Are People Demanding Too Much Safety for Self-Driving Cars?

Although AV technology promises numerous benefits, concerns over safety and trust have become the defining issue. The industry recognizes this. “People are ready to embrace new vehicle technology, especially if it will make driving safer,” said Greg Brannon, AAA’s director of automotive engineering and industry relations, upon the release of results from its latest public-opinion survey in February. As the technology improves, AVs promise extraordinary safety benefits. A 90% reduction in traffic fatalities—cited in a 2015 report from McKinsey—is the most frequently repeated figure.

But these promises are many years, if not decades, from being realized. The challenging part is how to get from here to there. A reasonable timeline for when AVs should be rolled out en masse is when they are at least as safe as the average driver. As soon as AVs exceed this threshold, then not only would we reap all their economic and social benefits, but we would also be saving lives. And importantly, we would rapidly begin to accumulate the data needed to drive AV safety levels up even further.

High expectations

Unfortunately, the broad acceptance of just-better-than-average AVs may be undermined by a host of psychological biases that fester in the minds of consumers. We investigated two of these biases in an article we published on March 15. We asked a representative group of Americans how much safer AVs would have to be, compared with human drivers, for them to be willing to ride in such cars. The results revealed that people’s demands for safety are extremely—potentially unrealistically—high.

If AVs were to eliminate 10% of today’s accidents, thus being substantially safer than the average human, only 11% of people would be willing to adopt them. Even if AVs caused accidents at half the rate of human drivers, only 37% of Americans would opt in. To make matters worse, about 15% of people would need AVs to have a perfect safety record before they opt in.

What might cause such inflated safety demands from AVs? We identify two primary reasons.

First, there’s overconfidence. Known in the psychological literature as “illusory superiority” or the “better-than-average effect,” the phenomenon is that people have an inflated perception of their own skills compared with others. Driving skill is literally the textbook example. Indeed, in our study, the majority of respondents thought if everyone drove as they did, 66% of accidents would be eliminated. This pattern was true for both men and women, in every age group, and for all education levels. And the higher people rated their own driving safety, the safer they wanted AVs to be before they were willing to adopt them.

In addition to overconfidence, we found evidence for a phenomenon called “algorithm aversion.” It refers to the reluctance of humans to use superior but imperfect algorithms. Suppose safety data was given not only for AVs, but also for humans, such as taxi and Uber drivers. Our study participants showed more aversion to riding with AVs than to riding with a human driver, even if these two modes of transportation were equally safe.

Is there anything we can do to reduce people’s excessive expectations for AV safety? We tried a de-biasing technique designed to help people recognize their tendency to inflate their own driving skills. Interestingly, this intervention succeeded in reducing people’s assessment of their own driving skills, though not completely: People now only thought they were, on average, in the 60th percentile of drivers. Unfortunately, however, they didn’t budge when it came to how safe they wanted AVs to be. This suggests that people’s safety requirements for AVs may be particularly sticky, and more research is needed to find ways to calibrate them.

This leaves decisions about safety thresholds for AVs in a difficult spot. A 2017 Rand Corp. study illustrated the consequences of an excessive safety threshold by simulating 500 possible future scenarios. In almost all scenarios, if we wait for AVs to be 90% (rather than 10%) safer than the average human driver, we risk failing to save thousands, if not hundreds of thousands, of lives in the long term that could be lost while we perfect AV technology.

However, what we show is that if regulators were to allow the broad deployment of 10%-safer AVs, psychological biases could interfere with large-scale adoption and thus prove to be a substantial public health hazard insofar as they prolong the era of relatively dangerous human driving.

A different message

Focusing the discussion too much on safety may be one of the problems. Given their overconfident, algorithmically averse psychologies, consumers are unlikely to be won over by exaggerated promises about the safety benefits—especially as these promises sit uneasily alongside the mounting numbers of AV-involved crashes. Casting the purpose of AVs as products that increase safety (rather than, say, convenience) could even cause a backfire effect.

We mentioned just two psychological biases above, but there are others. People experience what theorists Jonathan Koehler and Andrew Gershoff have called “betrayal aversion.” When the primary function of a product is supposed to be safety, people are especially incensed when it causes harm. In one study, people awarded larger punitive damages when a house fire was caused by a smoke alarm (meant to keep you safe) than when it was caused by a refrigerator (meant to keep your milk cold). Betrayal aversion also explains the oversize reaction people have to the side effects of vaccines. In another study, participants preferred a vaccine with a 2% chance of being ineffective to a different vaccine with a 1% chance of being ineffective but also a 0.01% chance of causing harm itself.

The clear implication is that the more AVs are marketed as safe, the more betrayed people may feel when the cars occasionally—and inevitably—fail to achieve that goal. That, in turn, could slow the spread of AVs, and fail to save even more lives in the long term.

Before the pandemic, the average American spent over 52 minutes a day commuting—time psychologists have found is the least happy part of their day. AVs promise to give us that time back—for working, for sleeping, for all sorts of activities. It’s the appeal of that, not safety, that inspires Tesla enthusiasts to shell out thousands of dollars for Autopilot. While safety should always be the priority of the AV industry, perhaps the communication about AVs should focus more on these other benefits.

Dr. Rahwan is a director of the Max Planck Institute for Human Development in Berlin. Dr. Shariff is an associate professor and Canada 150 research chair in moral psychology at the University of British Columbia in Vancouver. Jean-François Bonnefon, a research director at the Toulouse School of Economics in France and the former president of the European Commission independent expert group on the ethics of driverless mobility, also contributed to this article. They can be reached at [email protected].