Embedded Files
Weekly Learning Goals
Weekly Learning Goals
Understand how biases emerge both from training data and AI algorithms;
Understand the sources and implications of algorithmic (heuristic) bias;
Explore Narratives related to critical societal processes such as public safety and professional development.
Reading
In the first two weeks of our course, we laid the foundation for AI Ethics. In the first week, we discussed that AI is basically a predictive model based on training data or first principles that captures patterns to make inferences. In the second week, we explored how certain principles derived from an ethical tradition stemming from the Universal Declaration of Human Rights and Bioethics provide a conceptual framework for organizing the main ethical issues related to the impact of AI on the world.
Starting this week, we will begin to delve into specific areas of ethical issues related to AI, in line with the strategies of Floridi and the Alan Turing Institute. This first week, we will address issues related to fairness and bias mitigation. In this first reading of the introductory chapter of Joy Boulamwini's excellent book Unmasking AI,
Consider While Reading
What is algorithmic bias? Are there other types of biases?
If bias is a pre-existent social/human problem, what AI has to do with it?
How do the Facial Recognition examples connect to our first Unit final project?
Exploring Narratives
AI Biases
AI Biases
This week's narratives come in the form of a podcast conversation between Tristan Harris and Dr. Joy Buolamwini, who "argues that algorithmic bias in AI systems poses risks to marginalized people." This conversation revisits some of the points she raised in the chapter we read and prepares us for our in-class discussions.
On a perhaps interesting side note, I had the pleasure of talking to Dr. Joy and Bowdoin students on last Fall and I was impressed by her insightfulness, simplicity, and humanity.
Tuesday Class
Tuesday Class
Reading (for Thursday)
This week's second reading is a little longer, but extremely important, and I hope you will find it very insightful and thought-provoking. It sheds light on bias problems, explaining how they manifest in machine learning algorithms.
Consider While Reading
What are the causes of bias in machine learning?
What trade-offs need to be addressed?
What are the best solutions for algorithmic bias problems?
Addition Reading II (OPTIONAL)
In this second optional reading block, I invite you to read two chapters of Meredith Broussard's book called "More than a Glitch: Confronting Race, Gender, and Ability Bias in Tech." The entire book is fascinating, but the first two chapters provide an overview of how biases are created in artificial intelligence systems and share narratives about the real and terrible implications of this kind of problem. In a way, these chapters are a miniature version of the structure of our course because they provide an initial technical view, use narratives to highlight ethical problems, and discuss solutions from various perspectives.
Consider While Reading
What is data bias? Are there other types of biases?
If bias is a pre-existent social/human problem, what AI has to do with it?
How do the Facial Recognition examples connect to our first Unit final project?
Page updated
Report abuse