Linha de Pesquisa: Paradigmas para a Pesquisa sobre o Ensino Científico e Tecnológico
Turma: DB (Dante Barone)
Código da Disciplina: PIE00099
Título da Disciplina: Ethics of Artificial Intelligence
Professor Responsável: Dante Augusto Couto Barone (Disciplina compartilhada com discente de Tópicos Especiais XVIII (Bach. em Ciência da Computação) – Ethics of Artificial Intelligence
Disciplina destinada a:
- Alunos Regulares do PPGIE: 15 vagas
- Outros PPGs da UFRGS: 15 vagas
- Alunos PEC: 5 vagas
Dia da semana: quartas-feiras
Horário: 8:30 – 12:10
Local: Instituto de Informática (sala a definir)
Créditos: 4
Carga-horária: 60 h/a
Introduction
“What Is Fairness?” Operationalized Fairness Criteria
Calibration & Fairness , Learning “Fair” Models
Individual Fairness, Bias in NLP , Fairness and Causality
Fairness Socio-Technical Analysis
“Trust Framework”
Fair Models via Constrained Optimization
Explainability and Transparency
Feature Attribution and LIME , Shapley Values & SHAP
Saliency Maps , Exemplar-Based Explanations
Concept-Based Explanations , Counterfactual Explanations
The Legal Perspective”
Explainability Socio-Technical Analysis
Privacy and ML
Learning with Differential privacy
Federated Learning
Privacy Socio-Technical Analysis
Socially Interactive AI: Challenges and Perspectives
Studies on Avatars and our Future Society
Automated Decision Making for Safety Critical Applications
The Quest for the Perfect Image Representation
AI in Fintechs: Challenges and Opportunities
ChatGPT: Characteristics and Ethical issues
Closing Discussion
Objetivos: Artificial intelligence (AI) is developing at an extremely rapid pace. We should expect to see significant changes in our society as AI systems become embedded in many aspects of our lives. This course will cover philosophical and practical issues raised by current and future AI systems. Questions we consider include: ⋅ How do we align the aims of autonomous AI systems with our own? ⋅ Does the future of AI pose an existential threat to humanity? ⋅ How do we prevent learning algorithms from acquiring morally objectionable biases? ⋅ Should autonomous AI be used to kill in warfare? ⋅ How should AI systems be embedded in our social relations? Is it permissible to fall in love with an AI system? ⋅ What sort of ethical rules should AI like a self-driving car use? ⋅ Can AI systems present moral harms? And if so, of what kinds? ⋅ Can AI systems be moral agents? If so, how should we hold them accountable? Besides this, it might be possible to analyze practical case studies,where students will identify issues of fairness, justice and truth in AI applications. The students will be encouraged to understand and master recent techniques to detect and mitigate such algorithmic biases, along with methods to provide more transparency and explainability to state-of-the-art ML models.
Método de trabalho (principais atividades): The discipline has a theoretical-practical nature as it includes both the exposition of theoretical contents by the professor, as well as practical work on the development of algorithms and programming related to ethical issues of Machine Learning systems.Student seminars, public presentation of developments and or analysis of ML implementation works and discussion of advanced topics are also planned.The assessment will be done covering mainly the key topics related to Ethics of Artificial Intelligence: algorithmic bias and fairness,explainability and interpretability,privacy and robustness and Interdisciplinary & applications.
Procedimentos e/ou critérios de avaliação:
Written tests, theoretical-practical work, seminars and program implementations and or analysis of codes using advanced ML and Deep Learning techniques.
The final grade, which will later be converted into a concept, is formed as follows:
A – Ethics of AI’ sTheoretical aspects through:
A1. Questionnaire and/or Papers analysis
A2. Pitch of a topic chosen related to the discipline
B – Ethics of AI’ s Practical aspects through:
B1. Design of a project to address ethical issues in a specific ML application field
B2. Pitch about the chosen project to address ethical issues in a specific ML application field
The Final Grade (NF) is calculated as follows:
NF = 0.2 A1 + 0.2 A2 + 0.2 B1 + 0.4 B2
The possible grades are A (score ≥ 9.0), B (score ≥ 7.5 and < 9.0), C (score ≥ 6.0 and < 7.5), D (failed, score < 6.0), or FF (failed, the student attended less than 75% of the classes).
Complementary practical work or recovery examinations covering all content of the discipline are offered to students, who have not attained at least C as final note.
Bibliografia sugerida:
- N. Bostrom and E. Yudkowsky. ‘The ethics of artificial intelligence’. In W. M. Ramsey and K. Frankish, editors, The Cambridge Handbook of Artiûcial Intelligence, pages 316–334. Cambridge University Press, Cambridge, 2014.
- Robert C. Bartlett and Susan D. Collins. Aristotle. Nicomachean Ethics, trans. University of Chicago Press, 2012. ISBN: 9780226026756.
- Solon Barrocas,Moritz Hardt,Arvind Narayanan. Fairness and Machine Learning: Limitations and Opportunities.Estados Unidos,2022. Disponível em :https://fairmlbook.org/