, and recovered from the generated text without the original context or degrading quality.

What would settle it

A test generating multiple texts under different specified roles where the recovery method fails to distinguish them accurately above chance levels or where text quality drops measurably.

Figures

Figures reproduced from arXiv: 2605.00440 by Ching-Chun Chang, Hanrui Wang, Isao Echizen, Timo Spinde, Yuchen Guo.

Figure 1
Figure 1. Figure 1: An essay on a topic may be written by a human or generated by view at source ↗
Figure 2
Figure 2. Figure 2: Statistics of concept and content lengths across datasets. view at source ↗
Figure 3
Figure 3. Figure 3: Ternary classification accuracy across models. view at source ↗
Figure 4
Figure 4. Figure 4: Distribution of baseline features across models. view at source ↗
Figure 5
Figure 5. Figure 5: Distribution of role-specific p-values across models. larger values indicate stronger evidence of machine generation. While the human and creative samples are often reasonably separable, the assistive samples still overlap substantially with both. The data-driven detector yields scores that are highly concentrated near the two extremes, yet still provides only limited separation of the assistive samples. T… view at source ↗
Figure 6
Figure 6. Figure 6: Evaluation of robustness against synonym substitution across models. view at source ↗
read the original abstract

The evolution of artificial intelligence (AI) has rendered the boundary between humanity and computational machinery increasingly ambiguous. In the presence of more interwoven relationships within human-machine symbiosis, the very notion of AI-generated information becomes difficult to define, as such information arises not from either humans or machines in isolation, but from their mutual shaping. Therefore, a more pertinent question lies not merely in whether AI has participated, but in how it has participated. In general, the role assumed by AI is often specified, either implicitly or explicitly, in the input prompt, yet becomes less apparent or altogether unobservable when the generated content alone is available. Once detached from the dialogue context, the functional role may no longer be traceable. This study considers the problem of tracing the functional role played by AI in natural language generation. A methodology is proposed to infer the latent role specified by the prompt, embed this role into the content during the probabilistic generation process and subsequently recover the nature of AI participation from the resulting text. Experimentation is conducted under a representative scenario in which AI acts either as an assistive agent that edits human-written content or as a creative agent that generates new content from a brief concept. The experimental results support the validity of the proposed methodology in terms of discrimination between roles, robustness against perturbations and preservation of linguistic quality. We envision that this study may contribute to future research on the ethics of AI with regard to whether AI has been used fairly, transparently and appropriately.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper proposes a methodology to trace the functional role of AI in natural language generation within human-machine symbiosis. It infers the role specified in the prompt, embeds it into the probabilistic generation process, and recovers the role from the detached output text alone. Experiments are described in a scenario with AI acting as an assistive editor of human content versus a creative generator from a brief concept; the authors claim these results support role discrimination, robustness to perturbations, and preservation of linguistic quality, with implications for AI ethics and transparency.

Significance. If the central claim holds under controlled conditions, the work could provide a practical approach to assessing how AI participates in content creation, supporting ethical evaluations of transparency and appropriateness. The distinction between whether AI was used and how it was used addresses a timely gap. However, the preliminary and under-specified nature of the experiments limits the assessed significance at present.

major comments (2)
  1. [Abstract and experimental description] The abstract and experimental description assert support for discrimination, robustness, and quality preservation but supply no details on design, metrics, baselines, data, or statistical tests. Without these, the central empirical claim cannot be evaluated and appears under-supported.
  2. [Methodology and experiments] The two scenarios begin from qualitatively different inputs (full human-written text to edit versus brief concept for generation). Role recovery from detached text may therefore succeed by detecting content-type signals (e.g., residual human phrasing or topic breadth) rather than any role-specific embedding induced during generation. A content-matched control condition is required to substantiate that the functional role is reliably traceable without context.
minor comments (2)
  1. The manuscript would benefit from a diagram or pseudocode clarifying the inference-embedding-recovery pipeline.
  2. The discussion of linguistic quality preservation would be strengthened by explicit metrics or human evaluation protocols.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We are grateful to the referee for their insightful comments, which have helped us identify key improvements for the manuscript. We address each major comment below and indicate the revisions we will make.

read point-by-point responses

  1. Referee: [Abstract and experimental description] The abstract and experimental description assert support for discrimination, robustness, and quality preservation but supply no details on design, metrics, baselines, data, or statistical tests. Without these, the central empirical claim cannot be evaluated and appears under-supported.

