Thinking About AI and Work

Artificial Intelligence won’t make humans obsolete in the workforce, says Nik Martelaro, an assistant professor in the Human-Computer Interaction Institute (HCII).

Though powerful, algorithms are rigid, and at some point, technology always fails. As an example, he points to the iPhone, which was first introduced in 2007.

“This is a fantastic product. But still, it screws up all the time,” said Martelaro, whose research focuses on interaction design and interaction with autonomous systems. Instead, Martelaro sees AI’s role in the future of work as a collaborator with humans. We bring dexterity and cognitive flexibility, while AI delivers speed and precision.

HCII researchers explore how humans and computers communicate and interact, including some who are researching how AI can help people do their jobs. A few examples include Françeska Xhakaj’s work creating tools that allow teachers to better identify which students need help, and Adam Perer’s exploration of how machine learning might aid physicians in clinical decisions. Also, Haiyi Zhu researches how AI can address systemic bias to improve the well-being of gig economy workers, such as Uber drivers and Instacart shoppers.

But for AI to make work more efficient, the people already doing these jobs must be integral to designing the technology, said Sarah Fox, assistant professor in the HCII and Director of the Tech Solidarity Lab. In situations where AI is poorly designed, it could degrade or displace work. And that’s very concerning.

For example, Jodi Forlizzi, the Herbert A. Simon Professor in Computer Science and HCI, leads a project that illustrates the importance of centering workers and their experience in the design of AI, in her partnership with UNITE HERE!, the largest hospitality union in the U.S. Other collaborators for the project include Fox, as well as researchers at University of Illinois Urbana-Champaign, Michigan State University and New Mexico State University.

Hotel housekeepers reported frustration with a new algorithmic management software that determined the order in which rooms were to be cleaned. According to the employees, they’d sometimes be scheduled to clean several checkout rooms in a row. These rooms must be fully turned over for new guests, a more labor-intensive assignment than a room that has the same guest for multiple nights.

Fox explains that when housekeepers decide the order of their workloads, they say they are more efficient and tired less quickly. A housekeeper might accomplish this by spacing the cleaning of checkout rooms with other rooms with the same guests staying multiple nights, which requires a lighter touch. Additionally, workers might opt to make all the beds, and then clean the sinks, and then vacuum all the carpets for an entire block of rooms.

The housekeepers reported the platform seemed to be designed by someone who assumed that changing the sheets or cleaning the shower at their own home was the same as working in a hotel, which is not the case. The latter requires a high degree of precision and cleanliness.

“What [the housekeepers] impressed upon us was they believed the designers and technologists who developed this original software didn’t appreciate or recognize their work as expert work,” said Fox.

A more ideal scenario with AI design will include workers from the jump. That’s why Fox collaborates with the AFL-CIO Technology Institute, the Transport Workers Union (TWU) and the Amalgamated Transit Union (ATU) to investigate how AI can be used on city buses. The Technology Institute was established by the AFL-CIO, the national labor federation representing some 12.5 million workers from 60 unions, including TWU and ATU.

Collectively, both unions represent more than 300,000 U.S. workers, and their members include bus operators, train operators, conductors, trackworkers, car cleaners, mechanics and other frontline transit workers.

Martelaro, along with CMU research associate Hunter Akridge and Ph.D. student Alice Tang collaborate on the project.

Adding autonomous vehicle technology to city buses may seem simple since there’s ample research into using AI in four-door sedans. But that is not the case as public transport buses serve a different purpose. At 40 feet long, these accessible vehicles stop every few blocks, even during rush hour traffic.

To include the operators in the design of the AI, roughly 50 TWU and ATU drivers across North America recorded voice diaries of their workdays. In these diaries, the transit workers shared how AI could help them drive safer and increase awareness of their surroundings.

“I always felt like driving is like martial arts,” said one driver in Portland, Oregon. “You never know when a car is gonna pull out in front of you. You never know when a little kid is going to pull out in front of you.”

Drivers also gave suggestions, such as equipping buses with blind spot detection and improvements to lane departure warnings, which monitor when a bus gets too close to the edge of the lane. Fox noted that the current warning systems used in public transit do not factor in the size of buses, making the frequent alarms distracting.

But drivers need more than assistance with technical skills. These civil servants must also care for the welfare of their riders, who include children, those with disabilities and people without homes.

A recent HCII paper examining what would be lost if public transportation was fully automated explored the importance of passenger management. For example, one transit driver recalled an incident in which a frightened girl was being bothered by a visibly intoxicated male passenger.

“It just didn’t sit right with me,” said the driver, who intervened by instructing the girl to move and sit near the front door.

When the bus arrived at the girl’s stop, the driver only opened the front doors but kept the rear exit locked. The intoxicated man was seated toward the back; therefore, he couldn’t follow the girl off the bus. Then, the driver drove away and didn’t let the man exit for another eight blocks despite his yelling threats.

If there hadn’t been a human driver, that girl may not have been able to leave the bus safely. However, some visionaries foresee a future that advocates for fully automated public transportation. It’s already happening to a limited extent in Scotland, though buses are still required to have a human safety driver aboard in case of emergencies.

Part of the AFL-CIO’s motivation for working with HCII is that they don’t want AI to compromise the safety and dignity of transit work. In 2022, Amanda Ballantyne, director of the AFL-CIO Technology Institute, discussed her organization’s collaboration with CMU. She said “the current failed model of corporate-dominated research” too often fuels inequality and leaves workers behind.

That’s a legitimate concern as history is replete with examples of new labor-saving technologies leading to substantial reductions in employment for workers, said Brian Kovak, Professor of Economics and Public Policy in Carnegie Mellon University’s Heinz College of Information Systems and Public Policy. However, Kovak notes that it’s too early to know what tasks AI might be best at, and therefore, it’s hard to predict who is at risk of being displaced.

Rather than replacing  jobs entirely, AI will likely change what types of skills are required for different occupations,” said Kovak.

That’s also how Fox sees things potentially playing out with public transportation and points to some parallels with aviation. While AI has alleviated some stress and strain on pilots, these jobs now require more technical training and expertise.

Beyond having algorithms help with the tedious or labor-intensive aspects of a job, or even designing technology to make work safer, AI can also encourage people to be more creative.