Administrative Information
| Title | Decision theory |
| Duration | 60 |
| Module | A |
| Lesson Type | Lecture |
| Focus | Technical - Foundations of AI |
| Topic | Foundations of AI |
Keywords
consequentialism, subjectivism, probability theory, utility theory, decision theory, optimal decision, bounded rationality, satisficing, cognitive bias, effective altruism, off-switch game, sequential decisions, value of information, multi-armed bandit, exploration-exploitation dilemma,
Learning Goals
- The learner gets acquainted with non-quantitative and quantitative approaches to ethics
- students can define and calculate optimal decisions using univariate distributions and utility/loss functions
- students can know the elements of decision networks: stochastic/action/utility nodes
- students can know and apply Bayes' rule (multivariate)
- students can construct a Naive Bayesian Network (NBN) model
- students can expand NBN to an NBN-based decision network
- students can calculate the value of perfect information
Expected Preparation
Learning Events to be Completed Before
Obligatory for Students
- Probability distribution, conditional probability, expected value (e.g from AIMA4e or wikipedia)
- Influence diagram
Optional for Students
- Artificial Intelligence: A Modern Approach, 4th Global ed. by Stuart Russell and Peter Norvig, Pearson (AIMA4e):ch12-18
References and background for students
- AIMA4e:ch12-18
Recommended for Teachers
- AIMA4e:ch12-18
- Naive Bayes Classifier
Instructions for Teachers
- Adapt the basic triple node decision net
- Find the best and worst decisions
- Adapt the naive Bayes net
- Find the most valuable evidence (feature)
- During in-class activities, ask students to guess optimal decisions
Outline/time schedule
| Duration | Description | Concepts | Activity | Material |
|---|---|---|---|---|
| 5 | Sources of uncertainty and Interpretations of probability | uncertainty | ||
| 5 | Bernoulli and multinomial distributions | univariate distributions | ||
| 5 | Axioms of probability theory (additivity) | probability theory | ||
| 5 | Elements and graphical notation of a single-step decision problem: the decision network of stochastic→utility/loss←action nodes | decision problem | ||
| 5 | Utility and loss functions, common loss functions and matrices | preferences | ||
| 5 | Expected value, the maximum expected utility principle | optimal decision | ||
| 5 | Conditional probability and Bayes' theorem (for two variables and with condition) | conditional probability | ||
| 5 | Independence and conditional independence | independence | ||
| 5 | Example of a Naive Bayesian network | Naive Bayes net | ||
| 5 | Example of decision network based on a Naive Bayesian network | |||
| 5 | Posterior inference and selection of optimal decision | posterior inference | ||
| 5 | Between evidence inference and calculation of the value of information | value of information |
Acknowledgements
The Human-Centered AI Masters programme was Co-Financed by the Connecting Europe Facility of the European Union Under Grant №CEF-TC-2020-1 Digital Skills 2020-EU-IA-0068.