Cơn ác mộng tồi tệ nhất của Elon Musk, Missy Cummings, hiện đang hành hạ Waymo và Cruise - Autoblog

Được xuất bản lại bởi Plato

Người theo dõi: 0

Missy Cummings flew fighter jets for the Navy. Now, as a leading expert on automation and AI, she’s taking aim at self-driving cars. Chelsea Jia Feng/Người trong cuộc

Vào năm 2021, một kỹ sư tên là Missy Cummings đã khiến Elon Musk trên mạng xã hội lúc đó được gọi là Twitter. Là giáo sư tại Đại học Duke, Cummings đã tiến hành nghiên cứu về sự an toàn của xe tự lái và những phát hiện này khiến bà đưa ra một số cảnh báo nghiêm khắc về Tesla’s driver-assistance tech. The cars, she wrote, had “variable and often unsafe behaviors” that yêu cầu thử nghiệm nhiều hơn “before such technology is allowed to operate without humans in direct control.” On the strength of her research, Cummings was appointed to the Cục Quản lý An toàn Giao thông Quốc gia Đường cao tốc - để giúp điều chỉnh Robot ô tô.

Tesla fans reacted with their usual equanimity and sense of perspective, by which I mean they absolutely lost it. Their insistence that Cummings would attempt to unfairly regulate their boy Elon soon prompted Musk himself to join the thread. “Objectively,” he tweeted, “her track record is extremely biased against Tesla.” In response, Musk’s stans unleashed their full fury on Cummings — her work, her appearance, her motives. They accused her of conflicts of interest, signed petitions demanding her removal, and emailed death threats.

But the thing is, Musk’s bros of war were messing with the wrong engineer. As one of the Navy’s first female fighter pilots, Cummings used to fly F/A-18s. (Call sign: Shrew.) She wasn’t intimidated by the dick-wagging behavior of a few people on Twitter with anime profile pics. She posted the worst threats on LinkedIn, hired some personal security, and kept right on fighting. “I’m like, are you really going to do this?” she nhớ lại thinking. “I double down. The fighter pilot in me comes out. I love a good fight.”

She didn’t exactly win that particular engagement. A lot of whinging from Tesla đẩy NHTSA buộc Cummings phải tái sử dụng mình từ bất cứ điều gì liên quan đến công ty. Nhưng bạn biết người ta nói gì về bất kỳ cuộc đổ bộ nào mà bạn có thể bỏ đi. Cummings nhận được một công việc mới tại Đại học George Mason và mở rộng nghiên cứu của mình từ Tesla sang thế giới rộng lớn hơn. tất cả các các phương tiện tự lái. Với những công ty như Cruise và Waymo unleashing fully roboticized taxis on the streets of San Francisco and other cities, the rise of the machines has begun — and Cummings is on the front lines of the resistance. In a controversial new paper, she concludes that the new robot taxis are four to eight times as likely as a human-driven car to get into a crash. And that doesn’t count the way self-driving vehicles are causing ùn tắc giao thông kỳ lạ, chặn xe cấp cứu, Và thậm chí dừng lại trên người người đã bị một chiếc ô tô do con người điều khiển đâm phải.

“In the paper that really pissed all the Tesla trolls off, I actually say that this is not just a Tesla problem — that Tesla is the first one to experience the problems,” Cummings tells me. “For years I have been telling people this was going to happen, that these problems would show up in self-driving. And indeed they are. If anyone in the xe tự lái community is surprised, that’s on them.”

It turns out that serving in the Navy is a very good way to train for inbound ire from Muskovites. In her 1999 memoir, “Hornet’s Nest,” Cummings recalls how she loved flying jets, and says the excitement of getting catapulted off an aircraft carrier — or landing on one — never got old. But the environment was far from welcoming. Sexual harassment in the Navy was routine, and male colleagues repeatedly told Cummings she wasn’t qualified to fly fighters simply because she was a woman. When she and another female officer showed up at a golf tournament on base, they were told to put on Hooters uniforms and drive the beer carts. Cummings declined.

