Andrei Rogobete: “The Ethical Algorithm” by Michael Kearns & Aaron Roth

The Royal Academy of Engineering predicts that algorithms or Artificial Intelligence (AI), will become prevalent in “…most, if not all, aspects of decision-making” (April 2017). Algorithms benefit from a growing presence in key areas such as government, healthcare, education, financial industries, and of course, technology. Yet this is for good reason: they are highly efficient and effective, certainly far more than the average person. This is particularly true when dealing with large amounts of data where algorithms simplify complexities and present them in a more digestible, applicable form. In sectors such healthcare, algorithms quite literally save lives.

However, increasingly complex and penetrative algorithms can lead to some less desirable outcomes. They may lack nuance to changing scenarios, they may be unable to deal with subjectivity, or in certain cases their output may appear brutish.

This raises serious ethical and moral questions: can a degree of morality or ethical values be implemented within algorithms? Can they be made to reflect societal views and norms?  Can we even agree on any particular set of ‘common values’?  If so, how or to what degree might they be implemented within the structure of algorithms? These are just a few of the broader questions raised by “The Ethical Algorithm: the science of socially aware algorithm design” by Michael Kearns and Aaron Roth.

The authors bring together a comprehensive list of credentials. Dr Kearns has spent his career in the field of computer science and worked with AT&T Bell Labs where he was appointed head of the AI department. Dr Kearns is currently a Professor at the University of Pennsylvania and Chair of the National Center within the Department of Computer and Information Sciences. Dr Routh also has a background in computer science and is currently Associate Professor at the University of Pennsylvania. His research interests include Data Privacy, Game Theory, Machine Learning, and Algorithms.

The aim of the book is to dive “…headfirst into the emerging science of designing social constraints directly into algorithms, and the consequences and trade-offs that emerge” (page 16). More specifically, the book argues that in order to, “…make informed decisions we need to be able to understand the consequences of deploying certain kinds of algorithms and the costs associated with constraining them in various ways” – whilst acknowledging that technology alone may not be able to “…solve complicated social problems” (ibid).

An intriguing truth is that we are the data (page 2). We are not just users of data but through our activity we become creators of data. More importantly, the data in turn is being used to make decisions about us and sometimes, as the book points out, these can be very “consequential” decisions.

The structure of the book is devised in six chapters and the language is aimed at a non-specialist audience. However, there some “technical “parts that may require more of the reader’s attention and time. This will likely cause prospective readers to go through certain sections or chapters at a differing pace.

Chapters I and II look at issues surrounding privacy and the concept of ‘fairness’ within algorithmic design. There is an interesting reflection on how Netflix has used movie preferences of users to potentially reveal highly sensitive information such as sexual orientation, political affiliation and personal interests (pages 24-26). It demonstrates how an initial privacy agreement can rather quickly escalate into a much greater issue with significant consequences.

Chapters III and IV further the discussion by attempting to analyse various social outcomes of algorithmic design. This section also considers some of the shortcomings of the scientific data on algorithms. The internal structures of common navigation apps such as Google Maps and Waze are discussed and it becomes quite intriguing to discover how they can coordinate traffic around congested areas to yield the best possible outcomes in terms of time and distance of travel. For instance, the ‘Maxwell Solution’ (page 105) poses an algorithmic conundrum: the quickest route of each individual driver is not congruent with the quickest route for all drivers collectively.

Chapter V and the Conclusions reflect on the societal and ethical implications. The authors themselves recognise that algorithms are playing an increasing role in people’s lives. In new technologies such as autonomous transportation, healthcare, or defence, there may be decisions that “…we never want algorithms to make, period – even if they make them ‘better’ than humans” (page 176). It is argued for instance that the decision to kill another human being should never be taken solely by an algorithm (page 178).

“The Ethical Algorithm” by Michael Kearns & Aaron Roth is recommended for anyone with an interest in technology and the ethical implications of our increasing use of algorithms. That is not to say that it is flawless – one can sense that at various points throughout the book the authors become overly zealous in viewing the world exclusively through a computer science lens. Everything becomes a problem that computer science can fix (or at least try).

We must be realistic that there is a fine line between automation and human input. Managed poorly, it can result in catastrophes like the crash of two brand new Boeing 737-Max 8 jets. We all want to increase efficiency but what is the exact cost of losing the ‘human touch’?  Indeed, can we even agree on exact “societal norms” and “values”? Not only are they constantly evolving concepts, but they are also highly subjective in parts. Does this mean that certain algorithms would have to be continuously updated to “reflect” society’s shared values? And who is to determine what these values are? These are questions that require careful consideration and thorough answers. It is for our benefit because one thing is for sure: algorithms will play an increasing role in the public and private spheres.

 

“The Ethical Algorithm” by Michael Kearns & Aaron Roth was published in 2019 by Oxford University Press, (ISBN 0190948205, 9780190948207), 232 pp.


Andrei Rogobete

Andrei E. Rogobete is the Associate Director of  the Centre for Enterprise, Markets & Ethics. For more information about Andrei please click here.