Early Indications May 2019: The Next Infrastructure

If we look ahead into the medium term, to 2025 or 2030, there could be some subtle  but significant changes to US infrastructure to cope with the implications of several  emerging technologies. Many of these may be emerging in quiet locations, but if one  knows where to look, the landscape might be changing before our eyes. Let’s start with telecommunications, much of which is a matter of access. When you  look at the world’s nations ranked by land mass, it’s easy to neglect how vast Russia  is, even downsized from the former USSR: it’s twice as big as the #5 nation (Brazil).  Also, I was surprised to see that both Canada and the US are larger than China.  India, despite its massive population, is only 1/5 as large as Russia and about 1/3  the size of the US. The US total is deceiving: about 400,000 square kilometers are  water, and Alaska counts for about 1/6 of the #3-ranking total.  Why does this matter, apart from trivia night? Connecting people into the Internet is  expensive, difficult, and complicated. Rural broadband is emerging as a hot political  topic here in Pennsylvania: the state defines “broadband” as a measly 1.4 Mbps (the  FCC definition is 25), and many people can’t get that except by expensive and  unreliable satellite coverage. By way of comparison, Finland has established  broadband access as a right of citizenship, with the goal of getting every citizen  access to _at least_ a 100 Mbps connection by 2025.  A few hundred miles to Pennsylvania’s southwest, Kentucky shares our  prevalence of sparsely populated areas featuring rugged terrain. A statewide project  known as KentuckyWired that launched in late 2014 aimed to address the digital  divide: Kentucky is fifth worst in the US in high-speed Internet use and not  coincidentally fifth poorest as measured by median household income. To fund the  buildout, the state engaged an Australian bank with a long history of infrastructure  projects. In the five years since the deal was signed, construction progress is far  behind schedule and the complexity of the debt covenants is kicking in: the state  auditor estimates KentuckyWired will now cost $1.5 billion over the next 30 years,  or 50 times the original proposal. (ProPublica’s reporting in conjunction with the  Louisville Journal-Courier is detailed and dispiriting. ) This situation is common in the US: Take away the metroplexes (146 densely  populated counties) that are home to half of the population, and the other half is  scattered across the remaining 95% of counties. Cell coverage can be spotty in  these locales, which makes the prospect of 5G wireless broadband unlikely, given  the expensive necessity of more cells for the smaller coverage footprint of the new  technology. Whether for wired or wireless access, something as seemingly simple  as a telephone pole can be legally complicated, financially costly, and logistically  tricky to maintain: at our local airport I’ve seen helicopters that lower saws in flight  to cut away brush and branches from otherwise inaccessible wires. Absent Internet connectivity, millions of people can’t see doctors, access learning  materials, or work remotely. As robotics, telemedia, and Internet of Things  applications increase in importance, these citizens will fall still farther behind their  better-connected counterparts, both in the US and abroad. At some point, focused  effort on a national scale will be required to update the communications infrastructure.  This will happen as bridges and roads continue to crumble, and as the generation of  federal, state, and local workers who began to join public payrolls in the 1960s  continues to retire. Oh, and Social Security is due to run out of money in 2034.  Funding broadband will not be a simple matter in any state. Let’s turn from flyover zones to cities. Here, the combination of crushing population  density, demographic trends, and technology change will bring a very different set of  scenarios. Given the recent IPOs of Lyft and Uber, let’s start with cars. It’s currently  unclear what autonomous vehicles will do for traffic: some studies predict  increased delays while others foresee improvement. In any event, it would seem  intuitive to predict that the need for parking will drop, to take only one aspect of change.  