Động cơ hủy diệt chiến thuật bay cũng mang đến cho Cummings bài học trực tiếp về những mối nguy hiểm tiềm ẩn của máy móc, tự động hóa và giao diện người dùng. Vào ngày huấn luyện đầu tiên, hai phi công đã thiệt mạng. Vào ngày cuối cùng của cô, Hải quân đã trải qua thảm họa huấn luyện tồi tệ nhất từng xảy ra trên một tàu sân bay. Tổng cộng, trong ba năm Cummings bay, có 36 người chết vì tai nạn.

In 2011, while conducting research on robot helicopters for the Navy, Cummings had an epiphany. Even surrounded by nothing but air, those helos were far from perfect — and they relied on the same sensors that self-driving cars do while operating right next to cars and people. “When I got in deep on the capabilities of those sensors,” Cummings says, “that’s when I woke up and said, whoa, we have a serious problem in cars.”

Some of the dangers are technical. People get distracted, self-driving systems get confused in complicated environments, and so on. But other dangers, Cummings says, are more subtle — “sociotechnical,” as she puts it. What she calls the “hypermasculine culture in Silicon Valley” intertwines with Big Tech’s mission statement to “move fast and break things.” Both bro culture and a disruptive mindset, as she sees it, incentivize companies to gloss over safety risks.

All of which makes it even tougher for women when they level the kind of critiques that Cummings has. “When Elon Musk sicced his minions on me, the misogyny about me as a woman, my name — it got very dark very quickly,” she recalls. “I think the quân sự has made a lot of strides, but I do think that’s what’s happening in these Silicon Valley companies is just a reminder that we haven’t come as far in our society as I thought we would have.”

Một ví dụ: Tháng trước, người đứng đầu bộ phận an toàn tại Waymo đã giới thiệu một nghiên cứu mới của công ty ông trên LinkedIn. Nghiên cứu chưa được công bố và chưa trải qua đánh giá ngang hàng. Nhưng Waymo đã sử dụng nghiên cứu này để lập luận rằng những chiếc xe robot của nó thực sự ít có khả năng gặp tai nạn hơn những chiếc xe được điều khiển bởi các sinh vật sinh học như bạn và tôi.

A self-driving taxi from Waymo blocks traffic in San Francisco. “They got complacent,” says Cummings. “They lost their safety culture.” Terry Chea/AP

Cummings wasn’t having it. She had her new results — also still in preprint — which showed self-driving taxis to be way more crash-prone. So she went on LinkedIn, too, and said so.

Câu trả lời này đã quen thuộc với cô từ những ngày còn ở trong Hải quân. Kyle Vogt, CEO của Cruise, trượt vào phần bình luận. “I’d love to help you with this analysis,” he wrote to Cummings, questioning her number-crunching. “Would be great to connect and discuss this further.”

Cummings responded in kind. “I’d love to help you with your understanding of basic statistics, use of computer vision, and what it means to be a safe and responsible CEO of a company,” cô ấy viết. “Call anytime.”

Women, she figures, caught her vibe. “Every woman who read that was like: Mmm-hmm, you go,” Cummings says. But men — friends in Silicon Valley — did not. They thought she had been too mean to Vogt. “He was just trying to help you,” they told her.

“All the guys read it like: She’s such a shrew!” Cummings says. But, ever the fighter pilot, she was unfazed. “That’s how I got my call sign,” she says. “So I live with it.”

So who’s right: Cummings, or the self-driven men of Waymo and Cruise and Tesla? It’s hard to tell, for a simple reason: The data on the safety of robot cars sucks.

