Название: Real World AI Ethics for Data Scientists: Practical Case Studies
Автор: Nachshon (Sean) Goltz, Tracey Dowdeswell
Издательство: CRC Press
Серия: Data Science Series
Год: 2023
Страниц: 143
Язык: английский
Формат: pdf (true)
Размер: 10.1 MB

In the midst of the fourth industrial revolution, big data is weighed in gold, placing enormous power in the hands of data scientists – the modern AI alchemists. But great power comes with greater responsibility. This book seeks to shape, in a practical, diverse, and inclusive way, the ethical compass of those entrusted with Big Data.

Being practical, this book provides seven real-world case studies dealing with Big Data abuse. These cases span a range of topics from the statistical manipulation of research in the Cornell food lab through the Facebook user data abuse done by Cambridge Analytica to the abuse of farm animals by AI in a chapter co-authored by renowned philosophers Peter Singer and Yip Fai Tse. Diverse and inclusive, given the global nature of this revolution, this book provides case-by-case commentary on the cases by scholars representing non-Western ethical approaches (Buddhist, Jewish, Indigenous, and African) as well as Western approaches (consequentialism, deontology, and virtue).

In this book, we and the commentators refer at times to Artificial Intelligence (AI), ML, Data Science, and computer-assisted decision-making. The case studies and commentaries described in this book are intended to apply to each of these, overlapping, fields. Data Science is broadly “a set of principles, problem definitions, algorithms, and processes for extracting non-obvious and useful patterns from large data sets.” AI, on the other hand, can be defined generally as any type of Artificial Information-processing that carries out a psychological function – such as “perception, association, prediction, planning, motor control” – that up until now has been performed only by living beings. There have traditionally been two basic approaches to developing AI: Good Old-Fashioned AI (GOFAI) and ML. GOFAI, sometimes also called symbolic AI, takes a top-down view of intelligence.

ML is a subset of Data Science and a large and growing part of the field. Unlike GOFAI, ML is a form of AI that uses a statistical, rather than symbolic, approach to find patterns in a messy world of ambiguity.9 ML was in many ways a response to the early failures of symbolic AI, which quickly broke down outside of the controlled laboratory environments – unable to process the complexities of real-world situations. ML emphasises prediction – as opposed to the field of statistics, which emphasises associations and explanations.

ML algorithms are trained with the given successive layers of information, presented one at a time, and they can learn in ways that are relatively unsupervised, and relatively mysterious – much more like how our human brains learn. Supervised learning involves algorithms learning to use a function and applying it to a set of data – such as a function describing ‘spam’ being applied to a set of e-mails to filter out which ones are likely to be spam. The attributes of ‘spam’ and ‘not spam’ will be labelled for the algorithm by human beings, who already know the difference.13 Unsupervised learning, on the other hand, involves no labelling – indeed, we may not even know what qualities we are looking for. The algorithm first examines a set of data and figures out what the relevant attributes are before it applies them to new data. An unsupervised algorithm might, for example, look at numerous images of dogs and determine what set of attributes gives rise to the essence of ‘dogness.’ When presented with any new picture, it will then decide if it is or is not a dog.

Data Science tools have become more user-friendly, and this has opened the field to many new entrants – even those with little training. This means that “it has never been easier to do data science badly,” and the unethical consequences of badly designed or executed projects can and should be anticipated by those whose task is to design and deploy those systems.

We hope this book will be a lighthouse for those debating ethical dilemmas in this challenging and ever-evolving field.