Alex John London
Clara L. West Professor of Ethics and Philosophy
Carnegie Mellon University School of Computer Science Executive and Professional Education
Skip to Main Content
Responsible Artificial Intelligence
In this course, we explore the concept of responsible artificial intelligence, misconceptions surrounding it, and its recent rapid growth and development. We will discuss the benefits and potential applications of responsible AI and the ethical and social concerns surrounding its use. We will also discuss the importance of ensuring that AI is developed and used in a way that is fair, transparent and accountable.
This seven-week, online, self-paced course complements online content with live weekly office hours and live Q&A sessions with learning facilitators.
Modules: Introduction to Responsible AI, Robustness, Explainability, Privacy, Fairness and Bias, Responsible Generative AI and Large Language Models
In each module, we define a specific concept in the context of responsible AI systems, and discuss its benefits and limitations. We discuss how to account for this factor as part of responsible AI development, important considerations and common misconceptions for each, and recent developments in this space.
Prerequisites, Intended Audience and Size
This course has been designed for leaders, managers and executives who work on AI products or with AI product teams. The course does not assume any background knowledge of AI and there are no technical prerequisites.
We recommend a cohort size of 30 students. Custom cohorts can be launched with short notice.
Course Structure
Module 1: Introduction to Responsible AI
Upon completion of the module, participants will be able to:
- Define artificial intelligence.
- Describe and discuss common misconceptions about artificial intelligence.
- Define responsible artificial intelligence (RAI).
- State the common characteristics of a responsible AI system.
- Recognize that stakeholders need to attend to the goals of responsible AI at the various stages of system development or in the process of product procurement.
- Give an overview of the considerations relevant to RAI at each stage of development.
- Illustrate the importance of accountability by ensuring that each of the tasks associated with RAI is delegated to identifiable individuals or groups.
Module 2: Robustness
Upon completion of the module, participants will be able to:
- Define robustness in the context of AI systems.
- Explain why it is important for AI systems to be robust.
- Guide implementation of new AI systems that are robust.
- Critique existing AI systems developed by their organization on the basis of robustness.
- Evaluate and appraise AI products and systems developed by other organizations for robustness.
- Discuss organization-level policies for robustness.
- Develop a testing plan to measure robustness.
- State common misconceptions and red flags that indicate a lack of robustness in an AI system.
Module 3: Explainability
Upon completion of the module, participants will be able to:
- Define explainability and interpretability in the context of AI systems.
- Explain why it is important for AI systems to be explainable or interpretable.
- Discuss how explainability can help with trust and accountability.
- Guide implementation of new AI systems that are inherently explainable.
- Critique existing AI systems developed by their organization on the basis of explainability.
- Evaluate and appraise AI products and systems developed by other organizations for explainability.
- Discuss organization-level policies for explainability.
- Develop a testing plan to measure explainability.
- State common misconceptions and red flags related to explainability in an AI system.
Module 4: Privacy
Upon completion of the module, participants will be able to:
- Define the concept of privacy, its different dimensions, and how they apply in the context of AI and machine learning (ML) systems.
- Evaluate the design of different AI and ML systems in light of different interpretations of FIPPs such as those taken by regulations such as GDPR or CCPA/CPRA.
- Propose design modifications to AI/ML systems that comply with different interpretations of FIPPs.
- Apply the concept and methodologies of privacy by design, including data minimization principles and the LINDDUN framework, in the context of AI and ML solutions/systems.
- Demonstrate an understanding of privacy-enhancing technologies such as differential privacy and federated learning.
Module 5: Fairness and Bias
Upon completion of the module, participants will be able to:
- Define fairness and bias in the context of AI systems.
- Explain why it is important to consider fairness and bias while building systems.
- Discuss the different levels of fairness.
- Identify potential sources of bias in an AI system.
- Guide implementation of new AI systems that are inherently fair and where biases have been considered.
- Critique existing AI systems developed by their organization on the basis of fairness and bias.
- Evaluate and appraise AI products and systems developed by other organizations for fairness- and bias-related issues.
- Discuss organization-level policies and training regarding fairness and bias.
- State common misconceptions and red flags related to fairness and bias in an AI system.
Module 6: Connecting the Concepts
- This module will connect the concepts discussed throughout this course and will tie them together.
- Participants will receive actionable suggestions on how to implement the concepts from this course for their own use cases.
Module 7: Responsible Generative AI and Large Language Models
Upon completion of the module, participants will be able to:
- Define Generative AI and Large Language Models (LLMs)
- Explain the uses and abilities of Generative AI
- Critically assess and articulate the risks and ethical concerns associated with LLMs
- Describe how different techniques and tools can accurately identify AI-generated text
- Examine and interpret the copyright implications of using LLMs
- Design and recommend guidelines for the ethical and responsible use of LLMs in different contexts