Finding Genius

by Kunal Mehta

  Our society has debated the pros and cons of automation before. During the Industrial Revolution, the steam engine, conveyor belt, and power loom brought about mass-scale mechanization. A generation later, the world moved towards special-purpose machinery, factories, and mass production. Each of the industrial revolution and subsequent periods resulted in material improvements in the standard of living — from 1810-1850, real wages doubled, and over the course of the 19th century, the economy grew by 6x (the growth rate had been almost entirely flat during the previous 100 years). Moreover, the impact of industrialization wasn’t merely economic: mortality rates plummeted, efficient transportation systems bloomed, and high-speed communications systems connected towns and families across the country. Still, 19th century America wasn’t spared from short-term labor displacement. Unemployment levels reached as high as 8%, and working conditions left much to be desired. The transition from an agrarian to an industrial society tore at the fabric of American society. Familiar modes of work were destroyed, families uprooted themselves, small town ties disintegrated, and the religious and mystic were gradually supplanted by the secular and scientific. It’s difficult to comprehend the suffering and pain this transformation inflicted on generations of Americans.

  From the long lens of history, it is also difficult (and perhaps foolish) to quantify the impact of the industrial revolution. Of course, with a much shorter lens, this sort of analysis is even harder. What is clear though is that the technology developments of the 1800s transformed the world, and today, we’re at the precipice of another unavoidable revolution: the automation revolution.

  The inventions of our age will meaningfully expand the scope of automation in every aspect of life. Once again, society will need to reshape the workforce and adjust to new global economic realities. Opportunity is everywhere. Many will step in and forge a new path, taking advantage of the mass restructuring of the industrial and cultural landscape. Many more will be at risk. There are approximately 3.5 million truck drivers and 8.5 million Americans (~3% of the population) employed across the trucking industry. If certain prognosticators are correct, the majority of these jobs will disappear. And that’s only a single industry — unlike previous industrial transformations, ours might be broad and undiscriminating in its impact.

  In order to better grasp the potential impact of automation we must first understand whether predictions about the exponential improvements in machine automation are accurate. The logic for a compounding rate of machine prowess revolves around three concepts: the ability for machines to (i) sense their environment through better receptors, (ii) store, process, and communicate sensory information, and (iii) leverage ever-improving software to derive insights from organized data. Companies investing in autonomous vehicles (AVs), for example, are making a bet that the technology needed to enable autonomy across each layer of the data stack is sufficiently mature.

  At the first layer of the stack, the manufactured sensors needed to make AVs possible can be likened to human senses: the ability to see, to hear, to touch, to taste, and to smell. Just as humans with bad vision struggle to understand the letters on an eye test, an AV with bad sensors cannot understand its environment well enough to make an informed decision.

  One reason AV companies believe now is the right time is that hardware sensors have fallen in price by more than half over the past 30 years. Better yet, these sensors are now able to capture information with such high fidelity that vehicles can recognize the curve of the road, the speed of the car next to it, the speed limit of a road, and a person crossing the street. There are a few sensors in particular, such as cameras, sonars, LiDAR’s, and radars, that form the foundation of an AV:

  Camera: The vehicle has a number of cameras that capture context (e.g. what a sign says) and color. This data provides visual information that the other sensors cannot capture.

  Sound navigation and ranging (Sonar): A sonar system emits pulses of sound and listens for echoes to approximate the distance of surrounding objects.

  Light detection and ranging (LiDAR): A LiDAR sensor, which typically sits on top of the vehicle, sends a beam of light out and creates a colorless, context-less, three-dimensional replica of its environment, based on the speed with which the light returns to the sensor.

  Radio detection and ranging (Radar): A radar sends out radio frequencies to build maps of the environment, based on the time it takes for the radio signal to return. This system differs from LiDAR in that radio waves have less absorption and can therefore travel longer distances and identify items that LiDAR systems cannot recognize (e.g. LiDAR systems struggle to see through dust.)

  The combination of data coming from cameras, sonar, LiDAR, and radar sensors enables a vehicle to create a three-dimensional rendering of the world, similar to what we see with the human eye — a clear indication that data capture systems are approaching the necessary quality (and cost) to replace human tasks.

  The second layer of the stack dictates what happens when data is captured by autonomous machines. The challenges here are immense. Guido Vetter, the Head of Analytics at Daimler Motors, stated that “flexibility and scalability are what you need for AI and advanced analytics, and our whole operations are not set up for that.” If we take Daimler as an example, historically the company stored all data on premise (servers located on facility grounds.) However, given the amount of data generated by these AV sensors, the physical space and cost constraints of storing it on premise have become untenable. Daimler, along with many other companies, has begun to shift data storage to “the cloud,” a shift that should not be underestimated in significance. Whereas the company previously relied on in-house expertise to store information, it now mainly utilizes third-party providers such as Amazon Web Services (AWS) to store and process the logarithmically expanding amount of data required to enable autonomy in vehicles.

