- Level: L3
- Semester: SI6
- Duration: 12h
- Language: Fr
- Teacher: Pierre Parrend
Summary
The use of so-called “artificial intelligence” algorithms in sensitive environments requires “trusted AI”. Trusted AI involves two elements: the ethical use of algorithms, and control over their results, in particular through explainability. The goal of this course is to provide methodological tools and case studies to learn how to develop trusted AI and control its properties.
Objectives
After completing this module, students will be able to:
- Identify application domains that require trusted artificial intelligence
- Characterise the level of explainability of an algorithm
- Use and evaluate learning algorithms
- Develop their own tree-based ensemble learning algorithm
- Use data analysis development environments (notebooks)
Lecture outline
- AI and trust
- The challenges: how to trust AI
- The failure of uncontrolled AI: the Tay bot example
- Explainability and AI
- Evaluating trust
- Use cases
- Explainability of tree-based algorithms
- Tree-based learning
- Ensemble tree-based learning
- Evaluating learning
- Practical work
- Evaluating learning
- Ensemble tree-based learning
- Explainability and adversarial exploratory attacks
References
- Guide pratique pour des IA éthiques, Numeum, septembre 2021 http://ai-ethical.com/wp-content/uploads/2021/09/2021-SN-Guide-Me%CC%81thodo-IA-Ethiques-version-imprime%CC%81e.pdf
- Arrieta, Alejandro Barredo, et al. “Explainable Artificial Intelligence (XAI): Concepts, taxonomies, opportunities and challenges toward responsible AI.” Information fusion 58 (2020)
- Explainable Artificial Intelligence (XAI), DARPA-BAA-16-53, August 10, 2016
- Bostrom, Nick, and Eliezer Yudkowsky. “The ethics of artificial intelligence.” The Cambridge handbook of artificial intelligence 1 (2014): 316-334.