Guiding the Release of Safer E2E Conversational AI through Value Sensitive Design

A. Stevie Bergman; Gavin Abercrombie; Shannon Spruit; Dirk Hovy; Emily Dinan; Y-Lan Boureau; Verena Rieser

doi:10.18653/v1/2022.sigdial-1.4

Guiding the Release of Safer E2E Conversational AI through Value Sensitive Design

A. Stevie Bergman, Gavin Abercrombie, Shannon Spruit, Dirk Hovy, Emily Dinan, Y-Lan Boureau, Verena Rieser

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

Abstract

Over the last several years, end-to-end neural conversational agents have vastly improved their ability to carry unrestricted, open-domain conversations with humans. However, these models are often trained on large datasets from the Internet and, as a result, may learn undesirable behaviours from this data, such as toxic or otherwise harmful language. Thus, researchers must wrestle with how and when to release these models. In this paper, we survey recent and related work to highlight tensions between values, potential positive impact, and potential harms. We also provide a framework to support practitioners in deciding whether and how to release these models, following the tenets of value-sensitive design.

Original language	English
Title of host publication	Proceedings of the 23rd Annual Meeting of the Special Interest Group on Discourse and Dialogue
Publisher	Association for Computational Linguistics
Pages	39–52
Number of pages	14
ISBN (Print)	9781955917667
DOIs	https://doi.org/10.18653/v1/2022.sigdial-1.4
Publication status	Published - 1 Sept 2022

Access to Document

10.18653/v1/2022.sigdial-1.4

Cite this

Bergman, A. S., Abercrombie, G., Spruit, S., Hovy, D., Dinan, E., Boureau, Y.-L., & Rieser, V. (2022). Guiding the Release of Safer E2E Conversational AI through Value Sensitive Design. In Proceedings of the 23rd Annual Meeting of the Special Interest Group on Discourse and Dialogue (pp. 39–52). Association for Computational Linguistics. https://doi.org/10.18653/v1/2022.sigdial-1.4

@inproceedings{d00a6047da0f4146bd7b326b025c3044,

title = "Guiding the Release of Safer E2E Conversational AI through Value Sensitive Design",

abstract = "Over the last several years, end-to-end neural conversational agents have vastly improved their ability to carry unrestricted, open-domain conversations with humans. However, these models are often trained on large datasets from the Internet and, as a result, may learn undesirable behaviours from this data, such as toxic or otherwise harmful language. Thus, researchers must wrestle with how and when to release these models. In this paper, we survey recent and related work to highlight tensions between values, potential positive impact, and potential harms. We also provide a framework to support practitioners in deciding whether and how to release these models, following the tenets of value-sensitive design.",

author = "Bergman, {A. Stevie} and Gavin Abercrombie and Shannon Spruit and Dirk Hovy and Emily Dinan and Y-Lan Boureau and Verena Rieser",

year = "2022",

month = sep,

day = "1",

doi = "10.18653/v1/2022.sigdial-1.4",

language = "English",

isbn = "9781955917667",

pages = "39–52",

booktitle = "Proceedings of the 23rd Annual Meeting of the Special Interest Group on Discourse and Dialogue",

publisher = "Association for Computational Linguistics",

}

Bergman, AS, Abercrombie, G, Spruit, S, Hovy, D, Dinan, E, Boureau, Y-L & Rieser, V 2022, Guiding the Release of Safer E2E Conversational AI through Value Sensitive Design. in Proceedings of the 23rd Annual Meeting of the Special Interest Group on Discourse and Dialogue. Association for Computational Linguistics, pp. 39–52. https://doi.org/10.18653/v1/2022.sigdial-1.4

Guiding the Release of Safer E2E Conversational AI through Value Sensitive Design. / Bergman, A. Stevie; Abercrombie, Gavin; Spruit, Shannon et al.
Proceedings of the 23rd Annual Meeting of the Special Interest Group on Discourse and Dialogue. Association for Computational Linguistics, 2022. p. 39–52.

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

TY - GEN

T1 - Guiding the Release of Safer E2E Conversational AI through Value Sensitive Design

AU - Bergman, A. Stevie

AU - Abercrombie, Gavin

AU - Spruit, Shannon

AU - Hovy, Dirk

AU - Dinan, Emily

AU - Boureau, Y-Lan

AU - Rieser, Verena

PY - 2022/9/1

Y1 - 2022/9/1

N2 - Over the last several years, end-to-end neural conversational agents have vastly improved their ability to carry unrestricted, open-domain conversations with humans. However, these models are often trained on large datasets from the Internet and, as a result, may learn undesirable behaviours from this data, such as toxic or otherwise harmful language. Thus, researchers must wrestle with how and when to release these models. In this paper, we survey recent and related work to highlight tensions between values, potential positive impact, and potential harms. We also provide a framework to support practitioners in deciding whether and how to release these models, following the tenets of value-sensitive design.

AB - Over the last several years, end-to-end neural conversational agents have vastly improved their ability to carry unrestricted, open-domain conversations with humans. However, these models are often trained on large datasets from the Internet and, as a result, may learn undesirable behaviours from this data, such as toxic or otherwise harmful language. Thus, researchers must wrestle with how and when to release these models. In this paper, we survey recent and related work to highlight tensions between values, potential positive impact, and potential harms. We also provide a framework to support practitioners in deciding whether and how to release these models, following the tenets of value-sensitive design.

U2 - 10.18653/v1/2022.sigdial-1.4

DO - 10.18653/v1/2022.sigdial-1.4

M3 - Conference contribution

SN - 9781955917667

SP - 39

EP - 52

BT - Proceedings of the 23rd Annual Meeting of the Special Interest Group on Discourse and Dialogue

PB - Association for Computational Linguistics

ER -

Guiding the Release of Safer E2E Conversational AI through Value Sensitive Design

Abstract

Access to Document

Fingerprint

Cite this