    Authors: We agree that the current manuscript does not provide sufficient details on the experimental design, metrics, baselines, data, or statistical tests, making it difficult to fully evaluate the central claims. This was an oversight in the presentation. In the revised version, we will include a new section detailing the experimental methodology, including the specific models used for generation and role embedding, the metrics for measuring role discrimination accuracy, robustness to perturbations, and linguistic quality, the datasets employed, and the statistical tests applied. We will also revise the abstract to reference these supporting results more precisely. revision: yes

  2. Referee: [Methodology and experiments] The two scenarios begin from qualitatively different inputs (full human-written text to edit versus brief concept for generation). Role recovery from detached text may therefore succeed by detecting content-type signals (e.g., residual human phrasing or topic breadth) rather than any role-specific embedding induced during generation. A content-matched control condition is required to substantiate that the functional role is reliably traceable without context.

    Authors: The referee correctly identifies a potential confound in the experimental design. The differing input types could allow the role recovery to rely on content differences rather than the embedded functional role. We do not have a content-matched control in the current experiments. To address this, we will add such a control condition in the revised manuscript, for example by having the creative generation start from expanded concepts that match the length and style of human-written texts used in the assistive scenario, and vice versa. This will allow us to better isolate the contribution of the role-specific embedding. We will also include an analysis showing that the recovery performance drops when role embedding is removed, supporting that the signal is role-induced. revision: yes

Circularity Check

0 steps flagged

No circularity: methodology is an independent empirical proposal

full rationale

The paper presents a methodological framework for inferring, embedding, and recovering AI functional roles in NLG without any equations, derivations, or parameter-fitting steps that reduce outputs to inputs by construction. The central claims rest on an experimental design contrasting assistive editing versus creative generation scenarios, with reported results on discrimination, robustness, and quality treated as empirical validation rather than tautological consequences of prior definitions or self-citations. No load-bearing premises invoke uniqueness theorems, ansatzes from the authors' prior work, or renaming of known patterns; the approach is self-contained against external benchmarks and does not rely on self-referential chains.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on domain assumptions about probabilistic text generation allowing role embedding and recovery; no free parameters or invented entities are identifiable from the abstract.

axioms (1)
  • domain assumption AI's functional role is specified in the input prompt and can be made traceable in the generated output through the probabilistic generation process.
    This assumption underpins the entire methodology for inferring and recovering the role from text alone.

pith-pipeline@v0.9.0 · 5573 in / 1284 out tokens · 36548 ms · 2026-05-09T19:41:39.006334+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

50 extracted references · 50 canonical work pages

  1. [1]

    Computing machinery and intelligence,

    A. M. Turing, “Computing machinery and intelligence,”Mind, vol. 59, no. 236, pp. 433–460, 1950

    work page 1950

  2. [2]

    Presentation of a maze solving machine,

    C. E. Shannon, “Presentation of a maze solving machine,” inProc. Josiah Macy Jr. Found. Conf. Cybern., New York, NY , USA, 1951, pp. 173–180

    work page 1951

  3. [3]

    Deep learning,

    Y . LeCun, Y . Bengio, and G. E. Hinton, “Deep learning,”Nature, vol. 521, no. 7553, pp. 436–444, 2015

    work page 2015

  4. [4]

    Human-level control through deep reinforcement learn- ing,

    V . Mnihet al., “Human-level control through deep reinforcement learn- ing,”Nature, vol. 518, no. 7540, pp. 529–533, 2015

    work page 2015

  5. [5]

    Recurrent world models facilitate policy evolution,

    D. Ha and J. Schmidhuber, “Recurrent world models facilitate policy evolution,” inProc. Int. Conf. Neural Inf. Process. Syst. (NeurIPS), vol. 31, Montréal, QC, Canada, 2018, pp. 2455–2467

    work page 2018

  6. [6]

    On the morality of artificial agents,

    L. Floridi and J. W. Sanders, “On the morality of artificial agents,”Minds Mach., vol. 14, no. 3, pp. 349–379, 2004

    work page 2004

  7. [7]

    Research priorities for robust and beneficial artificial intelligence,

    S. Russell, D. Dewey, and M. Tegmark, “Research priorities for robust and beneficial artificial intelligence,”AI Mag., vol. 36, no. 4, pp. 105– 114, 2015

    work page 2015

  8. [8]

    The malicious use of artificial intelligence: Fore- casting, prevention, and mitigation,

    M. Brundageet al., “The malicious use of artificial intelligence: Fore- casting, prevention, and mitigation,” Apollo – University of Cambridge Repository, Tech. Rep., 2018

    work page 2018

  9. [9]

    AI4People–An ethical framework for a good AI society: Opportunities, risks, principles, and recommendations,

    L. Floridiet al., “AI4People–An ethical framework for a good AI society: Opportunities, risks, principles, and recommendations,”Minds Mach., vol. 28, no. 4, pp. 689–707, 2018

    work page 2018

  10. [10]

    On the dangers of stochastic parrots: Can language models be too big?

