Thanks, bot ... no thanks — why I’d think twice before getting into a driverless car

Be honest. If you needed to hail a ride, would you opt for a regular human-driven vehicle, or a driverless AI-piloted robotaxi, assuming the price was the same?

Me too. But it’s not the wisest choice. According to the stats from Waymo, one of the most well-known autonomous driving technology companies in the world, which operates on public roads in a number of cities in the US, including San Francisco, Austin, Phoenix and LA, after 15 million miles in every type of driving condition imaginable (day and night driving, snow, fog, rain, extreme temperature conditions) the autonomous vehicle was involved in 3.5 times fewer accidents than human-driven vehicles.

A further study conducted with insurance giant Swiss Re is even more startling, and we can assume that the stern eye of a huge insurer keeps the numbers honest. Huge reductions in at-fault collisions, property damage and bodily injury were recorded. 

We should swallow our preconceptions and hail a robotaxi, should the opportunity ever arise. At least, if safety is our primary concern.

There are two bigger stories here though. The first is that the problem of developing reliable and safe autonomous vehicles (AVs) is really, really hard. The world is an entirely unpredictable and constantly noisy place, and the stakes are as high as they come (get it wrong, and people can die). 

There is little room for error. AV designers cannot possibly imagine, let alone list, all the scenarios that could arise at any given moment on the road. And, frustratingly, while humans learn to drive easily, we cannot explain, in fine algorithmic detail, how we do it. We use our eyes, ears, hands, feet, common sense, memory, logic, intuition and our long history of being alive and moving through an ever-changing environment. 

All of which means, if we can’t tell the AV exactly how to drive, then it has to learn for itself. Ergo, we have a big, hairy, audacious, many-headed machine learning exercise. This is not your “predict-the-next-word” ChatGPT4. It is many orders of magnitude more impressive than that.

The development of AVs is AI writ large. 

There is a fascinating discussion about this in a podcast from venture capital company 16z, called “Building the World’s Most Trusted Driver”, in which 16z partner David George interviews Waymo CTO Dmitri Dolgov. 

Dolgov gets deep into the weeds of the tech, and what has to happen for it to all come together. Stuff like GPS, maps, radar, laser tech (LisDAR), cameras, light, AI models, accelerometers, gyroscopes, and so forth, all talking to each other in crisscrossing feedback loops, in real time, making thousands of micro decisions which may require millisecond-scale response times to an incalculable number of situations, both tested and untested. (Just one of the examples discussed – what does the car do if it has to stop for a passenger pickup and there is no parking spot available? Does it double park, and how does it judge when it is inconsiderate or dangerous to do so? Does it park in an open driveway? What if the gate or garage door starts to open?)

While listening to the podcast, I wondered about the people who chose to tackle this seemingly impossible challenge, which started in earnest in 2007 with a series of autonomous vehicle competitions designed by Darpa (Defence Advanced Research Projects Agency). I don’t mind a little challenge, and occasionally I take on a big one, fortifying myself with the tonic of ego, inspiration and aspiration. But this! Clearly these people are made of sterner stuff than me, and (presumably) most of you. 

There are more than 250 companies involved in building bits and pieces of AVs – software, radios, cameras, radar and other radio equipment, map providers, training data providers, etc – but a smaller number are actually putting cars on public roads. They are (among others) Waymo, Tesla, Cruise (GM), Zoox (Amazon) in the US, a surprisingly large cohort in China (Baidu, Pony.ai, WeRide, AutoX, Deeproute.ai) and, unsurprisingly, all of the major German auto manufacturers (Mercedes, BMW and VW). 

I was taken aback by the number of companies entering this race. These AVs, with all their fancy hardware and software and decades of R&D, are likely to cost significantly more than our current cars, even when manufactured at scale. And many of us will still want to drive our own cars, or would prefer to hail a car with the same species in control of the steering wheel. Why is all this money being spent? Is there actually a future which makes sense for AV investors?

The answer is yes, probably, maybe, we’ll see. MaaS (Mobility-as-a-Service) has been growing for a long time – ride-hailing apps like Uber and Lyft, bike and scooter rentals, long-distance ride shares, personal delivery apps. All have worked swimmingly. You don’t have to own your mode of transport, and it is often cheaper and more convenient not to do so. MaaS seems to be on solid ground; there are plenty of companies increasing both their revenue and profits in this space. 

But, at the risk of being a gadfly, I think self-driving cars are a different kettle of fish. Yeah, subways and other track-borne transports sometimes don’t need drivers, but there are not that many decisions to be made. Start. Stop. Slow down, Speed up. Open and close doors. Even aeroplanes are barely piloted these days. Pilots are basically sitting on the bench except for unusual situations, take-offs and landings – and even these last are near-autonomous. But we absolutely want those pilots in their seats, don’t we?

Perhaps we will always prefer the Uber driver too. A person, just like us. To chat with, or not. To make us feel like we are not alone, we are not at the mercy of a software bug or hack or a completely unprecedented driving situation, never before seen or anticipated. 

As a technologist, I am enthralled by these cars. But I think I will still want to ask a driver to take the scenic route when I feel like it, and then perhaps witter on about the unseasonably hot weather. DM  

Steven Boykey Sidley is a professor of practice at JBS, University of Johannesburg. His new book, It’s Mine: How the Crypto Industry is Redefining Ownership, is published by Maverick451 in South Africa and Legend Times Group in the UK/EU, available now.