Waymo’s autonomous vehicles have been navigating the streets of London for some time now. Currently, these cars are not ferrying passengers and are equipped with a human driver ready to take control if the situation demands. During my encounters, these AI-driven cars displayed cautious and predictable behavior, but there was a startling incident recently.

While cycling home from work, I approached a roundabout when a Waymo vehicle attempted to park right in front of me. I had the right of way, and it seemed like the car was heading straight for me without showing any signs of slowing down until it abruptly stopped.

This wasn’t the scariest experience I’ve had while biking, and I felt relatively safe because I saw it coming and was prepared. However, this incident was notable among the numerous close calls that often occur during my commutes, primarily because the driver was an AI, not a human.

Out of curiosity, I reported the time and location of the incident to Waymo, asking for clarification. In their response, an ADS (Automated Driving System) spokesperson stated, “We detected a bicycle approaching from the right side of the roundabout. The safety driver took manual control of the vehicle and applied the brakes at the roundabout entrance.” This highlighted that human intervention was necessary because the AI was not going to stop on its own.

I inquired how frequently humans have to step in during these situations in London, but Waymo reported that they don’t have that data and do not primarily use it as a performance metric. When I asked what had confused the AI, considering that my neon yellow bicycle was highly visible, I received no response.

Other cyclists in London reported positive experiences with Waymo’s behavior around them, though one mentioned a specific instance where a roadwork sign stating “narrow lanes—no overtaking cyclists” seemed to confuse the vehicle, leading it to pass uncomfortably close. Simon Munk from the London Cycling Campaign emphasized that, overall, these cars exhibit cautious behavior around cyclists.

Last year, I reflected on AI drivers when I discovered their introduction to London streets. I pondered whether having AI drivers, albeit imperfect, could be safer than human drivers who may be tired, distracted, or hostile. Having previously been knocked off my bike by human drivers, my reservations about Waymo still stand. Interestingly, the same AI driver once managed to navigate through police tape blocking a roadway, although the company claimed at the time that it was being monitored by a human.

While I harbor doubts about AI effectively navigating amidst human traffic, I also have a lack of trust in human drivers. The capacity for AI to learn quickly is encouraging, and advancements in sensor technology will further enhance their hazard detection capabilities. Whether AI drivers are safer than human counterparts remains to be seen; however, I believe they hold promise for the future.

Munk articulated that the rise of self-driving cars presents both advantages and challenges for cyclists and pedestrians, with the potential to make roads safer or, conversely, further endanger vulnerable road users. He noted that while Waymo has made strides in cities like San Francisco, London presents an entirely different level of complexity and unpredictability. Munk proudly shared a video showcasing one of Waymo’s cars adeptly navigating a chaotic San Francisco intersection, but admitted that his daily commute faced far worse traffic conditions.

Additionally, Munk expressed concern that once these self-driving cars commence operations, future software updates could alter their behavior, even if they initially seem safe. It’s essential that driverless cars undergo continuous monitoring to ensure safety and accountability for public welfare.

Other players in the self-driving car industry are expected to enter the London market soon. Wayve’s driverless taxis are anticipated to arrive in the upcoming months, though their approach remains uncertain. As with human drivers, the safety levels of AI can vary significantly; corporations design AI behaviors based on desired outcomes. Last year, Waymo even adjusted its algorithms, leading the vehicles to behave like “aggressive New York taxi drivers” to mitigate traffic congestion, a move that unsettled many San Francisco residents.

We have also witnessed dubious proposals aimed at alleviating road safety concerns, like suggesting pedestrians wear special vests to enhance visibility for autonomous vehicles.

Munk raised thought-provoking questions about the future: “In five or ten years, if these scenarios become commonplace, will companies prioritize speeding up passenger transit at the potential cost of safety?” This emphasizes the critical need for caution in developing and integrating AI in public transportation.