Intelligent Automation

31-12-2020

Alternate reading view (mobile friendly)

Holding a promise

Two months ago I posted the following message on LinkedIn:

To my surprise, the original authors responded and this post got around 10K views.

Now two months later, I have finally finished reading this book, with interleaved reading & writing of plenty academic research papers, which explains my tardiness.

With this blogpost, I will hold my promise to share feedback on Intelligent Automation.
One warning beforehand, this is my first (voluntary) non-fiction book review, so its quality might not do full justice to the reviewed book. The book did a marvelous attempt at structuring and mapping out higher-level trends in the Intelligent Automation landscape, with my book review I will try to approximate the same rigor.

This review is organized as follows: First, I will give a more objective summary of the book, introducing key concepts. Next, I present an overview of each of the 4 book chapters, structured in the format of summary/takeaways-comments. Finally, I will close the discussion with how I would advise to read this book, and who (in terms of job roles) would benefit most from reading it.

What makes automation intelligent?

Pascal Bornet, Ian Barkin and Jochen Wirtz have compiled a first reference book on Intelligent Automation (IA), alternatively hyperautomation, a term coined recently in 2017.

While having a lot in common with its vowel-swapped cousin Artificial Intelligence (AI), the authors clearly and necessarily differentiate between the two.

Intelligent Automation comprises a compelling class of technologies jointly capable of solving major world problems, when combined with people & organizations:

• A subset of AI for automation of knowledge work

• Robotic Process Automatic (RPA): the macro on steroids

• Workflow & Business Process Management (BPM)

• + people & organizations

IA allows for the creation of a software-based digital workforce, by mimicking four main human capabilities required to perform knowledge work:

1. Vision

2. Language

3. Thinking & Learning

4. Execution

Companies and workers can seriously benefit from implementing IA to build straight-through, requiring no manual intervention, business processes, which are more efficient (productivity, processing speed, cost) and often more effective (quality and logic).
Of course, depending on the complexity of the knowledge workers’ task to be automated, not all capabilities or technologies are always required. Alternatively, some tasks (the remaining 20%) might not lend themselves to automation at all, which is a nice thing, since IA only aims at taking the robot out of the human, not replacing human workers altogether.

Chapter by chapter

For each chapter in the book, I give an overview in the structure of summary/takeaways -comments.

Part 1: The promise of IA for a better world

IA cuts across many domains with foremost healthcare, education and business. Globally, more than 80% of the workforce are knowledge workers, which means the majority of humans can benefit from the support of IA. Certainly if we consider that most office workers are still working with “ancient tooling”, whose limitations are made more apparent with the recent COVID-19 crisis urging necessary digital transformations.

This chapter really convinced me of the value of this book, there are a plethora of references and statistics which have been meticulously collected.

Part 2: IA technologies explained

As explained before, IA comprises four key technologies and capabilities, for which this chapter gives detailed examples. Importantly, Natural Language Processing (NLP) is foundational to IA systems as more and more unstructured data is being created with organizations in dire need of structuring their data. While considering to automate complex end-to-end processes, it warrants the effort to build an IA roadmap where one prioritizes which capabilities should be implemented based on business needs.

This chapter was less interesting to me as an AI researcher, since there was not a lot of new material. Generally, it stays at an anecdotally high level, most probably by necessity, since each of these capabilities could have a whole book dedicated to them.

Part 3: How organizations succeed in implementing IA

Success at scaling IA transformations is not a given. The implementation of IA can fail due to gaps in management vision and support, improper change management, underutilized or unreliable data assets, technical limitations**, **high cost at possibly low efficiency, talent scarcity, and transformation complexity.
Then how can one succeed in IA transformations? Essentially, managing each of the above challenges, incorporated into a IA transformation roadmap, with appropriate prioritization & identification of IA potential, project preparation, and an ongoing effort of change & talent management will guide the way. The authors then take a more futuristic perspective as to how IA transformation roadmaps could look like in the future and what four levers will enable organisations to leverage more IA successfully.

This chapter presents the bulk of the book (90 pages vs. +- 60 per chapter) and distils the authors’ experience and expertise wonderfully. The first subchapters are introduced into a nice challenges–solutions format. In my view, these were more targeted towards project/innovation managers seeking to leverage IA within a larger company. I very much liked some initial quotes that “IA is a journey, not a project”, and that “IA is like medication to be administered to (sick) companies in small targeted doses”.

As an AI researcher I was most looking forward to the discussion on technical limitations, since removing these is supposedly my bread and butter. It was short, but to the point in that current IA/AI solutions are data-hungry and don’t do well in scenario’s they have not been given previous explicit supervision for. Further, they point to a symbiosis between humans and machines for handling technical limitations, in that machines can handle the regular bulk and pass on edge cases to humans for review. This co-bot perspective is something I very much believe in. However, this will require an advancement of AI technology to make their uncertainty or confidence assessment more reliable, which is exactly the scope of my PhD project.