The coexistence of cars (autonomous or human-driven) and bicycles will require new  urban designs, assumptions, and attitudes; infrastructure cannot be built in isolation,  as New York would seem to prove. No city in the US appears to be rethinking the role  of cars in the city core, apart from a congestion pricing experiment about to launch  in New York. I was in a meeting earlier this month where logistics experts discussed the future of  food. Meal delivery appears to be resonating with large audiences, which raises the  issue of where all that food will be prepared. Fast-food outlets are in much the same  situation as Macy’s stores attempting to implement omnichannel logistics: neither the  retailer nor the food franchises were built to support delivery of product to customers  off the store premises. Kitchens at McDonalds are already coping with more menu  options including salads and parfaits along with all-day breakfast; where and how  should UberEats and Grubhub customers be serviced in the existing kitchen and  cashier architecture? What should the new design seek to optimize? Some restaurants are pulling out of the delivery business, or are opting for branded  drivers rather than contractors who may or may not represent the restaurant with its  desired courtesy, promptness, and attention to detail. Dominos has long wrestled with  the dilemma of drivers who speed or otherwise poorly represent the company, and  there are no easy answers. The UberEats driver is a contractor to a contractor, so  allegiance is a fluid proposition: most Lyft drivers are also signed up with Uber, and  UberEats delivers for many restaurants. There’s a lot of room for competing priorities  in such a labor pool. Assuming the driver quality issue can be addressed, possibly by better pay, the kitchen  situation remains problematic. Around the world, multiple startups are launching  so-called ghost or cloud kitchens. These kitchens are commissaries that serve multiple  ethnic formats from a common infrastructure that has no visible presence or dining (or  carryout) area. Uber’s co-founder Travis Kalanick is involved with a firm called  Cloudkitchens that only services meal delivery firms. SoftBank’s Vision Fund has  invested $357 million in Zume Pizza, which uses robotics to prepare pizza for  delivery en route to the customer. On a smaller scale but already in market, Good Uncle is a food delivery startup  aimed at college students. Like Zume, it cooks pizzas en route to the customer,  which improves both speed and quality, but the service is also working on an  aggregated delivery model: rather than stop at every house or apartment that  has ordered food, it delivers at visible, logical campus locations. Right now the  service is operating in Bethlehem, PA, College Park, MD, and Hamilton and  Syracuse, NY. The talent behind the food includes chefs from Michelin- tar restaurants, and the service is in some cases replacing the student dining hall.  Good Uncle delivers groceries as well as cooked meals, does not add delivery  charges, and offers daily discounts, presumably to cut down on food waste of  slow-selling perishable items. Speaking of delivery, infrastructure will need to evolve to address the Amazon  issue of how to manage the last 50 feet. For apartment buildings without a  doorman or similar attendant, for free-standing houses with visible porches,  and for gig workers who might operate out of a Starbucks or other temporary  space, the delivery model (especially for frequently ordered items such as groceries)  doesn’t scale: UPS can’t keep coming back day after day if nobody will ever be  home during business hours. At the local FedEx ground location where I pick up  wine shipments that require an in-person signature, I was told that mine was one  of up to 50 such shipments per day; the holding room for packages sometimes  overflows into the main store. As autonomous vehicles reset the economics of automobile ownership, what happens  to tens of millions of US garages? At the same time, where will urban dwellers without  a garage charge electric cars at night? Will parking meters need to be rewired to include  high-voltage chargers? As the nature of the car and its relation to the city change,  everything from carry-out windows to parking lots to domestic architecture will change  in response. Will garage-door openers selectively allow access to deliveries, and will  refrigerators become commonplace in these post-garage spaces? The days of the  foyer, ice-delivery boxes, and milk boxes on porches may be circling back.  Will we see more use of e-bicycles that function like small pickup trucks? If so,  how will laws, norms, and space co-evolve? What will constitute a “bike lane”  for such chunky vehicles? All told, changes in connectivity, whether 5G or rural broadband, will bring with  them implications for how people get around, especially in relation to what the tech  analyst Horace Dediu calls “micromobility” centered around scooters, bicycles, and  related technologies. As old as the bicycle may be, it remains a wonder of technological  efficiency and with changes to batteries, motors, GPS, and other emerging technologies,  the mid-21st century could mark a new blossoming of interest and innovation. If Dediu is  right, and I believe he is, then many parts of the world will look very different (read: more  like Amsterdam) in a decade or two.

