Foreword

Author: Przemyslaw Biecek (Warsaw University of Technology and University of Warsaw)

0.1 Why?

Machine learning has a number of applications. Very often, however, machine learning predictive models are treated as black boxes which can be automatically trained without worrying about the domain in which they are used. This opaqueness rises many risks that are difficult to foresee during the model building process. Such as the model’s declining performance due to the data drift, poor performance on the out-of-domain problems or unfair biased behaviour learned on historical data.

The growing list of examples where black boxes fail spectacularly has led to an increased interest in XAI methods. Such methods allow to x-ray black boxes models for more detailed analysis on the local or global level.

According to Gartner Hype Cycle for Emerging Technologies in 2019 Explainable AI is on the verge of Innovation trigger and Peak of inflated expectations. It is a technology with a very high potential, which is talked about a lot in the media and which heats up the imagination as strongly as AI.

In the literature there are many articles arguing the need to use XAI methods as well as many ideas for new methods from the XAI family. However, it is much more difficult to find examples of successful implementations of XAI methods that have improved the business.

Missing elements are case studies of actual use of XAI methods in machine learning problems. Such case studies would allow a better understanding of what is possible today and what is not possible using XAI methods.

0.2 What?

This ebook collects examples of the use of different methods from the XAI family for different real-world predictive problems. In the following chapters, we show example applications of different XAI techniques to problems based on real-world public dataset.

These examples are called XAI stories and like every good story, each one has a structure. It starts with a description of the predictive problem, goes on to describe the proposed model or models. The models are x-rays using XAI techniques to finish the chapter with a point.

0.3 How?

For XAI stories to be credible they need not only a strong predictive model but also business validation of the proposed modeling and an explanation approach.

Each group of students was assigned two mentors from data scientists and experts within McKinsey Digital: a consultant and a data scientist. The mentors, together with the students, searched for the strengths and weaknesses of XAI applications in specific problems. At McKinsey Digital, we help our clients create change that matters—transformation, enabled by technology and sustained through capabilities. We drive transformation and build businesses by bringing together the capabilities needed to help organizations grow and thrive in the digital age. We help our clients harness the power of data and artificial intelligence, modernize core technology and capitalize on new technology, optimize and automate operations, fuel digital growth, create stunning digital experiences, and build digital talent and culture.

0.4 About academic partners

The Faculty of Mathematics and Information Science, Warsaw University of Technology

The Faculty of Mathematics and Information Science offers studies in data science at the level of the bachelor degree, master degree and doctoral studies.

The study program includes a powerful dose of mathematics and programming.

This book was prepared as part of the Interpretable Machine Learning 2019/2020 elective course. Find more details about MiNI classes, here and about the faculty, here.

The Faculty of Mathematics, Informatics, and Mechanics, University of Warsaw

The Faculty of Mathematics, Informatics, and Mechanics Department offers a master’s degree with a specialization in mathematical statistics or a specialization in machine learning. The curriculum includes many interesting subjects related to computational statistics or deep learning.

This book was prepared as part of the Interpretable Machine Learning 2019/2020 elective course. Find more details about the MIM classes here and about the faculty, here.

0.5 About business partner

McKinsey & Company around the world

McKinsey & Company is a global management consulting firm committed to helping organizations create Change that Matters. In more than 130 cities and 65 countries, our teams help clients across the private, public and social sectors shape bold strategies and transform the way they work, embed technology where it unlocks value, and build capabilities to sustain the change. Not just any change, but Change that Matters –for their organizations, their people, and in turn society at large.

McKinsey & Company in Poland

The Polish office of McKinsey & Company first opened its doors more than a quarter of a century ago. Since then we have become the largest strategic consulting firm in Poland, employing more than 1,500 people. We advise Poland’s biggest companies and public institutions and have played a part in the transformation of key enterprises, contributing to the development of companies that today are leaders in banking and insurance, consumer goods, energy, oil, telecommunications, mining and many other sectors. In total we have carried out almost 1,000 projects for our Polish clients. In 2010 we opened our Polish Knowledge Center in Wrocław, where we currently employ almost 250 top analysts. A year later we established a Shared Services Center in Poznań, where over a thousand of our people now work. A McKinsey Digital Lab has been operating in the Warsaw Office since 2017. Our developers, experts in Big Data and business consultants support companies undergoing comprehensive digital transformation and perform advanced data analytics. For more information, visit www.mckinsey.pl.

McKinsey Analytics

McKinsey & Company perform many engagements with strong analytics components. Our data scientists and architects, together with our machine learning and data engineers, complement our strategic and operational consulting and provide our clients with advanced and robust data-driven solutions. For more information, visit our Polish website and our global website.

Explainable Artificial Intelligence at McKinsey & Company

The importance of Explainable Artificial Intelligence (XAI) methods was recognized at McKinsey & Company from the beginning. Whenever needed, XAI methods are extensively employed in our analytical engagements. Moreover, McKinsey & Company contributes to the prevalence of XAI methods with its numerous publications and podcasts on this topic; to name but a few: