Roadkill of the Fourth Industrial Revolution

The emerging canon of disruptive tech is supposed to save the world, but it might have some unforeseen victims — all of us.

THERE ARE TIMES I WORRY about an impending crash between technology and all that we care about (our families, our home, our living world). When I do, I think about Elaine Herzberg and Rafaela Vasquez.

Elaine Herzberg was the 49-year-old grandmother of seven from Tempe, Arizona who on March 18, 2018 became the world’s first bystander killed by a self-driving car. The dashcam footage posted to YouTube shows Herzberg pushing a red bicycle, turning her head in surprise a second before impact — a human rabbit in the headlights of a fully autonomous Uber car.

In the released clip, the footage switches to the interior view. Rafaela Vasquez, employed by Uber as a “safety driver,” is looking down at her cellphone. Rather than scanning the road, she is trusting that the car’s sophisticated autopilot and sensors have her back. At the last second, she looks up. Her horrified look is the final black and white frame.

Under heavy spin from Uber and authorities, the media painted both Herzberg and Vasquez as expendable in the onrush of progress: Herzberg was described as a “homeless woman” and drug user. A previous felony conviction of Vasquez, the poor Latina “safety driver,” was also foregrounded. We were told this was just one flawed human being hitting another — like millions of car crashes before. Move on. There’s nothing new to see here.

But, of course, everything was new. Beside the two marginalized women was a third actor, not visible in the footage: the experimental system of sensors, robotic automation, streaming data, and artificial intelligence (AI) at the bleeding edge of what has been called the new “Fourth Industrial Revolution.”

The tricked-out blue Volvo that terminated Herzberg’s life wasn’t an isolated malfunctioning gizmo. (In fact, it didn’t malfunction at all.) It was the advance guard in a battalion of qualitatively different technologies now rolling out across transport, healthcare, agriculture, defense, energy, water, production, retail, and every other part of our lives. Harnessing a radical set of interlocking technology platforms, backers of these innovations have their sights set on re-editing genomes, hacking weather systems, and transforming our food systems, our bodies, and our democracies.

Also invisible at the crash scene was a system of collusion, greed, and economic adventurism. A municipal authority had let its citizens become test subjects for a rapacious tech corporation with cheap overworked “safety drivers.” Herzberg is among the first roadkills of the Fourth Industrial Revolution. And Vasquez is not the last of its scapegoats.

THE TERM “FOURTH INDUSTRIAL Revolution” or “4IR” was popularized by the World Economic Forum. In Davos circles, “4IR” is a buzzword that thrills and scares neoliberal elites in equal measure. Just as technologies clustering around steam, the combustion engine, and computing transformed economies, cultures, and our biosphere, so, the argument goes, we are now entering an even deeper technology-driven transformation structured around massive data and computation.

There is an emerging canon of 4IR technologies: artificial intelligence, particularly machine learning, tops the list, followed by robotics, nanotechnology, sensors, synthetic biology, blockchains, the Internet of Things, and other “disruptive” technologies. “Disruptive” is the Silicon Valley way of describing what the twentieth-century Austrian economist Joseph Schumpeter used to call “creative destruction” — the shaking up of the social and economic order to crack open new sources of value. Uber is the poster child disruptor. Facebook co-founder Mark Zuckerberg’s “Move fast and break things” motto is a celebrated mantra in Silicon Valley. But as the death of homeless pedestrian Herzberg shows, the outcomes can often be grim for those being “creatively destroyed.”

Which may end up being all of us.

In an age when seven of the top ten companies by equity value are now tech operations, there is plentiful cash to pour into 4IR technologies. Tech investors are floating a Cambrian explosion of radical tech startups looking to be the next “Uber for x,” where “x” may be anything that can be profitably disrupted. Increasingly, that includes seeking hi-tech solutions to the grand ecological and existential threats of our time: climate change, biodiversity loss, as well as energy, water, and food insecurity.

Trying to apply high technology to global problems is nothing new.

The illusion of the big technofix is always seductive, or at least useful as a PR tool.

Nuclear power was supposed to make energy cheap, GMO foods were to feed the world, and synthetic chemicals promised to end malaria. None of that has come to pass. But the illusion of the big technofix is always seductive, or at least useful as a PR tool.

Today, agribusiness’s efforts to digitize farming by replacing workers with robots and drones is sold as “sustainable” farming. Algorithms applied to make energy grids “smarter” are given climate change cred. Biodiversity recorded on a blockchain becomes a digital asset to track, manage, and manipulate like an Amazon stock inventory. Once you start seeing the world with 4IR hammers, global problems are re-forged as nails.

In this vein, PricewaterhouseCoopers — a multinational “professional services” firm that provides services to more than 400 of the world’s Fortune 500 companies — has established a “Fourth Industrial Revolution for the Earth” project in collaboration with the World Economic Forum. The project releases glossy reports — “Blockchain for a Better Planet,” “Harnessing 4IR for Oceans” — filled with sunny examples of high-tech applications. Part of the United Nations has embarked on cheerleading “tech for good” conferences as well. Like the “Atoms for Peace” propaganda of the 1950s, the chief task of these initiatives is to reassure the political class. Far from a threat, 4IR technologies are presented as an app store of ready-to-download policy options. No need for the heavy lifting of social change. “Don’t worry. Be app-y.”

