Value-Alignment problem. Building AI with human values FIL-SM>Przwzfn-c
1. Problem and context of the AI value alignment problem.
2. Machine Learning – problems with bias
3. Big Data, learning and representation
4. Fairness in AI, what it is and what it is not.
5. Facing the impossible in big data
6. Transparency of learning in AI
7. Deep Networks and learning human values
8. Agency and ML
9. Normativity and learning human values: imitation, inference, uncertainty
10. Unsafe AI systems, existential risks, and risks of misaligned AI
11. Honest AI, Emergent goals and complexity, Embedded AI agency
12. Dynamics of AI alignment, unpredictability and unwanted side effects.
(in Polish) Tryb zajęć
Course coordinators
Learning outcomes
Knowledge:
A student will gain an understanding of the concept of AI Value alignment problem, its philosophical background and assumptions.
Skills:
A student can explain what is AI Value alignment problem,, discuss the foundational assumptions behind the concept and critically review current discussions around the problem of value alignment (E_2)
A student will learn how to write and abstract for the conference, create presentation and present it to the audience.
Attitudes and transferrable (generic) competencies:
The student will be able to explain the concept of AI Value alignment problem,, its assumptions and its relations to concept of beneficial AI, human centered AI and AGI (E_3)
Assessment criteria
• (W_1) grading on the bases of the presentation to the class and a submitted two abstracts 300 words each on the topic of the lecture- 50% of grade
• Take-home exam – short-answer questions -20% of grade
• Penalty for the late work – 15% of the grade
• Student's activity during the class as well the discussion of the paper may raise one's grade. - 15% of grade
• Attendance is mandatory.
Practical placement
none
Additional information
Additional information (registration calendar, class conductors, localization and schedules of classes), might be available in the USOSweb system: