“The world hates change, yet it is the only thing that has brought progress” - Charles F. Kettering
Investors have become increasingly short-term focused due, in no small part, to today’s 24/7 financial news media and instantaneous access to information via the Internet. It is important to understand how current developments at the macro and micro level may impact our investment outlook, but it is also important to take a few steps back and consider longer-term trends that have the potential to create future secular headwinds and/or tailwinds for various companies, industries and indeed entire asset classes. While the opportunities (and risks) presented by these long-term themes may take years to unfold, it is important to monitor them now in order to be prepared to act when the time is right. We have written about such topics in prior Outlook Series articles including the potential for an American Manufacturing Renaissance, the impact of the aging Baby Boomer, and the prospects for energy independence. This month, we explore another long-term theme that has the potential to meaningfully impact future investment outcomes as well as the overall economic landscape: algorithms.
The light bulb, the telephone, the automobile, the assembly line, the boat, the train, the plane, the computer, the Internet, and the list goes on of extraordinary innovations that have changed the world. Some innovations have made it easier to communicate with one another; others have made it quicker to travel from one place to another, while others have made information much more accessible. These changes have not only made our lives more convenient, but they have also had significant impacts on productivity and the growth of the economy over the decades. Some of these innovations have been more disruptive than others, in some instances requiring major adjustments by the workforce to learn new skills and trades as some industries became obsolete and new ones emerged. Today, algorithms have the potential to take their place on the list of revolutionary innovations as advances in technology have enabled them to become increasingly complex and tackle tasks once thought to be strictly the domain of humans.
The Human Element
A majority of American adults have never taken a physics1 class, yet each time we sit behind the steering wheel of a car, we engage in an exercise of complex physics calculations, making judgments that range from how much distance we should leave between cars and how soon to brake when approaching a stop sign, to whether or not we have enough time to pull out into oncoming traffic.
We integrate active environmental risk assessments to these often subconscious physics calculations including the influence of weather, determinations as to if people or animals are likely to enter the field of traffic, attempts to discern whether the noise heard is a siren, and so on. These and many other assessments, recalculated minute-by-minute if not second-by-second, are made against a background of distractions from radio music, to the child in the car seat, to the beautiful view in front of us or the mental distractions of our own thoughts. When the act of driving is broken down piece-by-piece, it is astonishing that everything goes right more often than not. The brain’s cognitive processing abilities are impressive.
Yet in the end, we are only human and things do go wrong. According to the U.S. Department of Transportation’s Traffic Safety Facts 2011 Annual Report, there were 9.2 million accidents involving motor vehicles (excluding motorcycles). Thanks in part to technology innovations that help reduce the number of accidents (e.g., anti-lock brakes), the absolute number of motor vehicle accidents peaked in 1996 at 12 million. On an absolute basis, motor vehicle accidents have declined 22%. This is particularly impressive considering vehicle miles travelled (VMT) increased by 18.7% over the same period, suggesting people were driving more but getting into fewer accidents, a 35% decline in accidents per 100 million VMT. Similarly, technology innovations (e.g., airbags) have also reduced the physical consequences of accidents as fatality rates and injury rates have come down in lockstep (Chart 1). Looking back even further to the Unsafe at Any Speed era (published 1965), an even more dramatic improvement in ratios is clear (Chart 2).
Arguably the only thing that hasn’t drastically improved over time is the driver. Therefore, while we should anticipate continued technological improvements will further reduce these ratios, the rate of improvement appears to be moderating. For example, brake technology continues to improve, yet as long as motor vehicles are dependent on a human to press the brakes, the benefits are likely to be incremental, an issue that has not gone unnoticed.
Enter, the Autonomous Vehicle
Four states have now passed laws permitting driverless vehicles (California, Florida, Nevada and Texas) despite the program being far from commercialization. Google’s fleet of vehicles has logged over 300,000 autonomous miles using laser range finders to create detailed 3D maps of the environments, high resolution maps of the world for radars, cameras, GPS inertial measurement systems and wheel encoders, in addition to various other technologies to navigate by manipulating actuators and other automation technologies to accelerate, brake and steer.
But the secret sauce, as is true with Google’s more famous endeavors in search engines and advertising, are the algorithms running it all. While the hardware discussed above has definitely improved, the hardware involved is not per se new and innovative. What has changed is the ability to process the data generated by the “visualization” hardware accurately and quickly enough to respond in real time. This is no small task and has only been made possible recently by the Moore’s Law, driven by decreases in data storage and processing costs and the related increase in data collection.