This is what Evgeny Morozov, an American writer who studies political and social implications of technology, calls “solutionism” — “an unhealthy preoccupation with sexy, monumental, and narrowminded solutions — the kind of stuff that wows audiences at TED conferences — to problems that are extremely complex, fluid, and contentious.” Doing policy by technofix may feel good, but by failing to engage in the real causes and currents of our global challenges it distracts from the critical work we need to do to create real transformation.

The specific risks of each of the new species of 4IR technologies also raise concern. Each challenges existing oversight and regulatory arrangements in its own way. For example, the new materials of nanotechnology display different modes of toxicity (depending on shape, size, and surface effects) that government legislation of chemicals hasn’t been updated to account for. Gene editing and synthetic biology may set in motion unexpected changes in the genetic functioning of living things as they skirt existing biosafety laws that aren’t yet prepared to regulate them. Networks of sensors, drones, and the Internet usher in new states of surveillance that human rights laws haven’t been updated for.

Far too slowly, it is dawning on technology policymakers that managing the waterfront of the complex new societal, environmental, and health risks flowing from the current innovation explosion is a global challenge in itself. It will require serious investment in technology assessment, tracking, and monitoring.

BUT THE REAL CHALLENGE in addressing the synergistic changes will come as the full force of the Fourth Industrial Revolution begins to truly unfold. For example, what unifies all of the 4IR technologies is data. Moving, storing, and leveraging vast quantities of data is the heart of the 4IR vision. To enable technologies such as gene editing and synthetic biology, the world is currently amassing petabytes of genomic data and is expected to reach a zettabyte (a trillion bytes) by 2025. Add farm data, environmental data, health data, and more, and that further swells to 163 zettabytes of data expected annually by 2025. By 2025 the Internet of Things is expected to contain more than a hundred billion connected devices each with sensors hoovering up data to store on blockchains and process by artificial intelligence.

The planetary infrastructure for carrying, storing, and computing data is already the largest machine the world has ever seen, and is growing faster than anyone can track. In 2018 industry was laying fiber optic cable at the rate of 57,077 kilometers per hour — almost 50 times the speed of sound. Producing just the silicon chips for the 130 new digital devices being added to the Internet of things every second consumes small oceans of water, vast quantities of gases and toxic chemicals, and enough energy to power whole counties. Yet, the silicon itself looks set to run out by 2040. Powering data infrastructure is forecast to consume one-fifth of global electricity by 2025, emitting up to 5 percent of global CO2. Factor in the biodiversity impacts of mining metals and rare earths to build the components and the idea that the Fourth Industrial Revolution will be “for the Earth” begins to sound increasingly hollow.

Then there’s the efficiency problem. Most claimed environmental benefits of 4IR technologies come down to some sort of promised eco-efficiency gain.

There is no automatic link between efficiency and being lighter on the Earth.

But there is no automatic link between efficiency and being lighter on the Earth. Artificial intelligence is particularly relevant here. Most machine learning systems are optimization processes where the neural network observes data, finds patterns, and teaches itself to optimize a desired outcome. Google’s “DeepMind” AI was initially trained on video games including an electronic boxing game. DeepMind observed how the game was played and then evolved strategies until the AI could always reliably maneuver its opponent into the corner of the boxing ring and pummel them repeatedly — racking up points.

When I contemplate Bayer-Monsanto or John Deere’s harnessing of machine learning to further improve the productive “efficiency” of industrial agriculture I imagine them that way — forcing already depleted soils and farm economies into a metaphorical corner of the ring and pummeling them more efficiently.

But the real nightmare 4IR problem — the Uber car mowing down grandmothers scenario — is the black box problem. In airplane crashes, a black box — a flight recorder that records audio and flight data from the cockpit — can help explain what went wrong. In AI, the black box is the opposite — it’s the unknowable nature of why an AI agent does what it does.

Machine learning involves giving a computer large amounts of scenarios and data to evolve its own model of how to behave in any given situation. As of this moment, tools do not exist for AI systems to explain back to humans why any particular decision is reached. So that self-driving Uber car can’t explain why it could recognize a moose but failed to recognize and stop for a pedestrian pushing a bicycle. When those systems go wrong or create unexplainable side effects, the outcome in a self-driving car may be tragic at the level of an individual. But when AI decision-making is also being applied to managing and manipulating ecosystems, food systems, water supplies, and even the genetic basis of life on Earth, the collective tragedy of a bad decision could be both hard to trace and exponential in impact.

Early steps are already being taken to use artificial intelligence to determine how our climate system could be geoengineered to counteract global warming. What if an AI decision on the scale of changing the climate went wrong and couldn’t be explained? How might many parallel AI processes interact? And could that create further unexplainable outcomes? Proponents of the Fourth Industrial Revolution will contend that it will never get that far, that AI and other 4IR platforms are only tools to assist humans. They are confident there will always be a human “safety driver” making the ultimate decisions, ready to apply the brakes.

EVEN NOW I SEE HER — lulled into security by the hype that the car will take care of itself. At the last second she looks up. Her horrified look is the final black and white frame. She tries to turn the wheel. It is too late.

Subscribe Now

For $15 you can get four issues of the magazine, a 50 percent savings off the newsstand rate.