Early Indications April 2019: Review essay: Deep Medicine by Eric Topol

Note: I neglected last month to announce that MIT Press has published my book on 3D printing,  and that an article on some managerial implications of same came out in the Journal of  Organizational Design.  ***** Eric Topol is a cardiologist who is very visible on the front lines of the medical field as it intersects  with smartphones, distributed sensors, and other emerging technologies. His most recent book  looks at "how artificial intelligence can make healthcare human again," in the words of the subtitle.  While I have plenty of honest disagreements with Topol's book, I very much respect it, and hope  that it jump-starts some way-overdue discussions about the future of medicine, particularly in the US. Given that I don't work in medicine, I'm extrapolating my lay experience here. The phrase  "personalized medicine" conjures up, for me at least, micro-targeted pharmaceuticals: my dose of  heart, anti-depressant, or chemotherapy medication will be fit precisely to my weight, age,  metabolism, and holistic health picture. Topol makes it clear that I am off base here. First, time  -- of day, of season, of life -- matters a lot: precision dosing entails understanding time-series  data in entirely new ways.  Second, precision medicine may likely begin not with pharmaceuticals but with diet. Recent  research on the role of gut flora and related participants in digestion helps explain why nutritional  research lacks any consistent consensus: meat was good, then bad, now it's sometimes good.  Butter was normal, then evil, but now it's not nearly as bad as margarine. Sugar was portrayed as  benign (so was cigarette smoking, for the same reasons), but now the "a calorie is just a calorie"  fiction is being exposed. One reason that dietary advice is so unreliable is that the degree of  uniqueness in our digestion and metabolization patterns is only now coming into focus. A key factor in that biomedical research into the gut biome and related processes is a massive  increase in the size of the data sets being researched. Much as with Facebook photos or Google  searches, machine learning algorithms need enormous volumes of training data to become reliable.  If for no other reason, Topol's book is valuable for chapter 11, where he explains the rapidly  changing literature of diet and digestion research. One point is made quite emphatically in the book, though I don't recall the author saying so directly.  Given the vast number of AI-infused startups for medical applications, we are nowhere remotely  close to general AI: every company he discusses, some of them in depth and/or with great  enthusiasm, is a point solution. Diabetes-related readings are the specialty of one company, while  mammograms are graded somewhere else, and retinopathy is diagnosed somewhere else. In short,  AI will not replace doctors anytime soon. This is not to say, however, that physicians' jobs won't change. Certain tasks -- image recognition  in particular -- are done quite capably by machines. Note, however, that image recognition is not  disease diagnosis (aided by up-to-date knowledge of a vast literature), treatment planning, end-of- life counseling, arguing with insurance companies, or any of the dozens of other things doctors are  called upon to do. The challenge ahead will be to team doctors and computers in the most effective  possible proportions, letting each contributor provide inputs that will lead to the whole patient  experience amounting to more than the sum of its parts. Topol also provides useful context for the role of AI in employment: there are plenty of subspecialties  in the US that are far understaffed, and globally, many nations have staggeringly few doctors. AI  provides the quite realistic possibility, in dermatology care for example, of enhancing family doctors  and nurse practitioners as they address one condition among many they see in a routine workweek.  Up-skilling generalists is entirely different from threatening the livelihoods of the relatively small  number of specialists called upon to address the major issue of deadly melanomas in the U.S.  and elsewhere.  Compared to a Watson-like digestion of medical research (something that seems plausible, if not  yet realized), for machine learning algorithms to read patient records remains a daunting task. Even  with electronic medical records, the lack of standardization in people's lives, descriptions of their  symptoms, health professionals' acuity in observation and notation, interoperability between  institutions (despite potentially using the same software package), and, again, the time of  observation means that mining EMRs is for practical purposes impossible for the foreseeable  future. In the meantime, EMRs are linked to physician burnout and decreased patient satisfaction  -- doctors navigating screens and typing with their backs to the patient in the exam room are  hugely troublesome for all concerned. Speech recognition powered by machine learning could  help alleviate this issue, and Topol ends the book by imagining a "super-Siri" personal health  assistant that helps each of us live our best life, medically speaking, through an advanced form  of what the military calls data fusion. For all the many drawbacks of current EMRs and other technologies, however, it's time to  address the elephant in the room: managed care. To the extent that my genomic, diagnostic,  behavioral, and other data becomes a tool for prediction of my health (and by extension, the  cost to maintain it), there will be a fundamental conflict of interest between me and my insurer.  Topol nods at this issue but never engages it (he is, however, a consultant to some of those  insurers). As the old saying goes, "follow the money," so my conviction is that the initial large- scale deployments of AI in health care will not be to reduce diabetic retinitis or arthrofibrosis,  but rather in a reimbursement arms race between insurers and providers. It's entirely possibly  that both Aetna and a health network will have deployed IBM Watson systems as the two  parties dispute a billed transaction, so we may see algorithms at war with themselves. It seems far fetched, but perhaps the EMR + AI convergence will be the beginning of the end  of the U.S.'s unique model of health care, in which more is spent per patient than anywhere  in the world while most major outcomes rank well down the list. That is, the fundamental conflict  of interest noted above will lead to such complex statutes, regulations, and/or case law that the  current U.S. arrangements become unsustainable. Already far too much effort is expended in  paperwork that does nothing for patient care, and the implications of predictive algorithm-driven  medicine for privacy, doctor-patient confidentiality, and ethics make HIPAA look positively  antiquated and inadequate.  If Topol can envision a new "virtual medical coach," perhaps the American Medical Association  can help construct a 21st-century alternative to its previous objections to single-payer health  care. Such a design for medicine would acknowledge the new roles of genomic science, data  analytics, robotics, and machine learning, along with addressing end-of-life, opioid, and other  political hot-button issues. I'm not convinced AI can give physicians time and tools to become  "human again" (beware anyone selling a golden age fallacy), but medicine absolutely needs  to become individualized, beginning from a data foundation, for the first time.