What is true of autonomous vehicles applies to many other sectors. While the physical hardware of robots or unmanned aerial vehicles more easily captures the attention and imagination of the public, the real game changer is the increasing power and spreading influence of algorithms. Algorithms are fundamentally changing industries and social conventions. Understanding their role in industries and the economy as a whole is an increasingly important part of investment analysis.
Algorithms – the Great Enablers
Algorithms themselves are not new. The earliest known algorithm dates back to 2500 BCE Sumeria in the form of clay tablets that showed how to simply divide any size harvest across any numbers of parties.2 At the most basic level an algorithm is a binary decision tree; if A do X, if B do Y. Because almost any action can be broken down into a series of binary decisions, algorithms are incredibly useful to improve and automate processes. Even a checklist, a simplified manual algorithm, can drastically improve outcomes whether it is a pre-flight checklist for pilots or a surgery checklist in hospitals. Johns Hopkins’ success in virtually eliminating ICU bloodstream infections via a checklist is an example.3
Manual algorithms are limited by the human user’s ability to follow and work through them, are best used in defined circumstances, and are more likely to succeed if they are concise. They are most appropriate in situations where there are a limited number of variables to be accounted for. In contrast, driving is a massively multivariate activity and has historically required the massive processing power of the adult brain. For a motor vehicle to become autonomous, the algorithm needs to be automated which became possible with technology, enabling a new phase of algorithms to emerge.
Christopher Steiner, in Automate This, classifies algorithms under three phases. In phase one, a human creates an algorithm which a party uses to generate a decision recommendation. Depending on the context, the party may or may not act on that recommendation. A more automated “phase one” algorithm is the “Check Engine” light. Sensors in the vehicle detect a condition, automatically triggering a recommendation in the form of a light turning on in the dashboard, though whether or not you react to the recommendation is out of the algorithm’s control. Therefore, phase one can be labeled as full human-in-loop algorithms.
In phase two, the algorithm is still designed by humans, but now has power to directly act upon its own recommendations. Fail safes may be built in to bring a human back into the loop under specified or anomalous conditions, but in general the human factor is limited to initial design and upgrading. Many cars now come equipped with features enabled by phase two algorithms, ranging from variable speed windshield wipers that use sensors to detect levels of moisture and the vehicle’s speed, to sensors in the rear of a vehicle that automatically sense when an object has entered the vehicle’s path and thus pump the brakes.
These are relatively simple phase two examples that are already fairly commonplace in many products and industries. With the declining cost of processing power and storage, which has enabled the leap from the “Check Engine” light and semi-autonomous component functionality to fully autonomous vehicles, what is changing is the breadth and depth of decision trees that can be built and operated. Increased storage enables the increased algorithm complexity required, while increased processing power enables large quantities of inputs to be rapidly processed in order to derive decisions in real time. Both are critical to enabling increased motor vehicle autonomy. Consider the decision an autonomous vehicle would need to make if it detected something suddenly moving across the road. This would likely automatically trigger braking and honking, but may also require other decisions if braking alone is likely to be insufficient. The vehicle may need to steer itself out of danger, but in which direction? This would require processing any oncoming traffic, the object that is running in front of the vehicle as well as how it is likely to respond to the vehicle’s action. This is a complex decision requiring massive computing power to run a highly developed algorithm in order to make the right decision in time to have an effective outcome.
As storage and processing capacity further expand, adoption of phase three, self-learning algorithms, becomes increasingly possible and probable. While a human is involved in establishing the initial parameters, the algorithm crunches massive amounts of data to independently refine itself to improve its performance. An early example of a phase three algorithm was the breakthrough in translation software. Algorithms designed by experts in linguistics and strong teams of programmers had repeatedly failed to perform at satisfactory levels as human language often seems to be susceptible to more exceptions than rules. Eventually, a programming team with no linguistic expertise wrote a self-learning program that processed volumes of Canadian government transcripts, in both English and French, to derive a program capable of translating back and forth between the two languages. The key breakthrough was that the algorithm “learned” to translate groups of words together as opposed to translating words individually. Better context was helpful in identifying exceptions and other linguistic idiosyncrasies.