Part 4: Reinventing society with IA

Success in IA should be focused on people, since they are at the heart of any successful IA transformation. IA is mostly thought of in a business context, but it can impact people’s lives positively and negatively. The authors argue for 5 imperatives that any company seeking to leverage IA should respect. Thinking ahead both optimistically and pessimistically, the authors present the future of work in a roadmap for our society. The first 2 imperatives, evolving skills (adapting education, redesigning job roles), and sharing wealth (Universal Basic Income for countering wage inequality) are required in an optimistic scenario. The last 3 imperatives rethinking work (“much of today’s work sucks”), reinventing education (education should aim to help people find their purpose in life, less of a work-obsessed culture stifling creativity), and building a new society (making governments plan for changes that IA will inevitably bring).

This chapter is splendid in that it balances both the positive and negative perspective on increasing automation. What we often might forget while working with technology is that it’s all for and about people. This chapter reminds us that IA transformations should be people-centric, with mindfulness about how it will affect the future of work. What I am very happy about is that the book ends with this hopeful note, instead of lingering too much on the dangerous side of AI and IA, which too often gets picked up in popular media. It nicely wraps up the promise of IA that was started in Part 1 with a look towards the future.

Bonus: IA use cases library

This final book section provides a treasure trove of new and budding ideas for automation in different functions of business: finance, procurement, HR, legal, marketing, sales, …; and industry: medical and life sciences, banking, government sector, telecommunications, insurance, …. For each use case, it explains the end-to-end process and which capabilities (from Part 2) are required to actionably automate it.

For example in business, (1) process accounts payable is a most interesting finance use-case, requiring all 4 capabilities (V, L, T&L, E). Particularly, the processing of vendor invoices or receipts into accounting systems might involve OCR on scanned documents (V), which in turn needs to be interpreted by NLP algorithms to extract essential entities for payment (total amount, IBAN, invoice date, invoice number, VAT details, …) and detail item lines (L). When not all information is present, the solution could query external databases to fill in remaining gaps or enrich extracted information (T&L). In this complete end-to-end process, exceptions need to be identified and routed towards human workflows for processing and approval (E).

For example in industry, (2) smart toilets [1] can help with daily health monitoring and even real-time drug dispensing (V, T&L, E). The below prototype by Stanford University medical school can give you an idea 😉

[1] Park, Seung-min, et al. "A mountable toilet system for personalized health monitoring via the analysis of excreta." Nature Biomedical Engineering 4.6 (2020): 624-635.

Closing remarks

Reading guide

INTELLIGENT AUTOMATION: Learn how to harness Artificial Intelligence to boost business and make our world more human

The book has a really well-sculpted structure and reads like a train! What I did notice during my reading is that it’s best to read each chapter in its entirety, since picking up where you left off might be hard to orient yourself again. One exception is chapter 3, which has a deeper nested structure where it is easy to loose the thread. To ensure you really grasp the knowledge from that chapter, I would advise to read the first two subchapters (pages 162-199) on challenges and success factors for IA transformations. Then let this sink in, before continuing (pages 200-251) on which factors will enable IA success in the future. While I understand they are grouped together, the latter could have been distilled in a separate chapter.

My views on Intelligent Automation after reading

To me, this reference book on IA provides a necessary answer to my personal question: Why are we pursuing advances in AI and when are we successfully advancing towards IA?

While the answer lies in a symbiosis of technologies, it is nonetheless a welcome reminder of what could be achievable ex vitro – out of the lab environment --.

From a business perspective, it is insightful to see how to move towards increasingly more advanced “self-driving enterprises”. While at the same time, we need to be mindful of the impact that automation brings and prepare for it accordingly.

For fellow AI researchers working on language/vision, this book is a must-read. It reminds ourselves there are challenges beyond incrementally pushing state-of-the-art on an academic benchmark –MNIST anyone?--. We cannot solely rely on industry to find solutions to applied AI, and in consequence IA challenges, since in fact those challenges are what stand in the way of the true value the technology can bring.

While we are still working away technical limitations of AI technology, its potential value is almost limitless when embedding into the mindset of IA. Algorithms should not be infallible, they should be reliable and where uncertain the “better” human can take over. The true challenge for embedding more advanced AI into IA then becomes keeping mistakes during automation low, the so-called false positives, where the algorithm is convinced it is right, yet in fact isn’t. Or as Mark Twain puts it:

“It ain’t what you don’t know that gets you into trouble. It’s what you know for sure that just ain’t so.” – Mark Twain

What’s next for this physical copy?

What might be nice, is that the physical book owner receives a read-only, secure, non-distributable digital copy.😊 I and most probably others would like to refer to some of the nice illustrations in this book.

This physical book will find a new home in the Contract.fit library, once we hopefully return to our physical offices in Brussels. But before this, it will do the tour in our company, next up our company’s co-founders Pol Brouckaert, and then Bertrand Anckaert!