Posted on: Tuesday 15th of January 2019
I have a simple metric for how good I think a book is – the number of text highlights in my Kindle. The three top scorers in the data and digital category for 2018 are Digital Ape: how to live (in peace) with smart machines by Sir Nigel Shadbolt and Roger Hampson (45 highlights), Digital Human: the fourth revolution of humanity includes everyone by Chris Skinner (65 highlights) and in front by some way Prediction Machines: the simple economics of artificial intelligence by Ajay Agrawal, Joshua Gans and Avi Goldfarb (185 highlights).
I should point out there is some algorithmic bias even in this simple metric favouring Prediction Machines. At the end of each chapter it has a Key Points summary which I religiously highlighted (just in case I had missed anything in the chapter itself) so it has benefited from some double counting. But even adjusting for this, it is the comfortable winner.
I read Digital Ape and Digital Human in quick succession over the summer. Not surprisingly given the similarity in their names, they cover some of the same ground – how digitisation is changing our lives. But their dissimilarities are more informative.
Shadbolt and Hampson have their roots in academia and a track record in public service. Also, both have a strong focus on data (Shadbolt co-founded the Open Data Institute). Digital Ape focuses on the societal challenges that the continuing increase in computer processing is creating. Skinner is a financial services commentator, previously having been a technology practitioner in the banking sector for many years. His target audience is those people attempting to transform legacy businesses into digital ones.
Digital Ape begins with the story of our evolution, told from the viewpoint of our interactions with the tools that we have created. (The detail is such that you would expect one of the writers to have been an academic anthropologist.) The authors’ conclusion is that our tools have shaped us as much as we have shaped them. Given the power of the devices that we now have out our disposal, the social implications are huge. ‘The digital ape has created a new virtual universe that is still expanding by powers of 10. We now need imaginative help to understand the scale of what we have built and what it might forebode.’
As the authors point out, tools extend our capabilities. But when our tools are accelerating in power and increasing in complexity at a far greater rate than our cognitive competencies, it seems only a matter of time until we lose control and wreak unintended havoc. The risk they highlight is not some form of artificial super-intelligence taking control but from natural stupidity and arrogance, meaning we lose control of ever more pervasive and powerful machines. ‘The danger, in the financial world, is that immensely fast machines can be programmed by greedy fools, not properly supervised by the authorities, to do immensely silly things on an enormous scale in an immensely small interval of time.’
Despite this caution, overall they are optimistic that our new tools will augment our minds more than they will empower our suppressed biological urges, notably the desire to enrich ourselves as much as possible. They see data as delivering huge benefits when it is free to flow to where it can be of most value to the individual, highlighting the potential of Sir Tim Berners-Lee’s Solid project at MIT which is aiming to create a decentralised internet that delivers true data ownership to individuals and improved privacy.
The sub-title of Digital Human is ‘the fourth revolution of humanity includes everyone.’ It also covers the first three revolutions, but far quicker than Ape. As a technologist, Skinner is clearly excited by the dramatic possibilities that advances in technology can bring, highlighting how in many instances science fiction has become science fact. There are multiple references to Star Trek and he believes space travel will be as revolutionary to how we live in the future as air travel has been over the past 50 years.
The long-term predictions add colour but Digital Human’s real strength lies in the changes wrought over the last twenty years and what that suggests for the next ten. He is enthused by the potential for FinTechs but equally recognises the strengths that incumbent financial institutions have, pointing out that no start-up has yet to overthrow an incumbent. He equates technologists predicting disintermediation to the little boy who cried wolf, with every cry so far having been proved false. (When Liz Brandt, CEO of Ctrl-Shift, was working in banking in the early 1990s, there was a fear of disintermediation by the supermarkets – banking’s disintermediation wolf predates digitisation.)
Instead, Skinner sees increasing partnerships between start-ups and incumbents, with FinTechs providing widgets of capability in a new business architecture where back offices manufacture products and services, middle offices process payments and transactions and front offices retail intimacy and experiences. In technology terms, the core capabilities are analytics in the back office, APIs in the middle office and smart apps in the front office.
His central argument is that banks need to evolve from monolithic, vertically integrated, physically focused organisations to micro-service, open market, digitally focused structures in the next ten years to survive and thrive. But CEOs do not want to take the risk of dealing with their legacy challenges during their relatively short time in charge, so death through inertia is the most likely outcome. “The business model of banks was built for face-to-face interactions backed up by paper documentation; the business model of digital banks is for device-to-device interactions backed up by data. The two are completely different.”
