BRIO – Bias, Risk and Opacity in AI
National Research Project (PRIN MUR)
The workshop “Trustworthy AI – Conceptual Challenges and Applications” organised by Mattie Petrolo e Giacomo Zanotti will take place on February 16, 2024 at Politecnico di Milano (Room Alpha, Building 24, Politecnico di Milano, Via Golgi, 40, 20133 Milano).
For more information on the program and on attendance, here is the page of the event:
Trustworthy AI – Conceptual challenges and applications
Giuseppe Primiero will give a talk titled “The role of verification in Trustworthy AI” at the 19th Annual Conference of the Italian Association of Cognitive Sciences (AISC2023). The conference will take place at the University of Genoa, Italy the 14–16 December 2023. The talk will be on the 15th of December 2023, 14.45-15.05.
While a variety of tools are being developed to make ML systems more transparent, logical methods are coming back to play an increasingly important role: by their nature, they may help building and verifying transparent models of computations. A major aim is therefore to develop formal methods that will help the verification of these systems’ trustworthiness and fairness. In this lecture, I will present some recent work in this direction. I will overview the methodology and sketch some principles of the logics developed within the BRIO (Bias, Risk and Opacity in AI) Project (sites.unimi.it/brio) to reason within and about programs with probabilistic outputs obtained under possibly opaque distributions. Under this approach trustworthiness can be formally verified as an admissible distance from the behavior of a fair and transparent counterpart.
Roberta Ferrario and Emanuele B. Grifoni are giving a Laboratory for an inter-university Master’s Degree Program (University of Bergamo, University of Pavia and IUSS Pavia). The laboratory is about applied ontologies and it is titled “Philosophical Knowledge: Foundations, Methods, Applications”. The laboratory will take place from the 13 December 2023 until the 21 December 2023 at the University of Bergamo.
Emanuele B. Grifoni gave a talk titled “Lions we think we understand. A Wittgensteinian critique of Large Language Models” at the workshop Formalization, Modellization, and Representation of Social Pattern Analyses (ForMoRe). The workshop was organised by the Social Pattern Recognition Lab (SPaRe Lab) and by the Hans Schadee Methods Center (HSC). The workshop took place on November 20, 2023 in Trento, Italy.
https://webmagazine.unitn.it/evento/sociologia/116210/formore-workshop
Emanuele B. Grifoni will give a talk titled “Judgements, Definitions and Intrinsic Bias in Large Language Models”at the 19th Annual Conference of the Italian Association of Cognitive Sciences (AISC2023). The conference will take place at the University of Genoa, Italy the 14–16 December 2023. The talk will be on the 14th of December 2023, 17.30-17.50.
Through Wittgenstein’s “Investigations” we illustrate an intrinsic bias of Large Language Models (LLMs) at the design level. The current statistical approach cannot account for the normative aspect of rule-following and therefore of communication. This, we believe, is also at the basis of phenomena like “hallucinations” or “confabulations” in LLMs.