«Predictable AI» (predictable-ai.org) is an initiative to study the predictability of present and future AI systems, and their interplay throughout human-AI ecosystems. On the one hand, the predictability of AI systems requires the re-understanding of evaluation as an inference problem that anticipates how AI systems will behave for each (co-)operative condition (problem and stakeholders), by extracting cognitive capabilities or by learning ML assessors from test data. On the other hand, the predictability of future AI as an ecosystem calls for discovering robust patterns and scaling laws at different aggregation levels that ultimately connect predictability with control, liability and risks.
The initiative will start on 8th March 2023 with a singular event consisting of invited talks, panels, short highlights and time for networking.
The event is co-organised and will count with a numerous representation of colleagues from the CFI (http://lcfi.ac.uk/) and CSER (https://www.cser.ac.uk/) in Cambridge, and ValGRAI (https://valgrai.eu/) and VRAIN (https://vrain.upv.es/) in the Valencian Country.
More information about the initiative, the event and the agenda of the day at https://predictable-ai.org. Registration is open (and free of charge) at https://forms.office.com/e/r6yVZRC30Z.