Lecturer: Constantin Rothkopf
Fields: Cognitive Science, Artificial Intelligence
Content
Recent advances in artificial intelligence based on deep learning have led to the discovery of new medical drugs, the development of new materials, and the optimization of fusion reactor designs. However, claims about fundamental limitations persist: unpredictable blunders, limited robustness, and lack of explainability. The talk will present recent examples and studies contributing to the current debate on what current AI systems can do and what they cannot yet do. A central topic will be how to leverage Cognitive Science to understand the properties of such AI systems. The systems discussed include large language models, neural network models of economic decision-making, visual-language foundation models and the considered tasks range from the classic Bongard problems to sensorimotor control and planning under uncertainty to deontological ethical judgments. Topics will cover the anthropomorphization of AI systems, problems of data contamination and bias, Clever-Hans phenomena, inherent limitations of benchmarks, and fundamental limitations of evaluations and comparisons in terms of performance measures of behavior.
Literature
- Schramowski, P., Turan, C., Andersen, N., Rothkopf, C. A., & Kersting, K. (2022). Large pre-trained language models contain human-like biases of what is right and wrong to do. Nature Machine Intelligence, 4(3), 258-268.
- Binz, M., & Schulz, E. (2023). Using cognitive psychology to understand GPT-3. Proceedings of the National Academy of Sciences, 120(6), e2218523120.
- Mitchell, M. (2023). How do we know how smart AI systems are? Science, 381(6654), eadj5957.
- Valmeekam, K., Marquez, M., Sreedharan, S., & Kambhampati, S. (2023). On the planning abilities of large language models-a critical investigation. Advances in Neural Information Processing Systems, 36, 75993-76005.
- Thomas, T., Straub, D., Tatai, F., Shene, M., Tosik, T., Kersting, K., & Rothkopf, C. A. (2024). Modelling dataset bias in machine-learned theories of economic decision-making. Nature Human Behaviour, 8(4), 679-691.
- McCoy, R. T., Yao, S., Friedman, D., Hardy, M. D., & Griffiths, T. L. (2024). Embers of autoregression show how large language models are shaped by the problem they are trained to solve. Proceedings of the National Academy of Sciences, 121(41), e2322420121.
- Wüst, A., Tobiasch, T., Helff, L. , Singh Dhami, D., Rothkopf, C. A., Kersting, K. (2024). Bongard in wonderland: visual puzzles that still make AI go mad? Sys2-Reasoning, Neurips Workshops.
Lecturer
Constantin Rothkopf is a full professor (W3) at the Institute of Psychology in the Department of Human Sciences with a secondary appointment in the Department of Computer Science at the Technical University of Darmstadt. He is the founding director of the Center for Cognitive Science and a founding member as well as a member of the executive board of the Hessian Center for Artificial Intelligence (hessian.AI). He is also a member of the board of directors of the Center for Mind, Brain and Behavior (CMBB). He is a member of the European Laboratory for Learning and Intelligent Systems (ELLIS), a faculty member of the ELLIS Unit Darmstadt, and a member of the DAAD Konrad Zuse Schools of Excellence in Artificial Intelligence (ELIZA). He is currently co-speaker of the collaborative projects The Adaptive Mind and Whitebox. After obtaining a joint PhD in Brain & Cognitive Sciences and Computer Science at the Center for Visual Science at the University of Rochester, NY in 2009, he started a postdoc at the Frankfurt Institute for Advanced Studies (FIAS) working in the theoretical neuroscience group. In 2009 he started as a lecturer at the Goethe University, Frankfurt, and from 2010 to 2012 he was the principal investigator of the “beliefs, representations, and actions group” at FIAS. After a year as a substitute professor at the Institute of Cognitive Science at the University Osnabrück, he started as an associate professor for “psychology of information processing” at the Institute of Psychology at the Technical University of Darmstadt in 2013. During the winter semester 2017 he was a visiting professor at the Department of Cognitive Science at the Central European University, Budapest. In 2022 he received an ERC Consolidator Grant from the European Research Council for his project ‘ACTOR’. During the summer semester 2023 he was a visiting professor at the Zuckerman Institute, Columbia University, New York, USA.
Affiliation: TU Darmstadt
Homepage: https://www.pip.tu-darmstadt.de