The Ethical Algorithm | Michael Kearns

Summary of: The Ethical Algorithm: The Science of Socially Aware Algorithm Design
By: Michael Kearns

Introduction

In a world where data is constantly being collected and analyzed, the ethics of algorithm design become highly critical. In the book ‘The Ethical Algorithm: The Science of Socially Aware Algorithm Design’ by Michael Kearns, we explore the intricacies of creating algorithms that are fair, accurate, transparent, and ethical. Through various examples, Kearns delves into the challenges of designing systems that prioritize the FATE (fairness, accuracy, transparency, and ethics) principle while still maintaining efficiency. Get ready to explore the fascinating world of algorithm design, balancing privacy concerns, addressing biases, and discovering the applications of game theory in this insightful summary.

Ethical Algorithms

Companies and governments gather data using algorithms to provide services, but it can also lead to divulging private information and unfair decisions. Some algorithms use step-by-step instructions, while others use machine learning, making it difficult for humans to understand their internal models. Developing algorithms using the FATE principle, which involves fairness, accuracy, transparency, and ethics ensures ethical use. However, such implementation can reduce efficiency as compared to a design that does not follow this principle. Thus, regulations are necessary to ensure that systems are used according to their purpose.

The Illusion of Anonymity

Anonymizing data does not guarantee privacy as it is still possible to trace information about individuals. To protect privacy, algorithms should aim to minimize harm to individuals. This can be done through differential privacy. While effective, this method increases the amount of data needed to draw reliable conclusions. Google, Apple, and the US Census Bureau use differential privacy to protect data. However, it is not foolproof as it cannot protect information belonging to large groups. The desire of many people to know their DNA has made it possible to find family members of a serial killer and the Golden State Killer himself.

Biased Algorithms

Algorithms are not immune to bias, especially when trained on biased data. A Google research team found evidence of sexism in their own data. Similarly, Amazon’s machine learning algorithm had a gender bias in selecting job applicants. It’s difficult to prevent algorithms from producing biased output, and the problem is exacerbated when the data itself is biased. However, there are ways to make algorithms more fair, including explicitly demanding neutrality on certain criteria. Trade-offs exist between different definitions of fairness. While an algorithm may be fair for several large groups, it may still be unfair to a well-defined small one. Furthermore, algorithms that modify themselves in a game setting can guarantee fairness for a group but cannot completely eliminate bias. Ultimately, it is up to humans to make the trade-off for the desired level of fairness and choose the appropriate fairness type.

Algorithm meets Game Theory

The book explores the intersection of game theory with algorithms, which are increasingly shaping our daily lives. From dating apps, route-finding apps like Waze, shopping preferences, social media, to donor organ allocation: game theory can help design better algorithms for such situations. The book emphasizes that the challenge is not just creating an algorithm, but also how people respond to and interact with them. Game theory can help in developing algorithms that adhere to specified ethical standards.

Want to read the full book summary?

Leave a Reply

Your email address will not be published. Required fields are marked *

Fill out this field
Fill out this field
Please enter a valid email address.
You need to agree with the terms to proceed