Early Indications March 2019: Facebook Act 2?

On March 7 Facebook founder, CEO, and majority shareholder (for voting purposes) Mark Zuckerberg announced “A Privacy-Focused Vision for Social Networking.”  You might want to read it, if only to see that what he does NOT say is blindingly obvious.

There have been many commentaries on this missive, but I want instead to discuss a post from last April by Andreessen Horowitz’s Benedict Evans, called “The Death of the Newsfeed.”

There are many dots to connect here, with rather staggering implications. Let us begin.

1) Zuckerberg states that people value privacy, so Facebook will build on its WhatsApp franchise and offer end-to-end encrypted 1:1 chat: “I believe the future of communication will increasingly shift to private, encrypted services where people can be confident what they say to each other stays secure and their messages and content won't stick around forever. This is the future I hope we will help bring about.“ 

But wait: bad people can discuss and plan bad things in private, so Facebook has 

“a responsibility to work with law enforcement and to help prevent these wherever we can. We are working to improve our ability to identify and stop bad actors across our apps by detecting patterns of activity or through other means, even when we can't see the content of the messages, and we will continue to invest in this work. But we face an inherent tradeoff because we will never find all of the potential harm we do today when our security systems can see the messages themselves.” 

Whoa. Either communications is private or it is not.

2) Facebook stopped many copies of the New Zealand massacre video from being reposted, it is true, but there were still hundreds of thousands of copies that were circulated and of course recirculated. Facebook should not be congratulating itself on how much evil it has prevented any time soon. Recall that AT&T shared its bulk network traffic with US legal/intelligence sources via the infamous “Room 641A” in San Francisco. How will Facebook provide similar courtesy/compliance? Will Indian, or Iranian, or Russian governments get the same access? Whatever Facebook says about privacy, how much will educated observers believe? To whom is Facebook held accountable?

3) Even though Mr. Zuckerberg is the only Facebook shareholder whose vote counts, he has a fiduciary duty to the other investors. Not once does he mention what happens to ad revenue in these encrypted channels. If it goes away, does Facebook attempt to become the AT&T of the 21st century? If Facebook users stop being the product sold to advertisers, they likely become customers of a very different kind of service, one for which I predict there is limited paying appetite. 

4) Why is Facebook re-examining privacy given its steadfast dismissal of same? The Benedict Evans post from last year provided a useful prediction, in nearly Haiku form: 

"All social apps grow until you need a newsfeed

All newsfeeds grow until you need an algorithmic feed

All algorithmic feeds grow until you get fed up of not seeing stuff/seeing the wrong stuff & leave for new apps with less overload

All those new apps grow until..."

