Powerless in the Driver's Seat

I booked a Waymo in four minutes. It arrived in four minutes. I unlocked it from the Uber app, got in the back seat, and a voice welcomed me by my first name. The whole thing felt pretty seamless and professional. A tutorial video played as the car started moving, teaching me what to do during the ride. I watched the steering wheel turn on its own in awe. After two minutes, surprisingly, I felt comfortable enough to stop paying attention to the road. That first ride was a very good presentation of Waymo's objective. Just me and a car that seemed to know exactly what it was doing. I even discovered a forgotten set of keys in the backseat from the previous passenger, which Waymo's external speakers helpfully reminded me about before I left, assuming those to be my keys. Small details like this made me impressed with the level of innovation happening in front of my eyes.

However, just three rides and something shifted. I was running late to an event, and the Waymo was taking me around campus instead of toward my destination. The most direct route had a blocked-off section of road, so the car was trying to navigate around it. I could see the block was temporary, and if a human driver were taking me, I would have said, just drop me at the nearest point, I'll walk from there. Instead, the Waymo insisted on reaching my exact destination. It kept trying to turn right onto the blocked road. The car wouldn't accept that the road was, in fact, blocked. At some point, a police officer came over to the car, probably to tell "us"¹ to move along or find a different route. The officer was waving at "us" to move along, speaking to the vehicle in hopes it understood what "straight" meant, clearly trying to help redirect the vehicle. But the Waymo just kept trying to turn. The cop kept signaling. This dance went on long enough that the officer eventually gave up and moved the barrier to let the Waymo through. I was a passenger watching an algorithm do what it was programmed to do, and I was completely powerless to stop it or explain the situation. The moment I unbuckled my seatbelt to get out, the car stopped and alerted me to call for support if I wouldn't buckle back up. I sat there, feeling trapped, unable to leave a vehicle that wouldn't listen to reason.

A police officer had to physically move a barrier because a car wouldn't adapt.

When I started using Waymo, I was drawn to the promise they represent: a technology designed to remove human error, to save lives from drunk driving, distraction, and exhaustion. Those are real problems. Thousands of people die every year in car accidents caused by human negligence. Removing that risk sounds almost obviously good. But in that moment on the blocked road, I realized something else was happening. The algorithm made a decision and committed to it. It couldn't interpret the context, and it couldn't see that the barrier was temporary or understand that a passenger might have different priorities. I could not talk to it or give it more context. It just kept trying the same thing, over and over, because that's what the data told it to do. Humans see a blocked road and adjust. We notice a police officer trying to help us and recalibrate. We understand that sometimes the perfect solution isn't worth the cost of getting it. We make calls based on context, in the moment, on things that don't fit into a predetermined set of rules.

What's harder to sit with is that this matters differently depending on who you are. I was uncomfortable and late to an appointment. I had the option, however limited, to complain about the experience. I could write about it and I could choose not to use Waymo again. The pedestrian on the street corner didn't get any of those choices. Neither did the police officer who had to intervene. They didn't choose to interact with an algorithm that couldn't adapt. They were just in the environment where this machine was learning to drive, and they had to trust that someone, somewhere, had made good decisions about what it would do when things got confusing.

Let me be clear: I believe that Waymo will improve and I am very excited for the future of autonomous cars, and perhaps I was simply unfortunate in this experience, but I keep thinking about trust. When I get into a Waymo, I'm consenting to a kind of algorithmic decision-making I don't fully understand. But pedestrians walking down a street can't consent. They're just trusting that the technology was built safely. And that trust feels lopsided when the algorithm can't think in real time the way a human can. Look at the technology around you. What does trusting it look like for you?

---

1. While writing this, I debated on whether to call the Waymo, JARVIS, for much longer than you would think. ↩