Back

Responsible AI – Fundamentals, Policy, and Implementation

94-885

Units: 6

Description

As artificial intelligence technologies advance, so do the complexities and risks that organizations and policymakers
must monitor, in order to adhere with ethical principles. This course will cover several desired properties for
Responsible AI systems to embody, going beyond accuracy and reliability, to include interpretability, privacy,
security, safety, and sustainability. Along with this, we will discuss strategies, pitfalls, and limitations for corporations
aiming to identify and mitigate risks, such as organizational commitments to transparency via artifacts like model
cards. We will also cover governance frameworks aiming to enforce responsibility and prevent societal harms,
including the NIST AI Risk Management Framework (RMF), the EU AI Act, and several U.S. laws. This course will
be in a seminar/discussion format, and the use of case studies and discussion of real-world incidents will reinforce
the challenges in translating the nuances of ethical commitments to measurable organizational practice or
enforceable legal policy. Throughout the course, several types of AI systems will be discussed (spanning traditional
issues of fairness in predictive machine learning models, to recent high profile security issues or jailbreaks in
generative AI), as well as systems used in several domains of application. There are no formal prerequisites for this
course, but rigorous critical analysis, thoughtful communication, and reflection on integrity will be expected, as this is
a graduate level course at CMU.

Learning Outcomes

By the end of this course, students will be able to:
1. Lead discussions demonstrating depth of understanding on core risks in AI (ie bias, transparency, privacy,
safety, security, sustainability), across several types of AI systems, and communicate the gravity of these
risks using examples from real incidents.
2. Identify and mitigate AI risks at various points of the software lifecycle using current strategies, and critique
the capabilities and limitations of risk management practices.
3. Analyze existing regulatory mechanisms (including internal corporate governance strategies, US policies,
international policies), and the challenges and tradeoffs present in these regulations.
4. Design new governance proposals for AI systems, with consideration towards various stakeholders’ needs
and contradictory goals
5. Communicate reflections, reasoning on the anticipated future of Responsible AI needs, and future-proof
policies and practices for emerging threats.

Prerequisites Description

There are no formal prerequisites for this
course, but rigorous critical analysis, thoughtful communication, and reflection on integrity will be expected, as this is
a graduate level course at CMU.