Joint session with Helsinki Philosophy Colloquium:
Time: Thursday 25 January at 14:15–15:45 EEST/EET
Place: University of Helsinki Metsätalo, Unioninkatu 40, room B214 (room 4), 2nd floor, and online (Zoom link)
Time: Fridays (dates below) at 14:15–15:45 EEST/EET
Place: University of Helsinki main building, Unioninkatu 34, 4th floor, room U4078, and/or online (Zoom link). Please check the format details for each session separately.
The function of mindreading is to facilitate prediction of other people’s behavior and to enable appropriate action in social contexts (Dennett 1987). Mindreading in humans can proceed by attribution of mental states like beliefs and desires but also make use of stereotypes or scripts. Some think that we simulate others on the basis of our own experience (Goldman 2006), others think we rely on a rich body of knowledge and draw inferences to the best explanation (Gopnik & Wellman 2012).
In this paper, we suggest a form of pervasive mindreading that involves artificial intelligence. Social media platforms such as Facebook and Twitter, but also companies like Target or content providers such as Spotify or Netflix draw on a wealth of personal data from their users and make use of predictive analytics (Siegel 2013) to predict how and when individuals will vote, think, love or divorce, steal or kill, consume or click as well as what they will prefer or what moods they will be in.
We show that this is a genuine form of mindreading, where the mindreading targets remain individual people, but the mindreading subjects here are not other human individuals, but the big tech companies that have access to the big data which are needed and figure as the knowledge base on which predictive analytics works. They do not only use it to predict people’s behavior but also to engage in efficient mindshaping (Zawidski 2013).
After outlining the analogy to human mindreading according to the Bayesian version of the theory-theory, we discuss theoretical and ethical implications of this form of mindreading. This is collaborative work with Carissa Veliz (Oxford).
Tobias Schlicht is professor of philosophy at Ruhr-University in Bochum, Germany. His areas of specialization are consciousness and cognition and he has a special interest in how Kant’s theoretical philosophy can be informative in contemporary philosophy of cognitive science. He has worked extensively on situated and social cognition, is the author of Philosophy of Social Cognition (Palgrave 2022) and co-editor of What are mental representations? (OUP 2020).
Recent advances in machine learning provide us with the tools to model human perception and cognition in much higher detail and for more complex situations than before. To draw scientific insights from our models we need to design our models for this purpose and to adjust our methodology. Here, I will present three concrete projects towards this goal: First, we recently improved representational similarity analysis, a technique to compare high dimensional representations to each other, like a deep neural network to the brain for example. We developed better metrics, improved the accuracy of statistical inference, and are progressing towards a unification with other comparison methods. Second, we found a technique for using deep neural networks trained to predict human behaviour in large datasets to guide the development of understandable models for the same task. We recently applied this to 4-in-a-row, a complex two player game and found interesting improvements to a state of the art model of human play. Third, we found a new theoretical explanation for reward prediction error neurons in the midbrain. We propose that these neurons encode rewards efficiently, which explains additional response properties beyond the subtraction of the mean that is predicted by the classical theories based on reinforcement learning. The efficient code could still be learned based on local learning rules and the new predictions are confirmed in existing data. Steps like these will gradually move us towards generally applicable models that capture human perception and cognition in detail.
Heiko Schütt is associate professor for computational cognitive science and modelling at the Université du Luxembourg since April 2023, where he develops models of human perception and cognition, focused primarily on visual perception and methods to test these models. These methods are often available as programming toolboxes and are widely used. Before moving to Luxembourg he was a PostDoc in New York working with Wei Ji Ma on developing models of human planning in complex games and of visual perception and with Nikolaus Kriegeskorte on representational similarity analysis. Before that he completed a PhD on models of early visual perception and eye movements in Tübingen under the supervision of Felix Wichmann.
Human learning and predictive processing depend on multiple cognitive systems related to dissociable brain structures. These systems interact not only in cooperative but sometimes competitive ways in optimizing performance. Previous studies showed that manipulations reducing the engagement of prefrontal lobe-mediated explicit, attentional processes can improve non-declarative learning performance. Here, we present four studies – non-invasive brain stimulation, functional brain connectivity, lifespan development, local sleep, and mind-wandering - in which we investigated the competitive relationship between perceptual statistical learning and prefrontal lobe-mediated executive functions. Our result sheds light on the competitive nature of brain systems in cognitive processes and could have important implications for developing new methods to improve human learning and predictive processing.
Dezső Németh is a full professor and team leader at the Neuroscience Research Center in Lyon (CRNL) at INSERM in France. He earned his Ph.D. in 2005 from ELTE University in Budapest. He was promoted to Assistant and Associate Professorship at the University of Szeged, Hungary. In 2008, he joined the Department of Neuroscience at Georgetown University in Washington DC, where he worked with Professor Michael Ullman on procedural memory and language. In 2012, he was a visiting professor in Professor Russel Poldrack’s Lab in the Brain Imaging Center at the University of Texas, Austin. Nemeth received the most prestigious neuroscience research grant in Hungary from the Hungarian Brain Research Program by the Hungarian Academy of Sciences in 2015. His research focused on the neurocognitive background of non-declarative memory and procedural learning. He also received the Award for Excellence in Teaching for outstanding teaching and student mentoring. Professor Nemeth has published over 80 peer-reviewed papers as a first, last, or corresponding author. He was awarded the full professorship in 2017 at ELTE University in Budapest. In 2018, he was awarded the prestigious IDEXLYON Fellowship (1.2M Euro) and moved to Lyon, France. In 2022, he was appointed to the prestigious Chair Professor position at INSERM in Lyon, France. He continues his research in the area of learning, memory consolidation, and predictive processing.
I argue that dreams can contain perceptual elements in multifarious, heretofore unthought of ways. I also explain the difference between dreams that contain perceptual elements, perceptual experiences that contain dream elements, and having a dream and a perceptual experience simultaneously. I then discuss two applications of the resulting view. First, I explain how my taxonomy of perception in dreams will allow “dream engineers” – who try to alter the content of people’s dreams – to accurately classify different dreams and explore creating new forms of perception in dreams. Second, I consider the consequences of the view for the role of memory in dreaming and imagination. I argue that not every element of dreams or sensory imaginations must rely on memory. The resultant view of sensory imagination provides a counterexample to Hume’s account of sensory imagination, according to which sensory imagination must be built up from faint copies of sensory impressions stored in memory.
Fiona Macpherson is Professor of Philosophy and Director of the Centre for the Study of Perceptual Experience at the University of Glasgow. Her research is on perception, illusion, hallucination, extremes of imagination (aphantasia and hyperphantasia), and virtual and augmented reality. She is President of the British Philosophical Association; a Trustee of the Kennedy Memorial Trust; a Fellow of the Royal Society of Edinburgh; a member of Academia Europaea; a member of the UKRI Creative Industries Advisory Group; and sits on the editorial board of The Philosophical Quarterly.