With automated vehicles succeeding (so far) as taxis in San Francisco and Phoenix, it’s only natural to ask: What’s next? Pundits and researchers have long wondered whether autonomous vehicle technology can transform the world of public transit, which is beleaguered with staffing and financial problems despite a post-pandemic recovery in ridership. “Transit agencies are currently facing big financial deficits while also struggling to attract drivers at current wages and also facing pressure to add security personnel,” Matthew Yglesias recently wrote in his Substack newsletter. “So in this case, there’s no big labor market issue at all, it’s just a win.” I called up Sarah Fox and Nikolas Martelaro, researchers at Carnegie Mellon University and authors of a 2022 policy paper on automated public transit, to ask them exactly how close, or far, we might be from that “win.” As it turns out, it’s a little bit more complicated than simply taking the driver out of every bus. An edited and condensed version of the conversation follows: How satisfied are you with the government’s response to the recommendations coming out of the policy community as AV activity has ramped up? Sarah Fox: One thing we have consistently wanted is more robust reporting mechanisms. I still haven't seen a lot of improvement in that space. I'll just put it that way. With the reporting around some of the incidents in San Francisco, it seems like reporting mechanisms were not particularly well thought through. Nikolas Martelaro: Often the way we're learning about the results of pilot programs is through some type of report, but these are not mandated. Often the funding is given from the federal level, and then the states administer it, and they all have different standards on how they want to record and get data and information back and how they want to publish it. One of the most interesting reports we've actually read is from the National Park Service in Yellowstone. They deployed a small autonomous shuttle, and came up with all types of fun facts. You have to deal with bears, and GPS goes out, and there are fender benders, which help you fix blind spots. Given current workforce trends, is it possible that autonomous transit could be a boon both to the transit workforce and to riders? SF: That’s certainly a part of the rhetoric around these pilots. In a new paper, we concentrate on the work that bus operators perform right now, and how it could on a technical level confound the capacities of AVs, but also on a social level mean a lot for the fabric of our cities. Communities have to decide what role they see public transit playing. Operators do a lot, including, addressing medical emergencies, acting as first responders for health emergencies and road emergencies, and serving a population of people who largely don't have access to personal vehicles. Bus operations is the oldest on average profession within transportation, and employs a large proportion of Black transit workers. It's critical to think about the impact of this technology on the workforce and employability of trained operators, and who we’re really talking about when we talk about that. NM: The point about transit companies realizing they’re not able to hire workers, that’s real. If you were to talk to the major transit unions, their argument might be that if you pay better you'll be able to hire people. What data do you have on how drivers feel about parts of the prospect of their job being at least partially automated? SF: People take pride in operating a vehicle of this kind of size, weight and magnitude, and the difficulty of doing that job and how rewarding it is. Reshaping the role into something that is more akin to an attendant is not particularly appealing to them. It would also likely come with less pay, because it would require less in terms of a commercial license, credentialing or continued training. Fundamentally, if you’re going to keep a driver on a bus, why automate it at all? NM: There are technologies in the automation stack, like lane departure warnings or automatic front braking, that many drivers have in their personal vehicles, and have said they would like their bus to have. There are opportunities for well-designed automation that can support a driver. The challenge is in making sure that it supports a driver and doesn’t take away from their ability to successfully and safely navigate. SF: One driver said that she had a lane departure warning system on her bus that was not well calibrated for its size, but that if it was designed well, it would be useful. That’s a very different version of an automated future that’s not about driverlessness, but humans and machines learning and thinking about ways automation can specifically enhance safety. You’re both human-machine interaction researchers. What has this field taught you about how people interact with AI? NM: When systems are designed without this idea of a human-machine team, that’s when you have problems. One problem you have is called mode error, where people don’t understand how much automation is happening in a vehicle, and then something wrong happens where they were supposed to be in more control. Another is skill atrophy, which happens in aviation, where when you don’t manually have to pilot something you become less good at it because you’re not driving all the time. SF: But in aviation, you have more time to respond. In a car or a bus, you have seconds.
| 
