Ep 8: Journey from physiotherapy to computational neuroscience, Fairness in AI and the Deep learning Indaba with Siobhan Mackenzie Hall - Oxford Neural Interfacing, University of Oxford

Welcome to Episode 8! Science in Real Life Podcast’s host Nyasita Ondari and University of Oxford postgraduate student Siobhan Mackenzie Hall, discuss Siobhan's journey from physiotherapy to computational neuroscience, fairness in AI and the Deep Learning Indaba. Siobhan is also a member of the Oxford Artificial Intelligence Student Labs which works to measure and mitigate societal bias in existing AI models. Notably, she is a qualified physiotherapist and worked in state-run hospitals in South Africa as a frontline healthcare worker before pursuing a Masters degree as part of the Biomedical Engineering Research Group at Stellenbosch University, South Africa. Siobhan is an organizer of the Deep Learning Indaba, which is a grassroots movement with the mission to strengthen African Machine Learning and AI. Papers mentioned: A Prompt Array Keeps the Bias Away: Debiasing Vision-Language Models with Adversarial Learning - This work provides a method for steering the model in such a way that concepts, such as the intelligence of a person, don’t change with respect to the person’s gender. Gender can be swapped for any protected attribute of interest. Balancing the Picture: Debiasing Vision-Language Datasets with Synthetic Contrast Sets - This work synthetically modifies a person’s demographic attribute - specifically gender in this work, using models like Stable Diffusion. This allows the development of a balanced dataset with men and women equally represented VisoGender: A dataset for benchmarking gender bias in image-text pronoun resolution - This benchmark helps to test models for gender bias in occupational settings and provides metrics against which can be tested for improvements. Other references: Ethical and social risks of harm from Language Models by Laura Weidinger and authors from Google DeepMind which provides a framework for recognising representational and allocational harms in models Google’s Bard, which can be handy for helping with tasks supporting writing thorough documentation for code Andrew Ng’s The Batch is a newsletter that gives regular insight into AI advances as well as personal letters from Andrew Ng which helps shape ideas around how the developments of AI and how we can work to improve this for all The Deep Learning Indaba (https://deeplearningindaba.com/2023/) and the DLI mentorship program Coursera: https://www.coursera.org/(financial aid policy)