WATCH THE REPLAY BELOW
Here's a preview of the next issue of "About Everything." You're welcome to join a Hedgeye Q and A on this piece that I will do on Tuesday, July 12, at 2 PM ET.
The implications of this thesis are broadly negative for firms participating in the development of driverless cars. They're very slightly negative for the big tech giants: For Google (GOOG) and Apple (AAPL), driverless cars are basically a fun hobby. The impact is a bit more negative for the big auto firms, which need something new now that low interest rates and low energy prices have basically sucked consumption of their current product out of several future years. The five auto firms with the biggest stakes are Tesla (TSLA), Audi (FWB:NSU), Mercedes (FWB:DAI), BMW (FWB:BMW), and Honda (HMC). Not coincidentally, most of the action here is in foreign luxury cars, trying their hardest to push semi-autonomous to all-the-way autonomous. Very recently, Ford (F) and General Motors (GM) have also got on board: Even if just for PR purposes, they cannot afford not to be left behind.
A purer play on driverless cars would focus on sensor, software, and AI firms. Foremost here, in the public space, is Mobileye (MBLY), which totally rides the driverless wave: It careens up and down on rumors and now trades at a P/E of 139. Mobileye's chips are mostly supplied by Semiconductor Manufacturer International Corp. (SMI). Some other listed firms include Nokia (NOK), which (post cell phones) specializes in onboard mapping; NXP Semiconductors, NV (NXPI); Flextronics International, Ldt. (FLEX); Nvidia (NVDA); Autoliv (ALV); and of course Delphi Corporation (DLPH), which is happy to supply Big Auto with whatever Big Auto wants. Of all these, only MBLY, NVDA, and DLPH are based in the United States.
To recap, my take is negative for all of the above to the extent driverless cars feed their bottom line. My only partial qualification would be for Mercedes and Caterpillar (CAT: not mentioned above), whose efforts have disporportionately focused on driverless commercial trucks. Quite possibly, this is a product line which could be ready for industrial or long-haul commercial use (e.g., on U.S. Interstates) well before the driverless auto.
driverless cars: unsafe at any speed?
On May 7th, a man died after his Tesla crashed into a tractor-trailer while in autopilot mode. According to witnesses, the driver was watching a movie and never applied his brakes. While the crash wasn’t publicized until June 30, it has come out that Tesla CEO Elon Musk knew about the accident shortly after it happened.
Given the breathless excitement surrounding driverless cars, it’s no wonder Musk didn’t want investors learning about this tragic accident.
Autonomous vehicles, along with various autopilot add-ons for ordinary vehicles, have been under development for well over a decade. But in the past few months, the race to bring them into commercial production has accelerated. In March, General Motors (GM) bought self-driving startup Cruise Automation for $1 billion. In May, Toyota (TM) and Uber joined forces, Apple (AAPL) invested $1 billion into Chinese ride-sharing company Didi Chuxing, and Google (GOOG) partnered with Fiat Chrysler (FCAU).
The fate of self-driving cars impacts both the big automakers that have invested heavily in the space—from Ford (F) to Honda (HMC) to Tesla (TSLA)—as well as the tech companies working with these giants. Those with the most at stake, however, are the small firms specializing in driverless car technology. Nearly all are private, though Mobileye (MBLY) is a notable exception.
If you believe these companies, we’re getting close to the consumer-ready driverless car. Tesla says that it will have one commercially available by 2018, while BMW is planning for 2021.
At first blush, this timeline seems plausible. Semiautonomous add-ons like lane departure warnings and automatic parking are already in widespread use, after all.
My take on all this buzz? Check your long exposure. I don’t see a fully autonomous vehicle coming to the general marketplace for two decades at a minimum.
WHY IT’S HARD TO GO “FULL AUTO”
Yes, in the very long run, programmers and carmakers will certainly achieve their goal of producing an affordable driverless vehicle fit for widespread use. But it’s not going to happen on anything close to the timetable that these companies and the industry media are now touting.
Why? Some of the likely causes of delay are already well understood. These cars are still too expensive. (Today’s Google car, fully rigged with LiDAR, costs six figures to build.) They will require a whole new legal and insurance infrastructure. (After an accident with these cars, you sue the car company, not the other driver.) And they are unnervingly vulnerable to hacking (or mass hacking, which could make this a national security issue).
The most intractable obstacle, however, is the extreme difficulty of moving from semiautonomous to fully autonomous. Most of today’s semiautonomous features take over the most routine driving tasks. All higher-order thinking remains with the driver.
What will it take to replicate that higher-order thinking? Ah, there’s the rub.
Here are some critical difficulties that will challenge any version of driverless car likely to be available in the near future.
Staying on the road or lane in all circumstances. Driverless AI cannot cope with road edges or lane markings obscured by water, snow, gravel, or dirt. Same story with lane markings weathered by age or incorrectly veering off the road—as well as construction zones and detours where roads change.
