Typology of risks of generative text-to-image models

Charlotte Bird; Eddie Ungless; Atoosa Kasirzadeh

doi:10.1145/3600211.3604722

Typology of risks of generative text-to-image models

Charlotte Bird, Eddie Ungless, Atoosa Kasirzadeh

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

Abstract

This paper investigates the direct risks and harms associated with modern text-to-image generative models, such as DALL-E and Midjourney, through a comprehensive literature review. While these models offer unprecedented capabilities for generating images, their development and use introduce new types of risk that require careful consideration. Our review reveals significant knowledge gaps concerning the understanding and treatment of these risks despite some already being addressed. We offer a taxonomy of risks across six key stakeholder groups, inclusive of unexplored issues, and suggest future research directions. We identify 22 distinct risk types, spanning issues from data bias to malicious use. The investigation presented here is intended to enhance the ongoing discourse on responsible model development and deployment. By highlighting previously overlooked risks and gaps, it aims to shape subsequent research and governance initiatives, guiding them toward the responsible, secure, and ethically conscious evolution of text-to-image models.

Original language	English
Title of host publication	AIES '23
Subtitle of host publication	Proceedings of the 2023 AAAI/ACM Conference on AI, Ethics, and Society
Editors	Francesca Rossi, Sanmay Das, Jenny Davis, Kay Firth-Butterfield, Alex John
Place of Publication	New York
Publisher	Association for Computing Machinery (ACM)
Pages	396–410
Number of pages	15
ISBN (Electronic)	9798400702310
DOIs	https://doi.org/10.1145/3600211.3604722
Publication status	Published - 29 Aug 2023
Event	Sixth AAAI/ACM Conference on Artificial Intelligence, Ethics, and Society - Palais des congrès de Montréal, Montréal, Canada Duration: 8 Aug 2023 → 10 Aug 2023 https://www.aies-conference.com/2023/

Publication series

Name	Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society
Publisher	Association for Computing Machinery (ACM)
ISSN (Electronic)	2168-4081

Conference

Conference	Sixth AAAI/ACM Conference on Artificial Intelligence, Ethics, and Society
Abbreviated title	AIES 2023
Country/Territory	Canada
City	Montréal
Period	8/08/23 → 10/08/23
Internet address	https://www.aies-conference.com/2023/

Keywords / Materials (for Non-textual outputs)

generative AI
generative models
text-to-image models
responsible AI
AI ethics
AI safety
AI governance
AI risks

Access to Document

10.1145/3600211.3604722

BirdEtAl2023TypologyOfRisksAccepted author manuscript, 476 KBLicence: Creative Commons: Attribution (CC-BY)

1 Finished
1 Active

BRAID (Enabling A Responsible AI Ecosystem)
Vallor, S. & Luger, E.
14/11/22 → 13/11/28
Project: Research
Two studentships in Natural Language Processing
King, S. & Richmond, K.
Non-EU industry, commerce and public corporations
1/09/20 → 31/08/24
Project: Research

Cite this

Bird, C., Ungless, E., & Kasirzadeh, A. (2023). Typology of risks of generative text-to-image models. In F. Rossi, S. Das, J. Davis, K. Firth-Butterfield, & A. John (Eds.), AIES '23: Proceedings of the 2023 AAAI/ACM Conference on AI, Ethics, and Society (pp. 396–410). (Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society). Association for Computing Machinery (ACM). https://doi.org/10.1145/3600211.3604722

Bird, Charlotte ; Ungless, Eddie ; Kasirzadeh, Atoosa. / Typology of risks of generative text-to-image models. AIES '23: Proceedings of the 2023 AAAI/ACM Conference on AI, Ethics, and Society. editor / Francesca Rossi ; Sanmay Das ; Jenny Davis ; Kay Firth-Butterfield ; Alex John. New York : Association for Computing Machinery (ACM), 2023. pp. 396–410 (Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society).

@inproceedings{7fae0547dfed4c9386b425c814886987,

title = "Typology of risks of generative text-to-image models",

abstract = "This paper investigates the direct risks and harms associated with modern text-to-image generative models, such as DALL-E and Midjourney, through a comprehensive literature review. While these models offer unprecedented capabilities for generating images, their development and use introduce new types of risk that require careful consideration. Our review reveals significant knowledge gaps concerning the understanding and treatment of these risks despite some already being addressed. We offer a taxonomy of risks across six key stakeholder groups, inclusive of unexplored issues, and suggest future research directions. We identify 22 distinct risk types, spanning issues from data bias to malicious use. The investigation presented here is intended to enhance the ongoing discourse on responsible model development and deployment. By highlighting previously overlooked risks and gaps, it aims to shape subsequent research and governance initiatives, guiding them toward the responsible, secure, and ethically conscious evolution of text-to-image models.",

keywords = "generative AI, generative models, text-to-image models, responsible AI, AI ethics, AI safety, AI governance, AI risks",

author = "Charlotte Bird and Eddie Ungless and Atoosa Kasirzadeh",

note = "Acknowledgements: We would like to thank the UKRI Arts and Humanities Research Council (grant AH/X007146/1) for the policy fellowship that supported this work. We thank Shannon Vallor, Ewa Luger, and the members of Ada Lovelace Institute for helpful discussions. We also thank James Stewart, Lilian Edwards, Andres Guadamuz, and three anonymous reviewers whose comments improved our work. Eddie L. Ungless is supported by the UKRI Centre for Doctoral Training in Natural Language Processing, funded by UKRI (Grant EP/S022481/1) and the University of Edinburgh, School of Informatics. Charlotte Bird is supported by the Baillie Giﬀord PhD Scholarship at the Centre for Technomoral Futures.; Sixth AAAI/ACM Conference on Artificial Intelligence, Ethics, and Society, AIES 2023 ; Conference date: 08-08-2023 Through 10-08-2023",