    E. M. Bender, T. Gebru, A. McMillan-Major, and S. Shmitchell, “On the dangers of stochastic parrots: Can language models be too big?” in Proc. ACM Conf. Fairness Account. Transpar. (FAccT), Virtual Event, Canada, 2021, pp. 610–623

    work page 2021

  11. [11]

    Taxonomy of risks posed by language models,

    L. Weidingeret al., “Taxonomy of risks posed by language models,” in Proc. ACM Conf. Fairness Account. Transpar. (FAccT), Seoul, Korea, 2022, pp. 214–229

    work page 2022

  12. [12]

    On faithfulness and factuality in abstractive summarization,

    J. Maynez, S. Narayan, B. Bohnet, and R. McDonald, “On faithfulness and factuality in abstractive summarization,” inProc. Annu. Meet. Assoc. Comput. Linguist. (ACL), Virtual Event, 2020, pp. 1906–1919

    work page 2020

  13. [13]

    Survey of hallucination in natural language generation,

    Z. Jiet al., “Survey of hallucination in natural language generation,” ACM Comput. Surv., vol. 55, no. 12, pp. 1–38, 2023

    work page 2023

  14. [14]

    Large legal fictions: Profiling legal hallucinations in large language models,

    M. Dahl, V . Magesh, M. Suzgun, and D. E. Ho, “Large legal fictions: Profiling legal hallucinations in large language models,”J. Leg. Anal., vol. 16, no. 1, pp. 64–93, 2024

    work page 2024

  15. [15]

    Siren’s song in the AI ocean: A survey on hallucination in large language models,

    Y . Zhanget al., “Siren’s song in the AI ocean: A survey on hallucination in large language models,”Comput. Linguist., vol. 51, no. 4, pp. 1373– 1418, 2025

    work page 2025

  16. [16]

    RealToxicityPrompts: Evaluating neural toxic degeneration in language models,

    S. Gehman, S. Gururangan, M. Sap, Y . Choi, and N. A. Smith, “RealToxicityPrompts: Evaluating neural toxic degeneration in language models,” inProc. Conf. Empir. Methods Nat. Lang. Process. (EMNLP), Virtual Event, 2020, pp. 3356–3369

    work page 2020

  17. [17]

    Aligning AI with shared human values,

    D. Hendryckset al., “Aligning AI with shared human values,” inProc. Int. Conf. Learn. Represent. (ICLR), Virtual Event, 2021, pp. 1–29

    work page 2021

  18. [18]

    Challenges in detoxifying language models,

    J. Welblet al., “Challenges in detoxifying language models,” inProc. Conf. Empir. Methods Nat. Lang. Process. (EMNLP), Punta Cana, Dominican Republic, 2021, pp. 2447–2469

    work page 2021

  19. [19]

    Jailbroken: How does LLM safety training fail?

    A. Wei, N. Haghtalab, and J. Steinhardt, “Jailbroken: How does LLM safety training fail?” inProc. Int. Conf. Neural Inf. Process. Syst. (NeurIPS), vol. 36, no. 80079–80110, New Orleans, LA, USA, 2023

    work page 2023

  20. [20]

    Many-shot jailbreaking,

    C. Anilet al., “Many-shot jailbreaking,” inProc. Int. Conf. Neural Inf. Process. Syst. (NeurIPS), vol. 37, Vancouver, BC, Canada, 2024, pp. 129 696–129 742

    work page 2024

  21. [21]

    O’Neil,Weapons of Math Destruction: How Big Data Increases Inequality and Threatens Democracy

    C. O’Neil,Weapons of Math Destruction: How Big Data Increases Inequality and Threatens Democracy. New York, NY , USA: Crown Publishing Group, 2016

    work page 2016

  22. [22]

    Man is to computer programmer as woman is to homemaker? Debiasing word embeddings,

    T. Bolukbasi, K.-W. Chang, J. Zou, V . Saligrama, and A. Kalai, “Man is to computer programmer as woman is to homemaker? Debiasing word embeddings,” inProc. Int. Conf. Neural Inf. Process. Syst. (NeurIPS), vol. 29, Barcelona, Spain, 2016, pp. 4356–4364

    work page 2016

  23. [23]

