Human + Machine

Reimagining Work in the Age of AI

Paul R. Daugherty, H. James Wilson

• Publisher: Harvard Business Review Press
• Year: 2018
• ISBN:
  9781633693876
• Topic: Computer Science
• Subtopic: Artificial Intelligence (AI) & Semantics
• Index: Computer Science

Insights and advice from the field on how to respond to and leverage the radical transformation AI, robotics, and machine learning are having on all kinds of enterprise processes.

• Articulates the new environment in which AI, machine learning, and bots are increasingly commonplace tools being used across the organization
• Aids managers trying to understand how to help employees do their best work in a quickly changing tech environment
• Shows business leaders more concretely than heretofore how processes and definitions of jobs will change and clarifies potential opportunities at their organization or in their industry
• Provides concrete guidance for the many companies beginning to face the challenges of human-machine collaboration and management
• Engagingly written and accessible
• Broad range of stories from across industries (via Accenture's client list)

Audience: Broad audience of executives and managers interested in how they can improve processes, profitability, and overall impact via AI, robotics, and machine learning. This interest will range across core operations/manufacturing, back office, R&D, marketing, and HR.

Announced first printing: 20,000
Laydown goal: 4,000

PART ONE

Imagining a Future of Humans + Machines . . . Today

1

The Self-Aware Factory Floor

AI in Production, Supply Chain, and Distribution

For centuries, factories have been the paragon of automation. And the people who work in factories, as a result, have often been measured by the standards of machines. So it’s no surprise that the relationship between people and machines in industry has been fraught, with human workers feeling as if they’ve been dealt a losing hand. There’s ample reason for that feeling. Since 2000, the United States has lost five million manufacturing jobs, roughly half of that through efficiency gains and automation.¹

But things aren’t as clear cut as they might first seem. As we discussed in the previous chapter, the second wave of business transformation was all about automating existing processes, and it was during this era that many humans were replaced by machines. In contrast, the third wave relies on adaptive processes that are reimagined from scratch, and the goal here is humans + machines. In this current era, thanks to AI, factories are acquiring a little more humanity: jobs on manufacturing lines, for instance, have changed in nature and are increasing in number. And it’s not just manufacturing. AI is boosting the value of engineers and managers, too. The emergence of AI is also creating brand-new roles and new opportunities for people up and down the industrial value chain.

In this era of reimagining processes with AI, the great irony is that some of the most-automated environments—the factory and other industrial settings—are experiencing a renaissance of human labor. From the assembly-line worker and maintenance specialist to robot engineer and operations manager, AI is rejiggering the concept of what it means to work in an industrial setting. In many cases, AI is freeing up time, creativity, and human capital, essentially letting people work more like humans and less like robots. One implication of the fact that people can work differently and better with the help of AI is that companies are gaining efficiencies and saving money. But perhaps more importantly in the long term is that companies are also starting to rethink their business processes. And as they do, they uncover the need for new kinds of jobs for people, and wholly new ways of doing business, which is our focus in part two of this book.

But let’s not get ahead of ourselves. This is a complex journey. (For some historical perspective, see the sidebar “A Brief History of AI” at the end of this chapter.) Before we rewrite business processes, job descriptions, and business models, we need to answer these questions: what tasks do humans do best, and what do machines do best? There are jobs and tasks that will continue to shift to robots based on their comparative advantages in handling repetition and data processing power. But as we’ll see, the transfer of jobs is not simply one way. In this chapter, we survey a number of companies that have already answered the human-machine question for manufacturing, maintenance, warehouses, and agriculture. These early movers have put people and AI-enhanced machines into play, in the roles that they’re best suited to, and they are reaping the benefits.

The Arm That Learns

The third shift in a Tokyo factory is showtime for an emerging class of robotic arms that can learn new tricks overnight. Coupled with a camera and machine-learning software, these hinged and swiveling appendages can, on their own, figure out the most efficient ways to pick up parts and pieces and set them down somewhere else. No explicit programming is necessary.²

Robotic arms are used in factories to apply hot glue to widgets, to install windshields, and to smooth jagged metal edges, among other tasks. But, traditionally, engineers have preprogrammed them. Then, when robots’ jobs change, engineers must reprogram them. In contrast, the new robotic arms, developed by Fanuc in partnership with software-maker Preferred Networks (both based in Japan), adapt on their own. They do it with an AI technique called deep reinforcement learning, in which the robot is given a picture of the successful outcome and then uses trial and error to figure out its own solution.

