M. Beatrice Fazi
School of Media, Arts and Humanities, University of Sussex
Personal Home Page
Title of the talk: “Learning Why: Causality, Cognition and the Future of AI“
Abstract: This talk will explore causality as an abstraction that could be established within contemporary computational situations produced by artificial intelligence (AI). Is causality part of thinking? And, if causation is an expression of thought, should the machines that are said to simulate human cognition address causal reasoning too? Most scholars and scientists would agree that, in the past decade, AI techniques focusing on data-driven automated learning (for instance, deep neural networks) have been successful in performing tasks that were once difficult to achieve computationally. While there has been a lot of progress, the same scholars and scientists behind this success also express that there is still much to do. Importantly, researchers tend to agree that the learning performance of these AI techniques does not match that of human brains. Would the inclusion of causality, or of a representation thereof, help these computational systems to have better learning results? This talk will signpost some of the philosophical implications of this prospect and will link these implications to considerations about the future of AI.
Institute of Philosophy and Sociology, Polish Academy of Sciences
Personal Home Page
Title of the talk: “Putting it all together. A plea for theorizing in cognitive (neuro)science“
Abstract: In my talk, I argue for more stress on theorizing in cognitive (neuro)science. Progress in cognitive research does not rely only on having more reliable experimental evidence and new computational frameworks for modeling but also on the proper use of cognitive theories. However, what we currently lack is proper understanding of the role of theorizing, which could make cognitive (neuro)science a headless rider doomed to face ever new crises. Most crucially, theories should not be understood as mere speculation or as simple inductive generalizations.
By analyzing theories in terms of cognitive artifacts that support our cognitive tasks when we perform appropriate operations on these artifacts, I hope to shed more light on their nature. This perspective leads to the realization that there are multiple distinct kinds of theories dedicated to entirely different roles. By studying their roles, along with their possible epistemic vices and virtues, we can gain more insight into how theorizing should proceed. In short, I urge that we need to change the research culture so as to appreciate the varieties of distinct kinds of theories in cognitive (neuro)science. Having this insight, we can hope to put it all together systematically.
Abteilung Allgemeine Psychologie, Justus-Liebig Universitaet
Personal Home Page
Title of the Talk: “Inspiration loops between models of seeing and machines that see“
Abstract: Using visual perception as a case study, I will propose that questions in cognitive science are not passed from one discipline to the next, but are conversations among increasingly many disciplines. The question of “how vision works” has spread from the domain of philosophy, into that of psychology, then neuroscience, then computer science, without losing any of its previous footholds. In each generation, researchers are often drawn to harvest the low-hanging fruits newly revealed by some technological advance. When the methods of experimental psychology were formalised in the 1850s, it became natural and exciting to ask how visual perception relates to physical properties of visual stimuli. Likewise, recent successes at optimising computers to recognise objects in images make it natural and exciting to ask how computations in these artificial visual systems relate to those in brains. Old methods and concepts remain vital though, and ideas from psychology—visual attention, working memory, curriculum learning—feed back into the design of computer vision systems. My talk will celebrate the cycles of mutual inspiration over the past 80 years between efforts to model human vision, and efforts to engineer machines that see.