Take Cummings’ approach in her new paper. First she had to wrestle with NHTSA's dữ liệu toàn quốc về các vụ tai nạn không gây tử vong do người lái xe gây ra, để có được những con số mà cô có thể so sánh với California, nơi duy nhất mà ô tô robot chạy tự do. Sau đó, cô phải tìm ra những con số va chạm không gây tử vong và số dặm đã di chuyển của Waymo và Cruise, được theo dõi bởi các nguồn khác nhau. Kết luận của cô: Cruise gặp 8 vụ va chạm không gây tử vong đối với mỗi người và Waymo có 4 vụ - tương đương với tỷ lệ va chạm của những người lái xe mệt mỏi và làm việc quá sức tại các dịch vụ gọi xe như Uber và Lyft.

The purveyors of robot taxis argue that Cummings is wrong for a bunch of reasons. Chiefly, they say, the numbers for human crashes are actually undercounts. (Lots of fender benders, for instance, go unreported.) Plus, crash numbers for the whole country, or even just California, can’t be compared to those for San Francisco, which is way denser and hillier than the state as a whole. Looked at that way, Cruise argued in a recent blog post, its taxis have been involved in 54% fewer crashes than cars driven by humans. The company also maintains that ride-hail drivers get into one nonfatal crash for every 85,027 miles of driving — 74% more collisions than Cruise’s robots.

Cummings ain’t buying it. A blog post isn’t science; it’s a press release. “Every company has a fiscal interest in getting a paper out that makes them look good, and in the case of Cruise it makes rideshare drivers look bad,” she says. “So that’s what they’re doing.” This is exactly the sort of sociotechnical culture that Cummings is criticizing — that she’s uniquely qualified to criticize.

Other experts also discount Cruise’s claims, coming as they do from folks who are incentivized to welcome our new robot overlords. “If we were to believe the numbers Cruise is putting out there for ride-hailing drivers, those drivers would be having on average two crashes per year,” says Steven Shladover, a research engineer at UC Berkeley’s Institute of Transportation Studies. “How many drivers have two crashes every year? That is pretty extreme.”

But Shladover is also skeptical of the numbers crunched by Cummings. “Missy is assuming a human driver crash rate that’s too low for San Francisco, and Cruise is showing a human crash rate that’s too high,” he says. “The reality is probably somewhere in between.”

So maybe Cummings is right, and self-driving cars are a menace. Or maybe it’s not quite as bad as her new paper suggests. Until robot cars have traveled for hundreds of millions of miles, there’s no way to get a statistically significant, unequivocal conclusion. But the bottom line is: It shouldn’t matter. When the data on a product or device’s safety is equivocal, regulatory agencies are supposed to make and enforce rules that protect consumers, just as they do in other industries. If the data on robot cars is equivocal or incomplete, then those rules should keep them off the road. The burden of proof is on Waymo and Cruise and Tesla, not Missy Cummings. And if those companies want to put 2-ton robots on public streets, blogging about data benchmarks isn’t the way to show people they’re ready.

“One of the big things I’m on about now, pulling from my aviation years, is that all these companies need a chief AI pilot,” Cummings says. “They need to have somebody, one person, who stands up and says, ‘I’m responsible.’ We do that right now for aviation. That’s why so many heads rolled with the problems that happened with the Boeing 737 Max. They got complacent. They lost their safety culture.”

Cummings is a careful researcher. She’s also, as one transport-safety researcher put it privately, “provocative.” She is more than happy to strafe companies like Tesla and Waymo and Cruise, and to argue that tech bros need to be brought inside a stricter regulatory framework. In a sense, she’s Elon Musk’s worst nightmare. She has repeatedly and routinely risked her life to test the incredible capabilities — and the lethal limits — of human-machine interfaces. And she did it in an environment where the stakes are far higher than the battlefields of Twitter and LinkedIn. To her, the safety of self-driving cars is not an abstract question. It’s a matter of life and death.

“I’m a tenured professor. My work speaks for itself. I’m trying to save your life, right?" Cummings says. “And there’s the side of me where I’m like Don Quixote on steroids. There’s no windmill I don’t want to tilt at.”

Adam Rogers là phóng viên cấp cao của Insider.

Đọc bài viết gốc trên Business Insider