  The third layer of the stack is where companies such as Daimler leverage machine intelligence to derive insights from the data, a primary driver of autonomy. These companies use algorithms to run extremely complex probabilistic regressions on massive amounts of data — artificial intelligence (AI) — to turn plentiful, rich, labeled data into the autonomous functioning of machines. In order to better understand how these algorithms work, consider an analogy: four tennis balls are located on a white wall in the shape of a square. The balls alone are just that — physical objects. The square, of course, is not actually ‘there:’ our minds organize a pattern and abstract a shape. The mind allows us to form conclusions from the non-physical. Similarly, an autonomous vehicle’s ability to abstract conclusions is a result of applying algorithmic analysis to raw perceptual data. Recent advancements in the quality of these algorithms play an integral part in the process with which AVs navigate their environments. Thus, it is not surprising that demand for workers with AI talent has more than doubled — from 2015—2018 job postings for “machine learning engineer” grew 344%, and three of the top four paying jobs include computer vision engineer, machine learning engineer, and data scientist. This spike should only further accelerate innovation and the speed of commercializable automation.

  Technology optimists believe that with the emergence of this data stack — sensors, data infrastructure, and AI — autonomous vehicles, along with a number of other robotic systems, are poised to meaningfully change our world. These same individuals also point to the 1.2 million deaths annually due to automobile accidents, 93% of which are caused by human error. It should come as no surprise then that from August 2014 to June 2017, an estimated $80 billion was invested into AVs by the auto industry and venture capitalists. As of 2019, there are already three startups that have raised north of $1 billion from investors — Cruise, Nuro, and Aurora — with many more that have raised well over $100 million. Similarly, many large manufacturers such as Ford, GM, and Toyota have invested well over $1 billion into internal programs focused specifically on autonomy. New rideshare companies such as Uber, Baidu, and Lyft are following suit. Gab
e Cunningham, an investor at the mobility-focused venture capital fund, Fontinalis Partners, explained the interest in the autonomous vehicle sector: “this is a transformative shift in how society moves, touching trillions of dollars in the global economy. To realize that vision though, there is an expansive amount of research and development capital that will be needed to reach the necessary milestones. Ultimately, the big question is not if AVs will become a reality, but what companies have the resolve and technical capabilities to bring these vehicles to market at scale over the next decade and beyond.” Gabe’s sentiment is one that is generally shared across the industry: autonomy is an impending reality but the winner(s) will have to invest an incredible amount of capital to deliver on the market opportunity.

  The long-term implications of this transition are massive. Those immediately affected include 3.5 million truck drivers, and 1 million taxi / rideshare drivers. There will also be significant secondary effects. One fifth of organ donations come from vehicular accidents. One third of all civil trials have to do with motor vehicles. Highway motel occupancy will decrease drastically. The auto insurance and claims market will contract. Police forces will lose revenue generated from ticketing drivers. All said and done, millions of jobs will be indirectly affected, likely with a net negative impact, at least in the near term. These changes warrant concern but they shouldn’t completely obscure excitement. In transitioning to autonomy, millions of lives will be saved from auto accidents. Senior citizens, children, and the disabled will be able to use transportation solutions currently inaccessible to them. Vehicles will communicate and synchronize decisions that should improve road congestion — today Americans spend more than 6.9 billion hours a year sitting in traffic — and decrease CO2 emissions and fuel consumption. Finally, decreased transportation costs will open up access to more places for more people in a way that will produce benefits not currently imagined.

  While the potential impact of autonomous vehicles should not be understated, AVs are just one example of how the data stack will transform a cluster of industries. Julian Counihan, for example, is betting on this transformation. As a partner at Schematic Ventures, where he focuses on supply chain and manufacturing, Julian has invested in companies such as Plus One Robotics, Symbio, Root, and Azevtec. When discussing these opportunities, he focuses “on technology within industrial sectors where the automation problem is less challenging.” He says:

  “The industrial environment can be controlled to eliminate difficult variables, and infrastructure can be modified to simplify operational complexity. Still the same concepts apply regardless of business type. Industrial application systems will fuse sensor data to understand the environment and leverage machine learning to arrive at optimal decisions. Having been a part of a few robot companies, I find the difficulty of full automation can be understated. Unless the problem itself can be changed, the added cost to address the final one or two edge cases can push a solution past the point of commercial viability. The key to a robot company’s success is as much finding the right problem as building the right technology.”

