You may be aware of the spotlight that a recent New York Times piece has placed on working conditions at Amazon. Like me, you may also be wondering how a company that’s so smart about creating great customer experience can be so stupid about the experience it’s creating for employees. But the Amazon story is a much bigger story. It’s the story of how automation accelerates both blue-collar and white-collar work; Amazon just happens to be at its cutting edge.
Blue Amabot, White Amabot
This quote from the New York Times piece piqued my curiosity:
Company veterans often say the genius of Amazon is the way it drives them to drive themselves. “If you’re a good Amazonian, you become an Amabot,” said one employee, using a term that means you have become at one with the system.
In another context, being “at one with the system” might connote ecological wholeness or some transcendental state, but here, we’re talking about being a cog in a machine. When we think of organizations as machines, the cog is a robot, or in this case, an “Amabot.”
The best description I’ve seen of the “blue-collar Amabot” at Amazon is an excerpt from the book, “Mindless: Why Smarter Machines Are Making Dumber Humans.” It details some disturbing examples of Amazon’s use of systems for monitoring the minute-by-minute movements and performance of employees, including reprimands for box packers who occasionally failed to use the bathroom closest to their workstation. It paints a picture of a high-tech form of old-fashioned Scientific Management thinking (aka “Taylorism”), and Amazon uses it to minimize “time theft” and increase the overall throughput of its fulfillment centers and other blue-collar workplaces.
The systems for the salaried, “white-collar Amabot” differ from the time efficiency systems for hourly blue-collar workers. Most white-collar Amabots focus on knowledge work, and the overriding goal here is to harness individual intelligence so as to maximize the collective intelligence of the greater system that is Amazon. For white-collar workers, the system is designed to capture maximum share of mind, which translates to eighty-hour workweeks and harvesting time from employee weekends and vacations. Another critical element of the systems design is finding and promoting the most promising minds with a software-enhanced version of the “up or out” talent cultivation techniques used by consulting firms and other organizations that rely heavily on brain power.
Dehumanization
Dehumanization is easier to see with physical labor. When companies tag employees with satellite navigation systems to ensure their movements are optimally routed through a warehouse, it’s hard not to see these people as cogs in a machine.
Dehumanization in knowledge work is more subtle. To see it, it helps to first look at the more general relationship between human labor and automation. In our economic framework today, the role of human labor is to do the work machines have not yet learned.
[Tweet “Today, the role of human labor is to do the work machines have not yet learned.”]
If Amazon has humans carrying out machine-like work in its fulfillment centers, it’s simply because they haven’t yet figured out how to automate those particular jobs. In this sense, human labor provides a flexibility that fills in the edges of where our automation hasn’t yet reached. As machines gain scale and efficiency, the system becomes increasingly productive. That accelerating flow of work then adds more and more pressure on the inefficient human bottlenecks in the system. This is what we’re seeing with the Amazon stories.
Our initial efforts to automate knowledge work focused on email, websites, databases, and other tools for improving the generation and distribution of knowledge. As these tools took hold, information exploded. Machines weren’t yet capable of understanding this information, so human knowledge workers came under growing pressure as they raced faster and faster to maintain the accelerating throughput of the system. Stress increased, working hours increased, and work-life balance disintegrated, as we lost ourselves in the ensuing flood.
The effects we’re talking about here are two-fold: it’s a ‘de-humanization’ in the sense of a literal replacement of humans by machines, and it’s the more subtle erosion of our humanity from our frantic treading in that flood.
Rise of the Machines at Amazon
How will Jeff Bezos and Amazon’s top management respond to the current outrage over Amazon’s labor practices? Facing similar pressure when it too had a New York Times article about its work conditions, Apple’s contracted manufacturing partner, Foxconn Technology Group, decided in 2012 to embark on a massive robotics automation initiative. Amazon is already well down this road. By acquiring robotics maker Kiva Systems in 2012, Amazon Robotics has emerged as a major player in fulfillment center automation systems.
Over the last few years, Amazon has also made huge investments in artificial intelligence, primarily in the area of machine learning. A LinkedIn search for jobs related to “machine learning” now pulls up 669 posts for Amazon, compared to 232 for Apple, 266 for Facebook, 326 for IBM, 466 for Google, and, interestingly, 976 for Microsoft. It’s not a perfect proxy since it ignores already existing positions within these companies (and potential differences in the way these companies use LinkedIn for hiring), but there is enough activity with each company that it does say something about the state of their current, incremental investment plans in machine learning.
Amazon has a long history of using machine learning to analyze the massive datasets generated by its site. It’s what powers the company’s valuable shopping recommendations, and it’s helped Amazon software developers to build voice recognition capabilities for the Amazon Echo and solve wickedly difficult logistics problems in its fulfillment centers. In April of this year, the company announced its Amazon Web Services (AWS) platform will now include Amazon Machine Learning as a way to give external software developers access to its artificial intelligence prediction engine:
The Fate of the Amabots
Jeff Bezos is usually fairly impervious to outside pressure when it’s not strategic or that might negatively impact his ability to serve customers. The current focus on Amazon’s labor practices actually could shift customer perception of Amazon, however. Since automation is clearly already core to its strategy, the company could choose to solve its perception problem by simply accelerating its plans to replace blue-collar jobs with machine-learning-driven robotics.
The future of Amazon’s white-collar Amabots is trickier. If you want to see Amazon actively testing systems for automating a flow of work that machines haven’t yet learned, just look at the “Human Intelligence Tasks” of Amazon’s Mechanical Turk. This is a system designed to streamline and commoditize the white-collar tasks of humans, and it appears to be the brainchild of Bezos himself. Many of the pattern-matching tasks it was originally designed to handle ten years ago are now already being done by machine learning algorithms, by the way.
The role of human labor is to do the work machines haven’t yet learned, so systems that make machines learn faster speed up the pace through which machines take human jobs. This is why machine learning is so critical to understanding the future ‘de-humanization’ of work. Until now, the communications and decision making aspects of most knowledge work has simply proven too demanding for machines. But that is rapidly changing.
Machine learning techniques are making huge strides in Natural Language Processing, which means that machines will soon understand human information in ways that simply weren’t possible before. As this happens, machines will steadily take over more of the communications and decision making now done by knowledge workers. And as that happens, the human bottleneck now blocking dramatic jumps in organizational intelligence is removed.
What we’re experiencing today are the early birth pangs in this process.
Amazon already sees itself as a data-driven company. Today that means humans using machine-derived information to drive decision-making. Amazon’s large-scale investments in prime Seattle office space suggest the company doesn’t anticipate that changing anytime soon. But make no mistake, these decisions, and the workflow they enable, are increasingly shifting to machines. Fewer and fewer humans will remain in the decision loop over time. And the places where we will see this shift happen first are in large organizations with deep investments in machine learning.
Places like Amazon.
Pingback: Human vs API: When Remote Software Competes For Your Job | APIscene