ICYMI-AI — August 11, 2023 — AI Ethics & Society Edition

This week I attended the sixth AI Ethics & Society conference (AIES 2026), held this year in Montreal. The joint ACM/AAAI conference had 237 initial submissions and 68 accepted papers, with authors from the US (300+); UK (100+); Canada and Germany (35+); and Japan, Netherlands, Italy, France, Ireland, India and china (10+). The conference covers the fields of Computer Science, Law and Policy, Philosophy and Ethics, and the Social Sciences. I’m a proud member of the AIES 2023 Program Committee. You can download the entire program. Previous conferences are in the ACM Digital Library under AIES, which is where the current year’s papers will be archived within a few weeks.

For me, the key message of the conference is that there are two emerging thrusts for those working on ethical AI. The larger group of people, largely at universities and research labs, are laying the scientific groundwork for ethical AI, including developing algorithms, metrics and datasets. These people are mostly located at universities and corporate research labs in the US, Canada, the UK, Germany, and Northern Europe. Then there is a smaller group of practitioners being tasked by corporations and governments right now to certify that AI systems being deployed meet certain ethical requirements.

Currently, there are few players in this space, but I note that Trevor Hughes, CEO and President of the International Association of Privacy Professionals attended the conference. IAPP is making a major push into AI governance, and I expect other professional certification organizations soon will join in as well.

In celebration of AIES 2023, this week’s ICYMI-AI will be devoted to the student posters and papers, with links to arXiv papers and photos of the poster session where possible.

What does it mean to be a responsible AI practitioner? (Shalaleh Rismani, AJung Moon)

• Presents an ontology of roles and skills.
• “Due to the nascent nature of these roles, however, it is unclear to future employers and aspiring AI ethicists what specific function these roles serve and what skills are necessary to serve the functions.”
• Echos several other talks in that there is a desperate need for AI governance in practice. While an overwhelming number of AI governance frameworks have been published, there seems to be little agreement on how to move forward.
• https://arxiv.org/abs/2205.03946

Flickr Africa: Examining Geo-Diversity in Large-Scale, Human-Centric Visual Data (Keziah Naggita, Julienne LaChance, Alice Xiang)

• Sony AI conducted a large-scale study of metadata of Flickr images and associated metadata from Flickr between February 2004 and February 2022 (18 years). The data came from key-word searches using the Flickr API for several African and European countries.
• “For the African geotagged images we randomly selected for the crowdsourcing task, images were far more likely to be taken by foreigners whereas the opposite trend was observed for high GDP European nations, according to comparisons of geotags and user-reported location. For Sierra Leone, +169% of images were captured by foreigners, compared to Switzerland (-31%). The same trend applies to Djibouti and Cyprus (+335% and -49%) and CAF (Central African Federation) and Finland (+272% and -49%). Thus, images sourced from Africa may be more predisposed to bias resulting from an “othering” phenomenon, and less representative of African cultural viewpoints.”
• The study found far less representative data from Africa than from Europe. “Given that image content was fairly similar across most attributes annotated, and there exist far fewer geotagged images from Africa, we anticipate insufficient African data availability for certain computer vision tasks.”
• Unfortunately, Sony AI’s authors were unable to attend the conference in Canada due to visa concerns, so another member of the Sony AI team presented. (As a side comment, I was impressed that the Sony AI team presents their photos in random order to avoid the problem of biases due to alphabetical ordering.)
• https://r2hcai.github.io/AAAI-23/files/CameraReadys/13.pdf

ChatGPT Perpetuates Gender Bias in Machine Translation and Ignores Non-Gendered Pronouns: Findings across Bengali and Fine other Low-Resource Languages (Sourojit Ghosh, Aylin Caliskan)

• “We center this study around Bengali, the 7th most spoken language globally, but also generalize our findings across five other languages: Farsi, Malay, Tagalog, Thai, and Turkish.”
• “We find that ChatGPT perpetuates gender defaults and stereotypes assigned to certain occupations (e.g. man = doctor, woman = nurse) or actions (e.g. woman = cook, man = go to work), as it converts gender-neutral pronouns in languages to `he' or `she'.”
• “We also observe ChatGPT completely failing to translate the English gender-neutral pronoun `they' into equivalent gender-neutral pronouns in other languages, as it produces translations that are incoherent and incorrect.”
• “While it does respect and provide appropriately gender-marked versions of Bengali words when prompted with gender information in English, ChatGPT appears to confer a higher respect to men than to women in the same occupation.”
• https://arxiv.org/abs/2305.10510

It takes a village to raise an AI system - realising AI potential in healthcare (Hubert Dariusz Zaj?c)

While at the poster session, I had a conversation with Hubert Zaj?c, a Ph.D. Fellow in Health AI at the University of Copenhagen. His poster presented three already published papers that will be part of his dissertation, a cross-cultural look at AI in medical systems in Denmark and in Africa. One of the main findings is that there was more variation within countries than between them - that is, differences that result from hospital specialization and equipment vendor/age/design are more important than whether the same equipment is being used in Europe or in Africa (assuming that it’s being used by doctors who have received similar training.)

• Clinician-Facing AI in the Wild: Taking Stock of the Sociotechnical Challenges and Opportunities for HCI
• Designing ground truth for Machine Learning - conceptualisation of a collaborative design process between medical professionals and data scientists
• Ground Truth Or Dare: Factors Affecting The Creation Of Medical Datasets For Training AI Hubert D. Zajac, Natalia R. Avlona, Finn Kensing, Tariq O. Andersen, and Irina Shklovski (AIES 2023, not yet posted)?

Best Paper: The Bureaucratic Challenge to AI Governance: An Empirical Assessment of Implementation at U.S. Federal Agencies (Christie Lawrence, Isaac Cui, and Daniel Ho)

• Researchers at Stanford University reviewed the AI implementation strategies of US Federal Agencies required under various AI policies and executive orders.
• During the presentation, the speaker showed examples of one federal agency that has invested hundreds of millions of dollars on AI research and deployment, but stated “none” on an official form asking what it had done.
• Another federal agency did not include a highly publicized facial recognition and identification application in its inventory of AI use cases.
• https://dho.stanford.edu/wp-content/uploads/AIES-Bureaucratic-Challenge.pdf