Yes, driverless cars have GPS, but plus-or-minus 3.5 meters just isn’t good enough for the oak tree that’s only a couple feet from the curb. No car will dare abandon its LiDAR or camera for an accurate read of the immediate environment. Bad weather only makes things worse: In rain or snow, current autonomous cars are programmed to stop. Yes, that’s right, they just stop.
Making sense out of confusing and ambiguous situations. Think of parking lots, frontage roads, strip malls, and toll plazas where lines disappear and judgement takes over. Think of obscured or ambiguous signage. Or a police officer signaling traffic. Humans can figure out what to do when the rules are unclear, provisional, or absent. It’s hard for a driverless car to understand that it’s sometimes acceptable to drive in the opposite lane.
Interpreting disorienting changes in light. Today’s sensors can’t recognize the color of the traffic light through the sun’s glare, ignore a stop sign reflected in another car’s window, or determine whether the approaching dark spot on the road is a shadow or a shallow puddle or a deep pothole. Case in point: In the recent Tesla accident, the autopilot sensors couldn’t distinguish between the white side of the tractor-trailer and the brightly lit sky.
Engaging in higher-order object recognition. Humans know that it’s safe to drive through crumpled paper or plastic bags. Or slow down when hitting some low-hanging leaves. Or swerve entirely around tire debris and pieces of fender. How is a computer supposed know the difference and act accordingly? Humans are also able to figure out which signal in a complicated string of traffic lights belongs to their lane. GPS-plus-sensors are nowhere near smart enough to intuit this.
Comprehending human intentionality. Only a human can easily assess the intentions of other humans, whether it’s dealing with road-rage drivers or dodging cars taking evasive maneuvers from emergency vehicles. It takes a human to “read” the directions of a police officer, the gestures of a driver in distress, or the hesitation of a deer in the road. When a ball rolls into the road, only a human can sense when a child is likely to be following.
It's telling that many robotics experts are a lot more skeptical than the industry media. David Mindel is an engineer who has spent his career working with autonomous planes, submersibles, and space vehicles (and has written a book about it, Our Robots, Ourselves: Robotics and the Myths or Autonomy). Here's what he says about autonomous cars: "Full autonomy has… just proven to be a loser of an approach in a lot of other domains. I’m not arguing this from first principles. There are 40 years’ worth of examples.”
Many techno-enthusiasts will argue back: Wait, the driverless car doesn’t have to be perfect—it only has to be better than the human driver. Wrong. People have vastly less tolerance for a system or machine that arbitrarily kills people than they do for deaths that occur as the accidental result of personal human intentionality. That may defy cost-benefit analysis. But it is a fundamental truth about how real people make moral judgments. (See I, Robot: book by Isaac Asimov... or movie starring Will Smith.) The very fact that we would need to refashion the body of law governing car accidents (from personal injury tort law to product liability law) should give enthusiasts pause.
Analogy: While playing hockey, players often get hurt by flying pucks. Everyone accepts this is part of the game. Now imagine an otherwise “safe” puck which, from time to time, spontaneously jumps out and concusses people at random. Now we have a much more serious problem—public outrage and a wave of product liability suits that will quickly put that puck-making company out of business.
Watch out for a backlash against semiautonomous automakers. The race to fully autonomous vehicles is an all-or-nothing bet: Either people can safely take their attention away from the road or they can’t. Anything short of 100-percent autonomy is a big problem—which already poses an awkward tension in the marketing for current semiautonomous features. The company tells you that this cool feature enables you not to pay attention. But then they tell you always to pay attention. Imagine those very words in a closing argument to a jury.
Don’t believe any hype about “cars without steering wheels.” Google is determined to create a car that does not require a steering wheel or pedals. This concept is fine for specialized commercial purposes (like in mining and forestry)—but will never fly for everyday consumers. To accommodate such vehicles, the United States would need a total infrastructure remake. At the very least, the car would need to have an override in case of emergency.
Don’t expect onboard technology alone to create the 100-percent driverless car. Sure, better integration of satellite and LiDAR and better AI would be an improvement. But perfecting the driverless car requires much more. We will certainly need a new and improved highway system designed for a driverless society. We will probably also need a wireless communications infrastructure that allows all cars (even those with drivers) to signal to each other. Such a system would require massive public investment over at least a decade, the prospect of which isn’t even on the horizon yet.
The slower-than-expected rollout of driverless cars may be a good thing. While Boomers and Xers have spent a lifetime wedded to the concept of a car as a projection of their personal autonomy, Millennials are more comfortable ceding personal control to an intelligent system and are less attracted to ownership. A future devoid of drivers may only be possible when most of adults are younger than the median Millennial age, not older.
- Detroit and Silicon Valley make it sound like fully autonomous cars are imminent. Investment cash is flowing. Tesla and BMW promise a consumer-ready product within five years.
- This promise is mistaken. It will take at least two decades (possibly more) before driverless cars are ready for widespread adoption. Yes, the “low-hanging fruit” of semiautonomous driving has already been plucked. But full autonomy requires infallible higher-order thinking, a golden apple which will prove difficult if not impossible to grasp anytime soon.