  Not surprisingly, Julian shares a similar sentiment to Gabe and even expands upon the scope of what may be defined as automatable; but he also acknowledges that entrepreneurs must be very careful to figure out the right time and capital structure to achieve their goals.

  iRobot, manufacturer of the Roomba autonomous vacuum, is one of the first companies to create a consumer-grade autonomous robot. To date, the company has sold north of 15 million Roombas. Folks that have interacted with the product likely recognize the value of autonomous cleaning, but also understand the limitations: the robot misses certain areas, dies before re-docking for charging, and displaces items that it bumps. However, these limitations will likely lessen in the near future. Colin Angle, iRobot’s CEO, predicts a “big leap” where the Roomba “remembers what’s going on in the home.” Such a leap will be made possible by improvements the data stack. The Roomba i7+, launched in 2019, has a myriad of sensors, a camera, and an on-board computer processing unit, all of which allow the vacuum to see a room and memorize the layout. The Roomba will soon be able to clean every part of a room — even if the layout of the room changes — and map out a path back to the docking station. It’s not a reach to think that the Roomba might soon also be able to apply different cleaning techniques to certain floor surfaces and report incidents such as a spilled glass of wine in order to minimize long-term damage.

  Jacob Yormak, my partner, and managing partner at Story Ventures, agrees an immense opportunity exists around automation. At the same time, he dismisses the notion that AVs and other autonomous machines are anywhere near ready to replace the most complicated forms of human intelligence. For now, AVs fall into the bucket of artificial narrow intelligence (ANI), whereby they are able to complete a single, narrow task. This is compared to artificial general intelligence (AGI), whereby a machine would be able to complete multiple, generalizable, complex tasks (think: The Terminator). As Jacob says:

  “The current state of artificial intelligence is still very limited — we are just scratching the surface of ANI. Though the media may suggest otherwise, we are far away from machines replicating the most complex forms of human intelligence. Think about babies: they are born with complex sensors that are almost always on, incredible internal communication systems, and algorithms that have been honed over thousands of years. Improvements across the data stack have enabled machines to complete real-world tasks in narrow verticals such as selecting when to turn on a road, or how to lift a package on a conveyor belt, but the algorithms used to make these decisions quickly break down with even minor deviations from the originally intended purpose.”

  This recognition that machines do not have general intelligence highlights a key insight — in the near term, automation will be limited to highly repetitive tasks with predictable variations.

  These hesitations cited by Gabe, Julian, and Jacob provide a necessary reminder that robotic systems are not yet ready to take over the world. At the same time, the billions of dollars being poured into companies working on these systems clearly demonstrate the immense economic opportunity in commercializing autonomous systems. There are millions of repetitive, logic-based jobs carried out by humans today: cashiers (3.4 million), food preparation and serving workers (2.7 million), customer service representatives (2.1 million), and freight movers / laborers (2.0 million). These represent just a small sampling of jobs that will be susceptible to automation. Given the economic advantages, this transition to automation seems inevitable — as do the adverse impacts to millions of American lives.

  Going forward, we can resist this change, though history suggests that it is difficult to deter the momentum of technology. Conversely, we can embrace the efficiencies of this new wave and construct solutions to aid those at risk. To the latter, we are already seeing some progress. Lambda School, a trade school focused on coding, is one of the many institutions emerging with real signs of success to help individuals transition into jobs that will be in demand for the foreseeable future. This is part of a larger trend as venture capital funding for education technologies reached an all-time high in 2018 of $1.45 billion. Andrew Yang, a 2020 presidential candidate, is running on the platform of universal basic income in which every US citizen over the age of 18 would receive $1,000/month. Famed innovators including Bill Gates and Stephen Hawking have proposed a robot tax in which companies that leverage automation would have to pay higher taxes to compensate the US people for the labor displacement. Each of these solutions, if wielded properly, have merit and can partially remedy the challenges ahead. Yet, there is no silver bullet. A portion of the population will, once again, struggle to navigate the changing landscape.

  I believe that those investing in and building the future of automation are confident that the end result will be an improved material and non-material standard of living. However, it is important that as the world moves forward, we determine best practices to support those c
aught in the riptide of our impending revolution.

  ADRIEL BERCOW

  FLYBRIDGE CAPITAL PARTNERS

  The merits of universal basic income (UBI) — or a monetary stipend provided to all citizens of a country — is a hotly contested topic in the US in 2019. Candidates like Andrew Yang and Elizabeth Warren are establishing their platforms in preparation for the 2020 presidential election and have explicitly voiced support for this type of government program. This debate has become more relevant over the past two decades as automation has significantly impacted the livelihoods of truck drivers, factory workers, typists, farmers, miners, and other laborers whose jobs have been replaced or reduced by machines and automation. The differentiation between ‘unskilled’ versus ‘skilled’ labor is becoming less clear as the effects of automation are now far-reaching across all industries and geographies. In the developed world, even individuals with higher education or specialized skill sets are at risk of being out of a job as more startups seek to automate all kinds of work at a fraction of the cost. As this debate rages on, the central question is: What will happen to humans in a future where automation will replace most menial tasks? Rayfe Gaspar-Asaoka from Canaan Partners provided a perspective on artificial intelligence that shows that this reality is not far away. In this chapter, Adriel Bercow, an investor with Flybridge Capital, explores the factors driving this discussion in a thesis focused on the ‘future of work.’ He writes, a ‘combination of evolving demographics, a widening income and skill gap, and an education system that was not built for today’s digital economy have created an investment opportunity.’