AI with AI

In COVID-related AI news, Andy and Dave discuss an editorial in The Lancet Digital Health, which examines whether preliminary models add clinical value to health-care systems. In regular AI news, an Italian court rules that the European food delivery app Deliveroo used a “discriminatory” algorithm, potentially opening the door for liability even with unintentional algorithmic discrimination. A study from Google, OpenAI, Apple, Stanford, Berkeley, and Northeastern shows that large language models trained on public data can expose personal information, by making it possible to extract specific pieces of training data. In research, OpenAI combines the mini-GPT algorithm DALL-E with an image-to-text algorithm CLIP, to create an extremely powerful and flexible generative model, capable of generating high-quality images based on text instructions. The report of the week comes from the Connections 2020 Conference proceedings, which examined Representing AI in Wargames. The survey of the week looks at neural network interpretability. Kevin Murphy provides the book of the week, with Probabilistic Machine Learning: An Introduction. And Geoff Hinton speak on Eye on AI with Craig S. Smith about his latest research and the future of AI