Phase two algorithms may support significant advancements in autonomous vehicle development, but ultimately successful phase three developments are necessary for it to become a fully commercial technology. On a personal level, phase three may provide the owner of a new autonomous vehicle with a better ride to work. On the first morning in the vehicle, you will tell it to go to work and it will likely perform functions similar to a smart GPS system, considering various routes and known traffic conditions to plot the most direct route, which it is likely to also follow on the way home. But as the days go by, your vehicle may suddenly take an unexpected turn as it may have been comparing its actual results to the assumptions calculated in the original route determination and concluded that the route may be sub-optimal and thus it begins experimenting with alternative routes. Eventually your car may “know” that on Wednesday mornings it is better to take side streets, but on Thursday afternoons the highway is a better choice, automatically accounting for new information such as traffic accidents, construction or other delays.
The Winners and Losers in Disruptive Innovation
There are many economists who have written about the effects that innovation and progress have on the labor markets and the economy as a whole. The term “Creative Destruction” was coined by Joseph Schumpeter in his 1942 book Capitalism, Socialism, and Democracy where he wrote:
“…the contents of the laborer’s budget, say from 1760 to 1940, did not simply grow on unchanging lines but they underwent a process of qualitative change. Similarly, the history of the productive apparatus of a typical farm, from the beginnings of the rationalization of crop rotation, plowing and fattening to the mechanized thing of today–linking up with elevators and railroads–is a history of revolutions. So is the history of the productive apparatus of the iron and steel industry from the charcoal furnace to our own type of furnace, or the history of the apparatus of power production from the overshot water wheel to the modern power plant, or the history of transportation from the mailcoach to the airplane. The opening up of new markets, foreign or domestic, and the organizational development from the craft shop and factory to such concerns as U.S. Steel illustrate the same process of industrial mutation–if I may use that biological term–that incessantly revolutionizes the economic structure from within, incessantly destroying the old one, incessantly creating a new one. This process of Creative Destruction is the essential fact about capitalism. It is what capitalism consists in and what every capitalist concern has got to live in…”4
Schumpeter says that while jobs are lost, companies close, and entire industries cease to exist, there are significant benefits that come from these disruptions in the long-term. Over the long-term, countries in which “Creative Disruption” exists are wealthier and have higher productivity than those that do not, and the people enjoy higher standards of living as the jobs are better, the work weeks may be shorter, and so on. However, over the short to intermediate term, some individuals will be worse off as their skills are no longer applicable or desired in the workplace. These individuals need to re-train or they will be left behind. If countries try to protect these industries, they are only doing a disservice to the greater economy as they are trying to protect against the inevitable and subsequently slow progress.
What opportunities have been created from historical innovations and what disruptions have come? Innovations affect labor markets in various ways, including directly, indirectly, as well as potentially suppressing future job growth in certain industries depending on the innovation.
The telegraph led to direct destruction of telegram messenger jobs, and the telephone eventually replaced the telegraph, creating the telecom industry and thus a large number of jobs. Automobiles, railroads, planes and ships put many horses and mules out of work as they were no longer needed to pull plows, carriages, and boats along the canal. However, these vehicles created a huge transportation industry which employs millions in the U.S. today. Furthermore, this allowed numerous industries such as airlines, shipping, motor vehicle, travel, rail, and trucking (to name a few) to emerge and not only employ workers, but also fuel spending that has allowed the economy to grow.
But this may only be the tip of the iceberg when it comes to algorithms and their impact on employment. It is more difficult to calculate the impact when it is a service being automated. The Internet, though it has created whole new industries such as social media and Internet software services, has contributed to a decline in jobs in many occupations (e.g., travel agent, telephone operators, librarians). Internet-based companies such as Orbitz Worldwide now use algorithms to automate many of the services that travel agents once provided, from finding flights, hotels and rental cars to the vetting of the facilities beforehand to help ensure satisfaction (via online reviews). It is thus not surprising that the number of people working as travel agents has steadily declined (Chart 3).
The automation of a manufacturing facility is very visible, and it is theoretically reasonable to calculate the job losses either directly in the number of people let go in the case of an existing facility, or the jobs avoided (job growth suppressed) in the case of a new facility, likely also accounting for some additional job losses related to competitors becoming less competitive and either reducing capacity or exiting the market. Importantly, all those robots and other physical automation components are likely to be run by algorithms, so even these job losses should ultimately be accorded to algorithms.
