The new COPE Discussion document “Artificial intelligence (AI) in decision making”, was launched at the COPE webinar, discussing ethics issues relating to the application of artificial intelligence in editorial publishing processes. The session features guest speakers providing examples of the development of editing tools and perspectives on technology ethics.
The webinar is one of nine sessions as part of the COPE Seminar 2021.
COPE recommendations: AI in decision making
Marie Soulière, Head of Publishing Operations at Frontiers and COPE Council Member, provided the introduction and background to the session, with an overview of the COPE discussion document on AI in decision making. Marie defined AI and its characteristics, and its relationship to machine learning (ML) and natural language processing (NLP). Marie gave examples of the value of AI in editorial processes, in helping to solve problems that would be difficult or impossible for the human mind, to enhance efficiency and accuracy in processing, and to remove or minimise the potential of personal biases.
Marie gave examples of editorial checks which incorporate AI, such as plagiarism/text similarity checking, data falsification, image manipulation, language and image quality, citation relevance, adherence to journal scope, subject area relevance and impact, and discussed the role that the COPE document plays in attempting to understand under what circumstances s AI based decisions in the publishing process would be deemed unethical.
COPE recommendations: AI in decision making
Leveraging deep learning and natural language processing for publication readiness
Nishchay Shah, Chief Technology Officer and Head of Cactus Labs, the AI and ML division of Cactus Communications, presented on leveraging artificial intelligence to improve the quality of the publication process, detailing the principles and processes involved in the Cactus AI editing tools.
Nishchay compared the differences between traditional rule based editing corrections, based on handwritten linguistically-influenced rules, in contrast with the more sophisticated, flexible, semantic knowledge of the new Cactus neural network, developed through training on edits curated from over 300 million sentences.
Describing the complex processes that go into the development of AI models, Nishchay used flowcharts to illustrate the paths documents progress along when they are submitted to the neural machine editing. Examples are provided which show the corrections made, and the labels applied to the categories of edits, going beyond basic grammar corrections into the realms of readability, redundancy and wordiness.
Leveraging deep learning for publication readiness
Using AI for decision support: some ethical issues
Finally, Ibo van de Poel, Professor in ethics and technology, and Head of Department of Values, Technology and Innovation at TU Delft, the Netherlands, discussed the ethical issues of using AI for decision support. He referenced the EU expert group on AI, which has formulated four ethical principles for the foundations for trustworthy AI:
- respect for human autonomy
- prevention of harm
- fairness explicability
The Group has seven key requirements to guide the realisation and application of trustworthy AI, which call for an active awareness of human agency, technical robustness and safety, privacy and data governance, transparency, diversity and non-discriminatory practices, societal and environmental well being and accountability.
From these principles, Ibo focused on three key ethical concerns as they relate to editorial decision making in publishing:
- bias and fairness, in which he highlighted possible causes of biases, where they can be introduced or how they can develop;
- accountability, in which he discussed the role publishers and editors must play in mediating AI processes, provide a meaningful human element;
- and explainability, which refers to the necessity for AI.
Using AI for decision support; some ethical issues
Useful links
Related resources
COPE Artificial intelligence (AI) in decision making Discussion Document
About this resource
Full page history
-
10 January 2023
Published with access to all
-
10 January 2023
Published with member-only access.
-
18 October 2021
Sabah
-
13 September 2021
Sabah
-
31 August 2021
Sabah
-
31 August 2021
Sabah
-
26 August 2021
Sabah
-
26 August 2021
Sabah
-
26 August 2021
Sabah
-
26 August 2021
Sabah
-
25 August 2021
Sabah
-
23 August 2021
Sabah