Lecture 1 — Knowledge I: Can inference be automated? Intentional systems; problem-solving by search and heuristics; Monte Carlo Tree Search (MCTS); deep reinforcement learning (as motivation). Automata, formal languages, and logic; syntax and semantics; truth versus provability. Limit: the limits of formal proof 

Lecture 2 — Knowledge II: Can reasoning under uncertainty be automated? Uncertainty and the axiomatic derivation of probability theory from the coherence of belief. Probabilistic graphical models: naïve-Bayes models, Hidden Markov Models (HMMs), Bayesian networks, and stochastic context-free grammars. Independence model and independence map. The Markov boundary. Observational equivalence. Limit: exact probabilistic inference is computationally hard (NP-hard). 

Lecture 3 — Knowledge III: Can decision under preferences be automated. Utilities and loss functions; utility theory; decision theory; human cognitive biases; the value of information. Stochastic inference and Markov Chain Monte Carlo (MCMC), the laws of large numbers including Bayesian and dependent settings. Limit: Optimal decision. Approximate probabilistic inference is computationally hard (NP-hard). 

Lecture 4 — Knowledge IV: Can reasoning about causes be automated? The causal ladder: causal inference, the do-calculus, and counterfactual reasoning. Limit: conditional-independence relations admit no finite, complete axiomatization (an incompleteness result), and finding a minimal independence map is NP-hard. 

Lecture 5 — Induction I: Can inductive inference about observations be automated? Bayesian learning as inference: Bayesian Model Averaging (BMA). The prequential approach. The Minimum Description Length (MDL) principle and universal induction. The maximum a posteriori and the maximum likelihood principles. The relationship between optimization and Bayesian inference. The bias–variance decomposition. Complexity regularization. Bayesian Information Criterion. Limit: reference prior, Lindley-paradoxon 

Lecture 6 — Induction I: theoretical foundations. Universal consistency. Probably Approximately Correct (PAC) learning. Learning curves and sample complexity.  Computational learning theory. Transfer learning; learning with priors and informative priors; theory refinement; the fusion of data and prior knowledge; knowledge-based neural networks; multi-task and multi-modal learning.  Limit: in general, the Bayes error cannot be learned to a guaranteed accuracy. 

Lecture 7 — Induction II: Sequential learning. Multi-armed bandits, the exploration-exploitation dilemma. Active learning. Monte Carlo Tree Search. Reinforcement learning and deep reinforcement learning (sequential decision-making).  

Lecture 8 — Generative AI I: structural causal models. Learning generative models as Causal Bayesian Networks (CBNs). Learning from a mixture of observational and interventional data. Learning from incomplete data across the spectrum from Missing Completely At Random (MCAR) through Missing At Random (MAR) to Missing Not At Random (MNAR). Explanation generation. 

Lecture 9 — Generative AI II: transformers. The lineage of sequence models in natural-language processing, from Recurrent Neural Networks (RNNs) and Long Short-Term Memory (LSTM) networks to transformer models. The transformer architecture; inference and steering (controlled generation). Learning and fine-tuning. Distillation. 

Lecture 10 — Interpreting transformer models. Explainable artificial intelligence; neural probing; mechanistic interpretation; counterfactual analysis; circuit analysis. Machine teaching using model editing. Machine unlearning. 

Lecture 11 — Foundation models, agentic AI, and the AI ecosystem. Foundation models as the integrative substrate that connects domain-specific systems. Transfer learning