Retracing our steps back to autonomous vehicles, often the current focus is transportation freedom for those who can’t drive, particularly with a growing elderly population. Yet the early adopters of autonomous vehicles may be companies. Transportation fleet operators are already heavy adopters of algorithms in route planning and fleet maximization software, but labor is a major cost. A persistent shortage of drivers (which is skilled labor and often requires special licensing) is likely to get worse with the Big Crew Change from retiring Baby Boomers, and may be a strong incentive to automate fleets. As an example, for Wal-Mart, ground logistics is a key component of the company’s operation model with over 7,000 drivers, a fleet of 6,500 tractors and 55,000 trailers. How would autonomous vehicles ripple through the company and the larger retail industry?
Living in an Algorithmic World
Autonomous vehicles are a useful proxy to demonstrate why competitive analysis to identify investment opportunities will increasingly need to determine where algorithms are currently being used, what type (which phase) are being used, where they are likely to be adopted next, and whether there are substantive differentiations in adoption or uses of algorithms within industries or across value chains. Algorithms are likely to create new areas of competition within industries and between industries, particularly when they are enabling and leveraging network effects. At the same time, algorithms may commoditize previous competitive advantages. Providers of algorithms and algorithm technologies may be big winners, but beneficiaries of these new technologies may capture even more of the value. Algorithms will likely lead to the demise of some industries and the creation of others. It is important for investors to monitor these changes and determine if their impact will disrupt or sustain the competitive dynamics within an industry.
Understanding and accounting for these impacts will require adjustments and adaptations to the investment research process. For example, the Chief Technology Officer (CTO) may become as important, or more important, a member of the C-suite to interact with than the CFO or CEO in industries traditionally not considered technology-centric. The key will be to recognize that algorithms are not just about search engine optimization or social networking, but ubiquitous, strong currents beneath the surface that may fundamentally alter operations as adoption of increasingly complex algorithms occur. Because these developments will often not be as visible, it will be important to actively monitor for them.
Automation and algorithms have the potential to create disruptions, but also create opportunities in other areas. Automation is likely to increase productivity and serve as a deflationary mechanism driving down the cost of production and service. It will enable industries to replace some retiring skilled workers with machines, potentially offsetting some of the inflationary pressures from the Big Crew Change.
Predictions regarding the long-term impact of algorithms on the economy and the labor market range the spectrum from massive technology-driven unemployment to a modern utopian vision of greater leisure time as machines do all of our work. While these extreme predictions may be fascinating (or frightful) to contemplate, the reality is that this is a long-term phenomena where there are likely to be winners and losers as algorithms continue to evolve. The economy will likely not be the same a generation from now, just as it is not the same as it was a generation ago; and if history is a guide, the process of creative destruction brought on by the algorithms will make the economy stronger. In the words of the Greek philosopher Heraclitus, “There is nothing permanent except change”.
Analysis: Manning & Napier Advisors, LLC (Manning & Napier).
Manning & Napier is governed under the Securities and Exchange Commission as an Investment Advisor under the Investment Advisers Act of 1940.
Sources: U.S. Department of Motor Vehicles, National Highway Traffic Safety Administration, Research and Innovative Technology Administration (RITA), Techworld Worldwide, New York Daily News, Walmart.
1High school physics enrollment had only increased to a little over 30% by 2005 http://www.aip.org/statistics/trends/reports/hs05report.pdf
2Christopher Steiner, Automate This
3http://www.hopkinsmedicine.org/news/media/releases/Making_A_Better_Medical_Safety_Checklist
4Schumpeter, Joseph A. Capitalism, Socialism and Democracy (New York: Harper Perennial Modern Thought Edition, 2008) [orig. pub. 1942]
All investments contain risk and may lose value. This material contains the opinions of Manning & Napier Advisors, LLC, which are subject to change based on evolving market and economic conditions. This material has been distributed for informational purposes only and should not be considered as investment advice or a recommendation of any particular security, strategy or investment product. Information contained herein has been obtained from sources believed to be reliable, but not guaranteed.
This newsletter may contain factual business information concerning Manning & Napier, Inc. and is not intended for the use of investors or potential investors in Manning & Napier, Inc. It is not an offer to sell securities and it is not soliciting an offer to buy any securities of Manning & Napier, Inc.