The massive scale of Facebook means the average user (a dubious notion, but I’m merely quoting their figure) is eligible to see at least 1,500 posts per day. If published figures are true — I’ve seen extrapolations of 35 minutes, 27 minutes, 41 minutes — that means a user would have to click once per second for 30 solid minutes merely to stay current with the newsfeed (30 minutes x 50 clicks/minute).

Enter the machine learning solution: we can algorithmically determine what people want to see out of that impossible pool of 1500 eligible items. Alas, algorithms can be gamed, or stuffed (with echoes of the SEO arms race), or lag user sentiment, which can be expected to evolve over time. Hence the “pivot toward privacy” can be read at one level as an admission of defeat: machine learning cannot curate the newsfeed model indefinitely (or even temporarily).

5) This is an important moment. As industries from CPG to defense to pharma to transportation seek out consultancies and new hires to help embrace “big data” and “AI” (whatever that is in operational terms), an acknowledged leader pivots away from its core AI-driven product. The fact that Facebook is seeing an exaggerated departure of top executives of late underscores how much this matters: having made their first (or at least most recent) x million dollars inside Facebook, these people who are in a position to know are betting they can make their next x million _outside_ the social network leader.

6) Building on messaging is not Zuckerberg's idea. Chinese services such as WeChat, Viber (popular in former Soviet satellite states), and LINE (Japan/Taiwan) can include payments, gaming, GPS, and, yes, encryption. Consultants including Forrester have been predicting the rise of messaging for about five years. One must give Facebook credit for its acquisitions: Instagram and WhatsApp have proven to be brilliant buys. But can/will they cannibalize the original product?

The larger question involves revisiting Clayton Christensen's Innovator's Dilemma from 20 years ago: can Facebook shift its reliance from the revenue-rich but overbloated and mistrusted newsfeed toward lighter-weight and potentially ad-resistent messaging tools? If so, how will Wall Street value the company? Can Facebook leadership persuade investors, regulators, and users (both advertisers and consumers of the service) of the wisdom of moving with the platforms to new definitions of relevance, convenience, and trust? What happens to the future of news media, given Facebook's role in the current crisis of an informed body politic? Whatever happens next, this feels like the beginning of the end of an era.

Early Indications January 2019: The organizational context for machine learning

Much attention is paid to the necessity for innovation, typically at the level of a technology: a molecule, a machine, a circuit. While these types of innovations matter, they cannot succeed without institutional “scaffolding,” as it were, the organizational infrastructure that gets the molecule made into pills and dispensed through pharmacies or gets the electric car assembled, distributed, recharged, and maintained over its entire product life. We posit that this organizational infrastructure will be a critical factor in the adoption of a powerful new technology in the coming decades.

We are at the cusp of a period of broad adoption of machine learning in more and more technical, professional, and commercial fields. While the labor shortage of data scientists and employees with related skills is well documented, less appreciated is the need for the institutional scaffolding mentioned above. Three factors have contributed to the boom in machine learning since 2012 or so: larger scales of training data sets, improved algorithms, and broader availability of powerful computing capabilities, often through (or at) cloud providers including Google and Amazon. The need now is for business models, institutional processes, and deep functional knowledge to be coupled with the technical advances.

This has long been the pattern of technological innovation. Powerful steam engines drove the need for bigger, stronger ocean vessels; retrofitting schooners and other wooden sail-powered craft with the new engines was insufficient. The rise of the automobile would have stalled without consumer credit, adequate paved roads, and a network of gas stations. AT&T’s Picturephone never achieved market acceptance as a landline device, but video calling apps such as Skype and FaceTime are extremely popular as smartphone add-ons. Technological innovation without organizational infrastructure rarely succeeds.

