This transcript was prepared by a transcription service. This version may not be in its final form and may be updated.
Zoe Thomas: This is your Tech News Briefing for Friday, September 24th. I’m Zoe Thomas for the Wall Street Journal. Would you get into a robot taxi? Well, a company called Zoox wants you to. It’s developing an autonomous taxi that doesn’t look a whole lot like a regular car, more like a carriage but without the horses. Zoox’s technology caught the eye of Amazon. The e-commerce giant bought the startup last year. But while Zoox’s vehicles are being tested, the company hasn’t said when they’ll go to market. Meanwhile, some of its competitors are already offering rides in some locations.
On today’s show, a common with Zoox co-founder and CTO, Jesse Levinson, about the company’s technology, competition and the all important question, how do you get people to trust self-driving taxis. That’s after these headlines.
A federal judge ordered Facebook to turn over record related to accounts that the platform shut down in 2018 that were linked to the genocide in Myanmar. The government of Gambia is seeking the records to help it pursue an international criminal case against the government of Myanmar over its role in the killing of at least 10,000 people from the Muslim Rohingya minority. Our reporter Aruna Viswanatha has been covering the case.
Aruna Viswanatha: Aruna Viswanatha potentially gives the people who have been allegedly wronged by other users on Facebook, women who are trafficked, other potential victims of political violence. Facebook takes down these posts because they believe that they violate Facebook laws, but then it has been holding onto them and this ruling could potentially give an avenue to those alleged victims to get more information from Facebook.
Zoe Thomas: Facebook had objected to turning the information on deleted accounts over, saying it violated U.S. privacy law. A spokeswoman said the company was reviewing the decision. The ruling comes as Facebook faces scrutiny over the real world impacts its platform can have. A Wall Street Journal investigation found Facebook employees raised concerns about how the platform was being used in developing countries, including in ways that could spark violence.
We’ve been keeping an eye on the trial of Elizabeth Holmes, the founder of defunct blood testing company, Theranos. She’s accused of defrauding investors and patients by lying about the ability of her company’s technology. She’s pleaded not guilty to all charges this week. Prosecutors called former Defense Secretary Jim Mattis to testify. Mattis had advocated for using Theranos devices in the battlefield, and when he retired from the military, joined Therano’s board. Our reporter Sara Randazzo was in the courtroom.
Sara Randazzo: This testimony is crucial for prosecutors for a few reasons. First, in the indictment, they alleged that Elizabeth Holmes told investors that they had a profitable relationship with the Defense Department, and this testimony really got to the point that there was never a partnership that went anywhere with the military. It also shows some of the deceptions that she allegedly made to board members. So for that reason, Mr. Mattis’ testimony really got to the heart of both of those allegations.
Zoe Thomas: In the coming weeks, the prosecution is expected to call more former employees, patients, and medical practitioners. All right, coming up. In the future, jaywalking may get you a honk from a self-driving cab. We talked to a company working to make autonomous taxis a reality, and maybe even a little bit assertive. That’s after the break.
Imagine in the future, you need to hail a cab. You open up an app on your phone. You request a taxi from your location to where you want to go. When the vehicle shows up, it looks more like a rolling gondola, with two benches facing each other and no driver seat. It drops you at your destination and then wisps away to pick up another passenger. That’s the future that Amazon’s self-driving taxi company Zoox is trying to develop. But there are a lot of steps to making that system safer riders and ready for the highly competitive market. I spoke with Jesse Levinson, the CTO and co-founder of Zoox for the Wall Street Journal’s recent future of transportation event and put these issues to him. Welcome Jesse. Thanks for joining us.
Jesse Levinson: Hey, thanks so much. Good to be on.
Zoe Thomas: Let’s get into what’s probably the biggest issue that people think of when they’re talking about self-driving cars and that’s safety. There have been a number of incidents involving parts of self-driving cars, some crashes, some fatalities, and not involving Zoox. But how do you build up customer confidence in the safety of your vehicles given what’s going on?
Jesse Levinson: So the first thing we have to do is be safe. Then it’s really on us to explain to regulators and to the public why we’re very safe. We can do that through many mechanisms. We can actually show videos of the vehicle sensor data and show that it’s able to perceive its surroundings 360 degrees many, many times every second, and never lose track of any object, which is a pretty amazing superhuman capability. The other thing is our vehicle itself is designed to be the safest vehicle on the road, even from a passive crash perspective. So you’re kind of in this cocoon. There are airbags all around you inside the vehicle. We’ve done all the crash tests. This vehicle is designed to safely go up at 75 miles an hour in both directions.
Zoe Thomas: One of the other issues around safety has to do with the fact that Zoox are electric vehicles. A lot of other EV companies are having issues with battery fires. What is Zoox doing to avoid similar problems?
Jesse Levinson: Incidentally, sometimes there’s a trade off where if you try to pack the absolute most density you possibly can into a given space, that can increase the chances of thermal events. We’re not trying to push that envelope in the most extreme way possible. Partly because by reimagining the architecture and shape of the vehicle, we’ve been able to devote more space to the batteries and that means that we don’t have to make them insanely dense.
