Because of its interdisciplinary character, the variety of techniques used in artificial intelligence is considerable. In this course an overview is provided in three modules which focus on different aspects of artificial intelligence: techniques from logic and linguistics (module 1), from computer science (module 2) and from cognitive neuroscience (module 3). Reasoning being the general theme, the course shows what forms this can take in the different areas, ranging from idealization (module 1) to computation (module 2) to experiments (module 3).
In the first module the students will study the formal aspects of reasoning in artificial intelligence. After an introduction describing the emergence, through the ages, of formal reasoning in philosophy and the sciences, students will be introduced to various formal systems and methods of proof, such as natural deduction, sequent calculi, Hilbert-style systems and display calculi. It will be shown how different views on reasoning can be captured in these proof systems. As an example, intuitionistic logic, linear logic as used in linguistics and fuzzy logic will be discussed. The type theoretic Curry-Howard isomorphism that connects proofs to programs will be treated. It will be shown how one can in a precise way capture various aspects of reasoning, such as its complexity, by the structure and size of proofs. The relation to famous open problems in computer science is explained. At the end of this module students will be able to construct proofs in the various proof systems and to translate proofs from one format to another. They understand and can use the mentioned logics and understand the different views on reasoning underlying them. They understand the proofs-as-programs paradigm and know what a normal or cut-free proof is.
The second module covers various foundational techniques that are used for the development of intelligent systems in general and artificial agents in particular. We begin by refreshing the memory of the student by giving a crash course into modal logic. We begin with the general framework (syntax and Kripke-style semantics) and review applications such as epistemic logic, doxastic logic, temporal logic, dynamic logic and deontic logic. Also so-called minimal model modal logic with neighbourhood semantics with as application coalition logic. Students will do exercises with these standard techniques to be properly prepared for the other courses in the curriculum, in particular Intelligent Agents. Following the modal logic part, we focus on multi-agent programming techniques that can be used to implement multi-agent systems. We present a multi-agent programming language, present its operational semantics, and explain how its properties can be analysed by means of modal logic. The students will work with some programming exercise to master the use of the programming language. The final part of this module discusses probabilistic techniques for multi-agent learning, such as conditional expectation, Markov chains, Markov reward chains, decision reward chains, Markov decision processes (MDPs), stochastic games (multiple-player MDP's).
In the third module, students will be introduced to current methods in cognitive brain research with examples taken from vision and language research using classroom lectures and practicals. This module will cover the entire spectrum of skills and techniques available to the cognitive neuroscience community, including neurophysiological research methods (such as fMRI and EEG), psychophysics, experimental design, modeling and basics data analyses. The practicals will give students hands-on experience with a number of techniques to create experiments in vision and/or language, acquire data and analyze the results from these experiments using SPSS and Matlab.