Skinner sees a rapidly emerging world where banks do not make money from the services they currently provide and need to generate new revenue streams derived from a far deeper understanding of customers than they do today. But if there is a gap in Digital Human it is a fully developed customer angle. Customers don’t want a mortgage – or any other financial product for that matter – they want a new home, a new car, to afford to go on holiday. But banks focus on the product they are offering, not the outcome consumers want. This is why there is fear about FinTechs because they deliver a better product purchasing experience than the incumbents.
But when you start to plot the end-to-end new home journey from recognising the need to move to moving in, what becomes apparent is that the biggest blockers that customers face are information challenges. Where can I afford to live? What would my commute be like from this area? What are the schools like? How far am I away from a shopping district? Or a park where I can walk the dog? What broadband provider is available? These challenges aren’t solved by product-providing FinTechs but by what Ctrl-Shift has always called PIMS – Personal Information Management Services.
None of which negates his core argument that banks of the future should be comprised of micro-services linked by data. Indeed it just strengthens it by broadening the purpose of those micro-services to solving customers’ information challenges, in return for which banks will gain loads of insights into their intentions and priorities. And it gives a great place for incumbents to start the opening up of their platforms – they don’t have to introduce competing services that will cannibalise revenue, they can introduce complementary ones that improve the overall customer experience by helping to solve the customer’s real problem, not just the small part of the problem that they have traditionally solved.
Which leaves me with Prediction Machines. Agrawal, Gans, and Goldfarb are professors at the University of Toronto’s Rotman School of Management. Agrawal also founded the Creative Destruction Lab which is an accelerator for AI businesses. As economists, their take on AI is very different to the majority of books on the subject which revel in the technical aspects that excite data scientists. Agrawal and his co-authors demystify AI, distilling its importance it to one critical component of intelligence – prediction.
Recent advances in AI are reducing the costs of making accurate predictions in the same way that the advent of the first computers reduced the cost of doing arithmetic. ‘The advent and commercialisation of computers made arithmetic cheap. When arithmetic became cheap, not only did we use more of it for traditional applications of arithmetic, but we also used the newly cheap arithmetic applications that were not traditionally associated with arithmetic, like music.’ Challenges that were not originally framed as being a maths problems became so.
With falling costs of prediction, the same is happening thanks to AI. The example the authors cite is the airport lounge which exists because we are poor at predicting how long it will take to get to and through the airport. As the costs of missing a flight are far greater than the cost of time spent at the airport, the majority of us are risk-averse and arrive too early and so need to kill time. As prediction becomes cheaper and more widespread, the need for us to leave so early will reduce and the usage of lounges will decline in parallel.
The problem the falling cost of prediction helps with is uncertainty. This is a pet hobby horse of mine for a couple of reasons. Firstly because the traditional approach to strategy development created by the large consultancy firms fails to adequately deal with uncertainty. Secondly, because uncertainty doesn’t just exist in the future, it exists in the present (does someone have cancer or not) and in the past (why did something happen) as much as the future. Trying to understand causality is not something businesses spend much time on. Possibly because causality has traditionally been defined by the opinion of senior management, so a more scientific approach might prove embarrassing). The authors acknowledge the presence of uncertainty across time dimensions, but do not develop the impact of better understanding cause and effect as much as they might have done.
The falling cost of prediction raises the value of other skills, particularly judgement. Achieving an outcome requires four components – data, prediction, judgement, and action. (all of which will be familiar to anyone who has worked on personalisation or automation projects as those four components have their mirror in the solution being implemented). Judgement has traditionally been a human skill, but as decision technology becomes more sophisticated, it is also being increasingly automated at the operational level, though not at the strategic. How prediction upends the trade-offs associated with strategic decision making is one of the book’s strengths, as is their evaluation of the different risks the spread of AI brings.
Yet their overall tone is generally optimistic. As economists, they take comfort from how employment has survived waves of supposedly job-destroying technological innovation, albeit with some disruption in the short term. Using a simple economic prism – creating a country called Robotlandia and working out whether we would seek to trade with it or not – they explain how we can co-exist comfortably with increasing automation. But they also worry about how the profits from lower prediction costs will be distributed.
If you are a non-data scientist senior executive seeking to understand the impact AI will have on your business, this is the one book on AI you should read. If you are a data scientist, you should probably read it as well – it will help you reframe your enthusiasm for the possibilities AI brings in a way that will excite your less data-oriented colleagues.
Jack Springman, Head of Consulting, Ctrl-Shift