Within an organization, this cultural and organizational scaffolding is often invisible, making it difficult to change. A classic example is provided by the transition from steam and water power to electric motors in textile mills and similar facilities in the late 19th/early 20th century. Historically, water wheels and then steam engines powered overhead drive shafts that ran the length of the facility. Individual machines were thus located in relation to this overhead power source. Electric motors were initially used to drive the same shafts, slowing adoption: the case for new investment was weak. It took roughly 30 years for “group drive” to be replaced by “unit drive” in which individual machines each had their own electric motor(s) on board. Cost savings accrued as unused machines no longer were driven by the centralized power source, energy efficiency was improved with the removal of slippage in the belt-driven transmission system, and variable-speed motors allowed for each machine to run at its optimal RPM. Productivity improved dramatically as factory engineers repositioned machines, now freed from the proximity to the drive shaft, to facilitate workflow. In addition, removing the drive belts and overhead shafts allowed for the installation of overhead cranes, improving productivity further.

As machine earning is brought into mature organizations (as opposed to algorithm-centric companies such as Google or Facebook), managers will likely be tempted to overlay the new technology onto existing business practices. In the US, cellular phones were initially treated as mobile versions of stationary voice devices. In countries without a well-developed voice-device network (most of the world), such practices as texting took off much faster: the capabilities of the new technology were unconstrained by previous preconceptions and habits. More recently, 3D printing is often compared to mass-production molding and milling processes with which managers are familiar, and not surprisingly, 3D printing is found to make little business sense for long runs of simple geometries.

For machine learning to be applied to problems and processes where it can add distinctive and unprecedented value, it will need to be understood in the context of a particular market, a particular business model, and a particular business process. This level of understanding requires a rare combination of deep functional expertise, an open mind, adequate historic data for training the algorithms, and technical acumen. (Many efforts we have seen focus only on the latter aspect.) Weather forecasting — where IBM can claim deep expertise — has been reshaped by new algorithms and models (including ensembles) over the past 20 years, but medical reasoning has proven much more difficult to automate. Ad placement has been transformed by machine learning; production planning remains variable, volatile, and largely manual.

What kinds of scenarios are we seeing as machine learning is adopted in legacy business processes?

1) Unquestioned assumptions can be challenged by a naive algorithm

Because many more variables can be correlated across much larger samples than humans can cognitively manage, seemingly random inputs can be found to drive outcomes. At one hotel booking site, photos of guest rooms with the window shades open drove higher sales, but only in specific circumstances. In other scenarios, images with the shades closed performed better. One size rarely fits all, and bigger data sets can help identify which sizes fit which customers under which constraints.

2) Machine learning can cheat, often creatively

In one image-recognition application, the algorithm was supposed to be identifying sheep when it was in fact using grass as its decision tool, so a photo of a green field with rocks was scored as containing sheep. A race-car simulator found that the application’s physics engine did not penalize crashing, so the car would careen around the track, bouncing off the mid-point on the barrier wall of the straightaways. An algorithm designed to create locomotion simply built a tall, unstable tower that fell forward; a refinement allowed the tower to somersault for greater speed.

3) Knowledge of ground truth matters more than ever

As algorithms will be trusted to do more and more activities either unattended or with minimal human oversight, people will need to spot errors in data, logic, or translation into managerial levers. We have all been served online ads that are irrelevant, offensive, or otherwise a waste of the advertiser’s money. The “flash crash” of May 6, 2010 that shut down the US stock market was caused by a confluence of mischievous human trades and runaway algorithms. Our heavily digitized automobiles illuminate “check engine” lights that are as often a sensor error as a mechanical fault.

4) Ethics needs to catch up to the technology

Google, Facebook, and other machine learning leaders are realizing the potential for algorithmic harm and have funded both internships and full-time positions for ethicists. The costs are only beginning to be counted: of the speed and scope of algorithms that cannot be understood at human scale, or the “cheats” that may not be immediately obvious (as with the sheep/grass image recognition glitch), or the migration of computing from beige boxes (from which electronic money could be stolen, for example) to physical devices in people’s homes that both invade privacy and control door locks and other safety features.

These are in fact exciting times: machine learning has incredible potential beyond spam detection, credit card fraud monitoring, and energy efficiency gains. For these benefits to accrue, however, managers, investors, and institutions of training and education need to recognize the need for the organizational contexts, from budgets to risk mitigation, in which the technical gains will be situated. This path from recognition to action will take decades, be marked by many false starts, and unfold with an often frustrating sequence of relearning lessons that have already been taught elsewhere.