According to Shohei Hido, chief research officer at Preferred Networks, the arms take eight hours to become at least 90 percent accurate for this kind of task. This is virtually the same time and accuracy of an expert programming it, but because the arm is now autodidactic, the human expert is now freed to do other more complex tasks, especially those that require human judgment. What’s more, once one robot learns a task, it can share its knowledge with other robots in the network. This means that eight arms working together for one hour can learn as much as one working on a problem for eight hours. Hido, who calls this process “distributed learning,” says, “You can imagine hundreds of factory robots sharing information.”³

Now, imagine people working alongside these robots. Mechanical arms are great for highly repetitive tasks and heavy lifting, but in any factory, there will always be a subset of tasks that are just too complex to hand off to a robot—tasks like positioning numerous small wires or handling awkward or dynamic objects. A human is still needed in the loop.

So how do robot arms and humans work together? Historically, not so well. Robots, with their fast, decisive movements have been helpful and efficient, but also dangerous to people. They’ve often been cordoned off behind protective barriers. But that standard segregation is beginning to change. So-called collaborative robots from companies like Rethink Robotics, founded by robotics and AI pioneer Rodney Brooks, come equipped with sensors that allow them to recognize a range of objects and avoid knocking people around. When robots aren’t so clumsy, they can work well with people. Factories that use Rethink Robotics products often divide the work between the robot and the human worker, working side by side, performing tasks best suited to their abilities. (For further examples of embodied AI, see the sidebar “AI in the Factory.”)

AI in the Factory

For a century, factory floors have been at the leading edge in robotic automation. From conveyor belts to robotic arms to AI-infused operations systems, the factory is getting smarter every day.

• Hitachi is using AI to analyze big data and workers’ routines to inform its robots, which deliver instructions to employees to meet real-time fluctuating demand and on-site kaizen objectives. In a pilot, the company observed an 8 percent productivity improvement in logistics tasks.^a
• At Siemens, armies of spider-styled 3-D printed robots use AI to communicate and collaborate to build things in the company’s Princeton, New Jersey, lab. Each bot is equipped with vision sensors and laser scanners. In aggregate, they join forces to manufacture on the go.^b
• At Inertia Switch, robotic intelligence and sensor fusion enable robot-human collaboration. The manufacturing firm uses Universal Robotics’ robots, which can learn tasks on the go and can flexibly move between tasks, making them handy helpers to humans on the factory floor.^c

a. Dave Gershgorn, “Hitachi Hires Artificially Intelligent Bosses for Their Warehouses,” Popular Science, September 8, 2015, www.popsci.com/hitachi-hires-artificial-intelligence-bosses-for-their-warehouses.

b. Mike Murphy, “Siemens is building a swarm of robot spiders to 3D-print objects together,” Quartz, April 29, 2016, https://qz.com/672708/siemens-is-building-a-swarm-of-robot-spiders-to-3d-print-objects-together/.

c. Robotiq, “Inertia Switch Case Study – Robotiq 2-Finger Adaptive Gripper – ROBOTIQ,” YouTube video, 1:32 minutes, posted July 28, 2014, https://www.youtube.com/watch?v=iJftrfiGyfs.

Kindler, Gentler Robots

During the second AI “winter,” Rodney Brooks challenged one of the fundamental ideas that had driven previous AI research—namely, the reliance on predetermined symbols and relationships between symbols to help computers make sense of the world (see the sidebar “Two AI Winters”). He claimed a much more robust approach: instead of cataloging the world in advance and representing it with symbols, why not survey it with sensors instead? “The world is its own best model,” he wrote in a famous 1990 paper called “Elephants Don’t Play Chess.” (Brooks would later found iRobot, maker of the robotic vacuum Roomba, as well as Rethink Robotics. To date, iRobot has deployed the largest fleet of autonomous robots in the world: between 2002 and 2013, more than 10 million were sold.⁴)

Now, Brooks’s AI philosophy is alive and well in both research and industry. Rethink Robotics, in particular, demonstrates the power of an arm equipped with embedded sensors and algorithms for motion control that allow it to “feel” its way and adjust as it goes. The arm features elastic actuators and back-drivable joints, which means it can flex on contact to absorb energy. Consequently, even if it does knock into something (or someone), it wouldn’t have nearly the force of a traditional robotic arm.