year = "2023",

month = aug,

day = "29",

doi = "10.1145/3600211.3604722",

language = "English",

series = "Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society",

publisher = "Association for Computing Machinery (ACM)",

pages = "396–410",

editor = "Francesca Rossi and Sanmay Das and Jenny Davis and Kay Firth-Butterfield and Alex John",

booktitle = "AIES '23",

address = "United States",

url = "https://www.aies-conference.com/2023/",

}

Bird, C, Ungless, E & Kasirzadeh, A 2023, Typology of risks of generative text-to-image models. in F Rossi, S Das, J Davis, K Firth-Butterfield & A John (eds), AIES '23: Proceedings of the 2023 AAAI/ACM Conference on AI, Ethics, and Society. Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society, Association for Computing Machinery (ACM), New York, pp. 396–410, Sixth AAAI/ACM Conference on Artificial Intelligence, Ethics, and Society, Montréal, Quebec, Canada, 8/08/23. https://doi.org/10.1145/3600211.3604722

Typology of risks of generative text-to-image models. / Bird, Charlotte; Ungless, Eddie; Kasirzadeh, Atoosa.
AIES '23: Proceedings of the 2023 AAAI/ACM Conference on AI, Ethics, and Society. ed. / Francesca Rossi; Sanmay Das; Jenny Davis; Kay Firth-Butterfield; Alex John. New York: Association for Computing Machinery (ACM), 2023. p. 396–410 (Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society).

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

TY - GEN

T1 - Typology of risks of generative text-to-image models

AU - Bird, Charlotte

AU - Ungless, Eddie

AU - Kasirzadeh, Atoosa

N1 - Acknowledgements: We would like to thank the UKRI Arts and Humanities Research Council (grant AH/X007146/1) for the policy fellowship that supported this work. We thank Shannon Vallor, Ewa Luger, and the members of Ada Lovelace Institute for helpful discussions. We also thank James Stewart, Lilian Edwards, Andres Guadamuz, and three anonymous reviewers whose comments improved our work. Eddie L. Ungless is supported by the UKRI Centre for Doctoral Training in Natural Language Processing, funded by UKRI (Grant EP/S022481/1) and the University of Edinburgh, School of Informatics. Charlotte Bird is supported by the Baillie Giﬀord PhD Scholarship at the Centre for Technomoral Futures.

PY - 2023/8/29

Y1 - 2023/8/29

N2 - This paper investigates the direct risks and harms associated with modern text-to-image generative models, such as DALL-E and Midjourney, through a comprehensive literature review. While these models offer unprecedented capabilities for generating images, their development and use introduce new types of risk that require careful consideration. Our review reveals significant knowledge gaps concerning the understanding and treatment of these risks despite some already being addressed. We offer a taxonomy of risks across six key stakeholder groups, inclusive of unexplored issues, and suggest future research directions. We identify 22 distinct risk types, spanning issues from data bias to malicious use. The investigation presented here is intended to enhance the ongoing discourse on responsible model development and deployment. By highlighting previously overlooked risks and gaps, it aims to shape subsequent research and governance initiatives, guiding them toward the responsible, secure, and ethically conscious evolution of text-to-image models.

AB - This paper investigates the direct risks and harms associated with modern text-to-image generative models, such as DALL-E and Midjourney, through a comprehensive literature review. While these models offer unprecedented capabilities for generating images, their development and use introduce new types of risk that require careful consideration. Our review reveals significant knowledge gaps concerning the understanding and treatment of these risks despite some already being addressed. We offer a taxonomy of risks across six key stakeholder groups, inclusive of unexplored issues, and suggest future research directions. We identify 22 distinct risk types, spanning issues from data bias to malicious use. The investigation presented here is intended to enhance the ongoing discourse on responsible model development and deployment. By highlighting previously overlooked risks and gaps, it aims to shape subsequent research and governance initiatives, guiding them toward the responsible, secure, and ethically conscious evolution of text-to-image models.

KW - generative AI

KW - generative models

KW - text-to-image models

KW - responsible AI

KW - AI ethics

KW - AI safety

KW - AI governance

KW - AI risks

U2 - 10.1145/3600211.3604722

DO - 10.1145/3600211.3604722

M3 - Conference contribution

T3 - Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society

SP - 396

EP - 410

BT - AIES '23

A2 - Rossi, Francesca

A2 - Das, Sanmay

A2 - Davis, Jenny

A2 - Firth-Butterfield, Kay

A2 - John, Alex

PB - Association for Computing Machinery (ACM)

CY - New York

T2 - Sixth AAAI/ACM Conference on Artificial Intelligence, Ethics, and Society

Y2 - 8 August 2023 through 10 August 2023

ER -

Bird C, Ungless E, Kasirzadeh A. Typology of risks of generative text-to-image models. In Rossi F, Das S, Davis J, Firth-Butterfield K, John A, editors, AIES '23: Proceedings of the 2023 AAAI/ACM Conference on AI, Ethics, and Society. New York: Association for Computing Machinery (ACM). 2023. p. 396–410. (Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society). Epub 2023 Aug 10. doi: 10.1145/3600211.3604722

Typology of risks of generative text-to-image models

Abstract

Publication series

Conference

Keywords / Materials (for Non-textual outputs)

Access to Document

Fingerprint

Projects

BRAID (Enabling A Responsible AI Ecosystem)

Two studentships in Natural Language Processing

Cite this