    Semantics derived au- tomatically from language corpora contain human-like biases,

    A. Caliskan, J. J. Bryson, and A. Narayanan, “Semantics derived au- tomatically from language corpora contain human-like biases,”Science, vol. 356, no. 6334, pp. 183–186, 2017

    work page 2017

  24. [24]

    Big data’s disparate impact,

    S. Barocas and A. D. Selbst, “Big data’s disparate impact,”Calif. Law Rev., vol. 104, no. 3, pp. 671–732, 2016

    work page 2016

  25. [25]

    Crawford,Atlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence

    K. Crawford,Atlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence. New Haven, CT, USA: Yale University Press, 2021

    work page 2021

  26. [26]

    Deep fakes: A looming challenge for privacy, democracy, and national security,

    B. Chesney and D. Citron, “Deep fakes: A looming challenge for privacy, democracy, and national security,”Calif. Law Rev., vol. 107, no. 6, pp. 1753–1820, 2019

    work page 2019

  27. [27]

    Will deepfakes do deep damage?

    S. Greengard, “Will deepfakes do deep damage?”Commun. ACM, vol. 63, no. 1, pp. 17–19, 2019

    work page 2019

  28. [28]

    Deepfakes and disinformation: Exploring the impact of synthetic political video on deception, uncertainty, and trust in news,

    C. Vaccari and A. Chadwick, “Deepfakes and disinformation: Exploring the impact of synthetic political video on deception, uncertainty, and trust in news,”Soc. Media + Soc., vol. 6, no. 1, pp. 1–13, 2020

    work page 2020

  29. [29]

    The creation and detection of deepfakes: A survey,

    Y . Mirsky and W. Lee, “The creation and detection of deepfakes: A survey,”ACM Comput. Surv., vol. 54, no. 1, pp. 1–41, 2021

    work page 2021

  30. [30]

    Man-computer symbiosis,

    J. C. R. Licklider, “Man-computer symbiosis,”IRE Trans. Hum. Factor Electron., vol. 1, no. 1, pp. 4–11, 1960

    work page 1960

  31. [31]

    Floridi,The Fourth Revolution: How the Infosphere is Reshaping Human Reality

    L. Floridi,The Fourth Revolution: How the Infosphere is Reshaping Human Reality. Oxford, UK: Oxford University Press, 2014

    work page 2014

  32. [32]

    Directions in hybrid intelligence: Complementing AI sys- tems with human intelligence,

    E. Kamar, “Directions in hybrid intelligence: Complementing AI sys- tems with human intelligence,” inProc. Int. Jt. Conf. Artif. Intell. (IJCAI), New York, NY , USA, 2016, pp. 4070–4073

    work page 2016

  33. [33]

    Machines as teammates: A research agenda on AI in team collaboration,

    I. Seeberet al., “Machines as teammates: A research agenda on AI in team collaboration,”Inf. & Manag., vol. 57, no. 2, pp. 1–22, 2020

    work page 2020

  34. [34]

    From human-human collaboration to human-AI col- laboration: Designing AI systems that can work together with people,

    D. Wanget al., “From human-human collaboration to human-AI col- laboration: Designing AI systems that can work together with people,” inProc. ACM Conf. Hum. Factors Comput. Syst. (CHI), Honolulu, HI, USA, 2020, pp. 1–6. 10

    work page 2020

  35. [35]

    Roles of artificial intelligence in collaboration with humans: Automation, augmentation, and the future of work,

    A. Fügener, D. D. Walzner, and A. Gupta, “Roles of artificial intelligence in collaboration with humans: Automation, augmentation, and the future of work,”Manag. Sci., vol. 72, no. 1, pp. 538–557, 2025

    work page 2025

  36. [36]

    Wittgenstein,Philosophical Investigations

    L. Wittgenstein,Philosophical Investigations. Oxford, UK: Basil Blackwell, 1953

    work page 1953

  37. [37]

    Attention is all you need,

    A. Vaswaniet al., “Attention is all you need,” inProc. Int. Conf. Neural Inf. Process. Syst. (NeurIPS), vol. 30, Long Beach, CA, USA, 2017, pp. 6000–6010

    work page 2017

  38. [38]

    Language models are few-shot learners,

    T. Brownet al., “Language models are few-shot learners,” inProc. Int. Conf. Neural Inf. Process. Syst. (NeurIPS), vol. 33, Virtual Event, 2020, pp. 1877–1901

    work page 2020

  39. [39]