What’s possible when robot arms can learn on their own, as with Fanuc’s products? Or when an arm operates in a kinder, gentler way, as with Rethink’s products?

On the assembly line, workers can collaborate with a self-aware robot arm. Say a worker is putting together a car and needs to put an interior panel on one of its doors. The robot can lift the panel and position it into place, while the worker performs fine adjustments and fastening without fear that a clunky machine will clock him in the head. AI helps both robots and people play to their strengths, and in the process, the assembly line changes shape.

Two AI Winters

The path to human-machine collaboration—a hallmark of the third wave of process improvement—was far from smooth. AI was initially greeted with considerable enthusiasm, only to be followed by results that didn’t live up to the initial hype, and then more progress, leading to a second wave of hype then disappointment. Those down periods have become known as AI’s two “winters.”

The field of AI began in the 1950s, and during the decades that followed any research progress came only in fits and starts. By the 1970s, funding had dissipated so much that the era became known as the first AI winter. Then, during a few years in the 1980s, some researchers made progress in so-called expert systems—computer systems loaded with code that allowed a machine to perform a kind of rudimentary reasoning using “if-then” rules rather than following a strict, predetermined algorithm. But the desktop computer revolution was under way, and attention was diverted toward personal computers as they became increasingly affordable and practical for the average person. Again, money for AI dried up, and the second AI winter descended. It wasn’t until the 2000s that AI began to draw major investment again.

One way that assembly lines can be reconfigured is through AI itself. Engineers at the Fraunhofer Institute of Material Flow and Logistics (IML) have been testing embedded sensors to create self-adapting assembly lines in car plants. Essentially, the line itself can modify the steps in its process to fit the demands of various features and add-ons for highly customizable cars. Thus, instead of engineers designing an assembly line to make one kind of car at a time, these lines can adapt as needed. What’s more, says Andreas Nettsträter, who coordinates strategic initiatives at IML, “If one station has a failure or is broken down, the others could also do what should have been done in this assembly station.”⁵

This means that assembly-line workers are doing tasks that are less robotic (saving those tasks for the robot) and more nuanced, while process engineers don’t need to reconfigure the line every time there’s a change in demand or breakdown of a machine. They can spend their time working on more creative tasks to eke out further efficiencies, for instance.

Follow the Data

What starts with smart arms can extend to an entire factory line and beyond: AI-enabled processes throughout manufacturing and industrial environments are freeing up human potential in a variety of contexts. Maintenance work, for instance, has been forever upended by AI. Sophisticated AI systems predict machine breakdowns before they occur, which means that maintenance workers can spend less time running routine checks and diagnostics and more time fixing a company’s assets. (For other applications, see the sidebars “AI for Faster Machine Onboarding” and “AI in the Field—Unmanned Vehicles.”)

AI for Faster Machine Onboarding

Sight Machine, a startup in San Francisco, uses machine-learning analytics to enable its customers to reduce down-time when adding new machines to a factory floor. In one case, the technology was able to reduce a customer’s downtime, usually inherent in breaking in new robotic systems, by 50 percent. In addition, the net gain was a 25 percent increase in performance when all the assets were up and running. Furthermore, not only does the technology help improve factory efficiency, but it also allows engineers and maintenance workers to spend more time tackling other, higher-value tasks.^a

a. “Jump Capital, GE Ventures, and Two Roads Join $13.5 Million Series B Investment in Sight Machine,” Sight Machine, March 22, 2016, http://sightmachine.com/resources/analytics-news-and-pr...

Table of contents

• Cover
• Title Page
• Copyright Page
• Dedication
• Contents
• Introduction: What’s Our Role in the Age of AI?
• Part One: Imagining a Future of Humans + Machines . . . Today
• Part Two: The Missing Middle
• Conclusion: Creating Your Future in the Human + Machine Era
• Postscript
• Notes
• Index
• Acknowledgments
• About the Authors