Why Uber CEO Dara Khosrowshahi Is Ready to Slow Down to Get Autonomous Vehicles in the Right Lane

It’s coming up to a decade since Uber Technologies announced its first major push into autonomous vehicles (AV), announcing in 2016 a deal with Volvo to deliver self-driving cars, albeit ones with a human behind the wheel for safety. Flash forward nine years and while the firm that disrupted the traditional hired taxi market has diversified operationally, the autonomous vehicle space remains a priority in the AI age, with CEO Dara Khosrowshahi staking a claim to differentiators from other contenders in the field: > What we're seeing as it relates to the development of AV models and physical AV in particular is that end-to-end models are replacing the heuristics approach, A bunch of these companies, including Waymo, Nuro, etc, they have been building AV for years and years, but it was based on heuristics. It was based on if-then kind of logical functions. That is quickly being replaced by end-to-end models that take in enormous amounts of data and then make decisions as humans do. What Uber wants to do with its AV Labs is accelerate the development and the training of these models on an L4 basis [the external escalation boundary of IT support], efficiently with real kind of data from rideshare-specific scenarios. To that end: > We're building out hundreds of cars that are riding in rideshare-specific scenarios with robo-taxi-grade sensors, and we're collecting a super-set of data that then we can provide to all of our partners. Partnering That partner ecosystem brings its own issues to factor in, he points out: > Each partner has to collect the unique data sets, to go out and acquire all of the tail data that you need to train on to make sure that your AV driver is safe. We can go out, collect one set of data that's rideshare-specific at very high fidelity with advanced sensors, and we can provide that data to all of our partners so that we bring the benefit of scale to AV development. As of today, a number of key milestones have been passed on the AV journey, Khosrowshahi argues, with more to come: > We're live in seven cities, and we're on track to be live in 15 cities by year-end. We've got a Nuro/Lucid launch coming in the Bay Area. We've got Zoox coming in Vegas. We've got Wayve in London and Tokyo, Baidu also in London and then Pony and Verne in Zagreb and potentially more. More partners will enter the ecosystem, he predicts, citing Rivian as one possibility: > This is a full stack kind of a build, which is software and hardware with a very affordable bill of materials. We expect to be in perhaps San Francisco and Miami in 2028 for Rivian. And then with NVIDIA, we expect to be in L.A. and San Francisco in 2027 and then 28 different cities globally by '28 as well. Not that all partnerships have gone as smoothly as they might it seems. Uber has recently confirmed that it is endings a three-year robo-taxi partnership with Alphabet’s Waymo operation in Phoenix, Arizona. That said, similar collaborations between the two in Austin and Atlanta remain intact for now. Waymo remains a “very, very important partner”, insists Khosrowshahi: > We continue to operate in Austin and Atlanta. We believe we'll continue to operate next year in those marketplaces. It's a terrific product and the on-the-ground partnership continues to be very strong. At the same time, we want to make sure that we're not dependent on one partner. We're absolutely seeing a plethora of newer players in the AV ecosystem, just like you see in the foundation model space. And while we continue to provide a great service with Waymo in Austin, Atlanta, we'll continue to build our services with our other players as well. Early days So, grand ambitions ahead, but running through this is firstly quality of the service and secondly, the utilization of these vehicles, contends Khosrowshahi: > What we've seen is that launching with us as a partner, with the built-in demand that we've got, we can drive very significant utilization-per-vehicle, often mid- to high 20s, low 30s, in terms of trips-per-vehicle-per-day, which is quite substantial in terms of the needed monetization as it relates to these vehicles. So it's getting partners in market, quality of service and then obviously, the economics of the service that we're looking at. It’s all still very early days for Uber in the robo-taxi game, admits Khosrowshahi: > AVs are doing hundreds of thousands of trips per week. [As Uber] we’re at 300 million trips per week, so it's even less than 0.5% of our overall trip volume. You compare that again to the foundation model space, people estimate that 20% of search now has gone to AI, 40% of users are using AI search one way or the other. So the penetration of kind of physical AV is going to be slower. It's going to some extent be more deliberate. It's way below where it is at AI. The regulatory environment is a key consideration here, he posits: > We are a highly regulated business. We routinely talk to lawmakers, whether they're Governors, Mayors, council members who have real concerns about both the effect of AI and AVs with their constituencies - and some of these concerns are real. There are concerns about job loss, there are concerns about safety, there are concerns about congestion. > > > While AVs have been incredible in the markets in which we've introduced them, they've had their fair share of issues, whether it's safe driving through school zones or next to school buses, or how they interact with emergency response vehicles, or how they react to large power failures where the traffic lights aren't working. These are real issues that have to be discussed. For example, D.C. is talking about this stuff and we can't have AVs blocking the streets of D.C. when there's a Presidential or Vice Presidential motorcade going through. These are real issues. We've got to have the proper dialogue with all the constituencies to make sure that any new law reflects the needs of all these stakeholders. There is time to get this right, he suggests, with lessons to be learned from the wider AI sector: > We see sometimes the results of trying to go too fast and some of these AI companies with data centers, they were pushing through, you could argue, too quickly and there's been a huge public blowback against it. What we think is you need to have the dialogue. You need to have smart regulation and dialogue with our shareholders so you can actually enable innovation going forward, and we can drive AV regulation in a way that's sustainable, that doesn't kind of have the same blowback that you're seeing in AI. He concludes: > Sometimes you’ve got to slow down to drive sustainable regulation. We're very much pro AV, but we want any regulation to kind of address the needs of stakeholders and any model that we have is a model that truly lasts. My take Over 2,000 words on AVs and not a mention of Elon Musk! Damn, spoiled it now!!! This is a fascinating sub-culture of wider AI society and while I remain an inherently nervous passenger - regardless of who's behind the steering wheel - it's clearly a major opportunity for someone to crack one day. Before then there are, as Khosrowshahi notes, lots of regulatory hurdles to jump and they need to be jumped cleanly and as uniformly as possible. There's already an alarming degree of Balkanization of rules and regs around AVs, both internationally and within countries themselves. That's perhaps inevitable at this nascent stage of market development, but it's something that needs to tackled head-on to succeed. I'm off to San Francisco in a few weeks. Will I be tempted to try a Waymo this year? Well...not if there are other options, I admit. Maybe next year?