Zoe Thomas: I want to ask you a scenario question. For anybody who’s been to busy, dense cities which Zook says it’s designed for, you know that sometimes pedestrians don’t exactly follow the rules. So if you’re in a Zoox vehicle and you’re trying to make a turn through say a crowded walkway, how is it going to do that? Because a human driver could nudge its way through. It could maybe signal to people that are walking through, but without a human driver, how is the computer going to know to take those risk calculations?
Jesse Levinson: We’ve given ourselves some really handy tools. One of them is using light and sound to communicate with other agents. We have this light bar that can change colors and it can blink, it can flash, it can be on, it can be off. You can use the light bar to signal intent. What’s perhaps even cooler is we have a multi speaker array on both front end of the vehicle and you can do something called beam forming. What this means is you can shoot sound in any particular direction you want and it’s quite loud in the direction you’re shooting the sound, but it’s almost inaudible everywhere else. so if there’s a particular agent like maybe a pedestrian who maybe hasn’t paid attention and doesn’t realize what’s going on and they step out into the road, you can actually ping them.
Zoe Thomas: Are you talking about directional honking? You’re just going to honk right at me?
Jesse Levinson: That’s exactly right, except it doesn’t necessarily have to be an obnoxious honk. So that’s one of the tools. Then the AI has to be good at negotiating those types of scenarios, including being assertive. If you’re trying to drive in the suburbs, you can probably get away with waiting for everything to clear and then you’re like, okay, there’s nothing there. I’m going to proceed. If you’re driving in downtown San Francisco, for example, you can’t necessarily wait for there to be nothing around you because you’ll be waiting forever and then people will be honking at you and they’ll be trying to drive around you. That actually creates a less safe situation overall.
So our vehicle has to be trained to nudge, to be assertive, to negotiate with pedestrians. What’s really cool is we can actually do that better than humans because we always tracking all the agents around us 360 degrees, sometimes several hundred agents at the very same time.
Zoe Thomas: I want to stay on this topic of crowded cities, Zoox is going to be operating in places with diverse populations and under different conditions. We know that AI systems can have problems with diversity. For example, if developers don’t maybe put enough variety into the datasets that they’re dealing with, how are you thinking about that and accounting for that in Zoox’s development?
Jesse Levinson: So the first thing is to make sure we’re training our system on huge amounts of data. So we drive our vehicles all over the geo-fence in different weather conditions, lighting conditions. The next thing we can do is in simulation, we can create scenarios and types of agents and people that we maybe haven’t even seen in real life. We can use those to train our system. Probably the most important thing we can do is actually have a diversity of sensor modalities that perceive the world in different ways.
Computer vision is probably the most sensitive to some of these diversity issues because it’s very, very sensitive to the mix of training data that you feed the algorithms. If you see an object that looks meaningfully different from the ones you’ve seen before, it or may not work well. This is one of the fundamental limitations of computer vision as we know it today. If you have a camera-based system and then you have somebody with dark skin wearing dark clothing at night, it’s possible that you just don’t get enough photons reflecting back for your computer vision algorithm to be confident that there’s an object there.
What our system does is it actually shoots light out. So if you think about radars and LiDARs, they’re what’s called active sensors. That means that they’re emitting energy into the environment and they’re looking for reflections. So you can have somebody with arbitrarily dark clothing, dark skin and it can be pitch black outside with not a single light and we’re still not only going to see them, but we’re going to measure their exact distance, shape and size, and that gives us extraordinary confidence that we are not going to run into them.
Zoe Thomas: Some of Zoox’s competitors are already on the road. So when is Zoox going to be?
Jesse Levinson: We don’t have a arbitrary window to develop and commercialize our technology. If our competitors have a widespread robo taxi networks in three years and ours come out in 50 years, then that’s too late, right? We are not conceding, by the way, the opportunity to be the first company to deploy ground up robo taxis in cities at scale. I think we have some tremendous advantages. I mean, we’re the first company in the world that’s shown a ground of vehicle that actually moves and drives autonomously. So we’re pretty proud of that. Not taking that for granted, but that’s a pretty good accomplishment so far.
The other thing is that it’s not always the very first technology that’s the most commercially successful. If you look at the iPhone, for example, it was far from the first smartphone, but the iPhone was such a thoughtful integrated approach to solving the problem and it was such a better device that they end up pretty much taking over the industry.
Zoe Thomas: Jesse Levinson, CTO and co-founder of Zoox. Thank you so much for joining us for this conversation.
Jesse Levinson: Thanks so much for having me on.
Zoe Thomas: All right. That’s it for Tech News Briefing this week. Our producer is Julie Chang, our supervising producer is Chris Insley, our executive producer is Katara Yokum, and I’m your host, Zoe Thomas. Thanks for listening and have a great weekend.