    Training language models to follow instructions with human feedback,

    L. Ouyanget al., “Training language models to follow instructions with human feedback,” inProc. Int. Conf. Neural Inf. Process. Syst. (NeurIPS), vol. 35, New Orleans, LA, USA, 2022, pp. 27 730–27 744

    work page 2022

  40. [40]

    Auto-regressive next-token predictors are universal learn- ers,

    E. Malach, “Auto-regressive next-token predictors are universal learn- ers,” inProc. Int. Conf. Mach. Learn. (ICML), Vienna, Austria, 2024, pp. 1–15

    work page 2024

  41. [41]

    Detecting fake content with relative entropy scoring,

    T. Lavergne, T. Urvoy, and F. Yvon, “Detecting fake content with relative entropy scoring,” inProc. Int. Conf. Uncovering Plagiarism Authorship Soc. Softw. Misuse (PAN), vol. 377, Patras, Greece, 2008, pp. 27–31

    work page 2008

  42. [42]

    GLTR: Statistical detection and visualization of generated text,

    S. Gehrmann, H. Strobelt, and A. Rush, “GLTR: Statistical detection and visualization of generated text,” inProc. Annu. Meet. Assoc. Comput. Linguist. (ACL), Florence, Italy, 2019, pp. 111–116

    work page 2019

  43. [43]

    DetectLLM: Leveraging log rank information for zero-shot detection of machine-generated text,

    J. Su, T. Zhuo, D. Wang, and P. Nakov, “DetectLLM: Leveraging log rank information for zero-shot detection of machine-generated text,” in Proc. Conf. Empir. Methods Nat. Lang. Process. (EMNLP), Singapore, 2023, pp. 12 395–12 412

    work page 2023

  44. [44]

    De- tectGPT: Zero-shot machine-generated text detection using probability curvature,

    E. Mitchell, Y . Lee, A. Khazatsky, C. D. Manning, and C. Finn, “De- tectGPT: Zero-shot machine-generated text detection using probability curvature,” inProc. Int. Conf. Mach. Learn. (ICML), vol. 202, Honolulu, HI, USA, 2023, pp. 24 950–24 962

    work page 2023

  45. [45]

    TURINGBENCH: A benchmark environment for Turing test in the age of neural text gen- eration,

    A. Uchendu, Z. Ma, T. Le, R. Zhang, and D. Lee, “TURINGBENCH: A benchmark environment for Turing test in the age of neural text gen- eration,” inProc. Conf. Empir. Methods Nat. Lang. Process. (EMNLP), Punta Cana, Dominican Republic, 2021, pp. 2001–2016

    work page 2021

  46. [46]

    A watermark for large language models,

    J. Kirchenbauer, J. Geiping, Y . Wen, J. Katz, I. Miers, and T. Goldstein, “A watermark for large language models,” inProc. Int. Conf. Mach. Learn. (ICML), vol. 202, Honolulu, HI, USA, 2023, pp. 17 061–17 084

    work page 2023

  47. [47]

    Learning word vectors for sentiment analysis,

    A. L. Maas, R. E. Daly, P. T. Pham, D. Huang, A. Y . Ng, and C. Potts, “Learning word vectors for sentiment analysis,” inProc. Annu. Meet. Assoc. Comput. Linguist. (ACL), Portland, Oregon, USA, 2011, pp. 142– 150

    work page 2011

  48. [48]

    A thorough examination of the CNN/Daily Mail reading comprehension task,

    D. Chen, J. Bolton, and C. D. Manning, “A thorough examination of the CNN/Daily Mail reading comprehension task,” inProc. Annu. Meet. Assoc. Comput. Linguist. (ACL), Berlin, Germany, 2016, pp. 2358–2367

    work page 2016

  49. [49]

    Pointer sentinel mixture models,

    S. Merity, C. Xiong, J. Bradbury, and R. Socher, “Pointer sentinel mixture models,” inProc. Int. Conf. Learn. Represent. (ICLR), Toulon, France, 2017, pp. 1–15

    work page 2017

  50. [50]

    On the use of arXiv as a dataset,

    C. B. Clement, M. Bierbaum, K. P. O’Keeffe, and A. A. Alemi, “On the use of arXiv as a dataset,” inProc. Int. Conf. Learn. Represent. (ICLR) Workshop, New Orleans, LA, USA, 2019, pp. 1–7. Ching-Chun Changreceived the PhD in Computer Science from the Uni- versity of Warwick, UK, in 2019. He is currently affiliated with the National Institute of Informatics...

    work page 2019