Governance, Regulation and Policy
Keystone
Simone Stumpf (Project Lead), Monika Harvey, Yunhyong Kim, Craig Macdonald, Graham McDonald, Iadh Ounis, Alesandro Vinciarelli, Mark Wong, Evangelia Fringi, Zoe Bartliff, Jane Blanchard, Paul McMillan, Adam Turner, Vyron Christodoulou, Elizabeth Black, Elena Simperl, Daniele Quercia, Dan Hunter, Kathryn Simpson, Leonardo Bezerra, Alexander Brownlee, Leif Azzopardi, Ian Ruthven, Yashar Moshfeghi, Siamak Shahandashti, Laura Whyte, Patrizia Di Campli San Vito
“Our project will put auditing power back in the hands of people who best understand the potential impact in the four fields these AI systems are operating in. By the project’s conclusion, we will have developed a fully-featured workbench of tools to enable people without a background in artificial intelligence to participate in audits, make informed decisions, and shape the next generation of AI.”
A significant barrier to reaping the benefits of AI is their unassessed potential for harms. Hence, AI auditing is imperative but it has been haphazard, unsystematic, and left solely in the hands of experts. Our project aims to fix this challenge through participatory AI auditing where diverse stakeholders without an AI background undertake audits of predictive and generative AI, either individually or collectively. We will produce workbenches that support stakeholders in auditing AI, embedded in methodologies which guide these audits, grounded in four use cases. We will train stakeholders in participatory auditing and work towards a certification framework.
For predictive AI, we tackle two distinct use cases, Health and Media Content. In Health, we will work initially on two datasets with two different sets of stakeholders; this will enable us to investigate how our findings generalise across contexts in the same domain. Scottish Patients at Risk of Readmission and Admission (SPARRA), developed by NHS NSS and Public Health Scotland (both involved in this project), helps healthcare professionals by predicting a person’s likelihood of being admitted to hospital as an emergency in-patient within the next year. This is calculated monthly for approximately 80% of all people living in Scotland. Previous investigations by AI experts have already revealed bias in SPARRA but participatory auditing is hampered because the detailed dataset is not available publicly. Stakeholders for this context are healthcare professionals, NHS decision makers and patients.
The second dataset, the School Attachment Monitor Corpus contains videos of children undergoing the Manchester Child Attachment Story Task. It aims to help with the identification of children with insecure attachment by using a variety of social signal processing methods. This dataset suffers from bias, privacy issues and a lack of agreement about ground truth. Stakeholders include psychologists, social workers, educators, and parents of children undergoing assessment. In the Media Content use case, we will address fairness in a search engine setting in collaboration with Istella.
An unfair search engine can result in amplified misinformation and biased perspectives being propagated through society, as well as discriminating against groups that are marginalised in the search results leading to a loss of real-world opportunities and a negative economic and social impact. Potential stakeholders in this use case are search engine developers and users. We also will work on the publicly available Jigsaw dataset, which has been used to detect hate speech in social media and suffers from bias in hate speech classification. Stakeholders will include social media moderators as well as social media users.
For generative AI, we will work on two use cases. For Cultural Heritage, we will employ datasets provided by project partners National Library of Scotland, Museum Data Service, and the David Livingstone Birthplace, involving collections of historical material including newspapers, books, manuscripts, records, and poem collections. Generative AI allows users to ask questions about the content, offers new interpretations of the material, annotates the material, and offers speculative explanations of phenomena. The risks of generative AI in this situation include misrepresentations, such as hallucinations, historical bias arising from the data, or lack of consideration of the context in which historical events have occurred.
Stakeholders include collections’ curators as well as their users. For Collaborative Content Generation, project partners Wikimedia Foundation and Wikidata will support our research into generative AI used to write articles, focusing on under-resourced languages. There are more than 300 language editions of Wikipedia, but their number of articles and editors varies greatly. For example, while Arabic is the fifth most spoken language by number of speakers, Arabic Wikipedia has only 10% of the articles compared to the English language version, created and maintained by fewer than 4k editors. Given Wikipedia’s role as a trusted information source, this can entrench existing inequalities in accessing and sharing knowledge, hamper cultural diversity and heritage efforts, create unfair erasure of marginalised voices and representation, and contribute to the spread of misinformation. More perniciously, foundational AI models also use Wikimedia articles as training data, leading to a potential feedback loop of harms. We will focus on articles about health, linking to use case 1. Stakeholders will include article editors, and well as users of article and information written by AI, including fact checkers, journalists and researchers, supported by project partners Full Fact and the Open Data Institute.
Outputs will include workbenches that support stakeholders in auditing AI, with audits embedded in methodologies that define how, when and who carries out the audit. Following rigorous evaluation, the focus will move to training stakeholders to carry out the audits, to embed project outputs in practice.
The PHAWM project kicked off with a Launch and Networking Event at the University of Strathclyde’s Technology and Innovation Centre. The event brought together project researchers, partners, and stakeholders from across the UK and beyond.
Professor Dame Muffy Calder set the tone with an inspiring welcome, drawing on her expertise as RAI UK Skills Pillar Chair and Vice-Principal of the University of Glasgow. Our project team then took centre stage, sharing insights into our key objectives and an overview of our first of the health and cultural heritage use cases that the project will explore.
This was followed by an interactive Responsible Research and Innovation (RRI) activity. Our approach to RRI has been inspired by UKRI Trustworthy Autonomous Systems Hub’s approach of building it into the core of our activities.
The afternoon finished with a networking event with our stakeholders, connecting with other RAi-funded projects, researchers (including early-career and PhD), and the wider AI ecosystem across the UK.
The team will work towards a certification framework for AI solutions, democratising AI development and ensuring that AI is safe and trustworthy.
RAi UK will store your data in accordance with the General Data Protection Regulation 2017 (GDPR). We will not share your data with any third parties and you are given opportunities to unsubscribe at any time within the electronic communications you receive, or by emailing info@rai.ac.uk