Pump the breaks: Why Autonomous Vehicles Aren’t Ready
Introduction: AVs following the trend of EVs
We talk about AVs — or autonomous vehicles — as though they’re already a part of daily life. As you read this, they’re replacing buses; they’re freeing up DMV lines and parking lots while cutting emissions, costs, and accidents; they’re further advancing the notion that AI is killing jobs (and/or creating new ones).
…not quite. Remember when the first highway-legal electric vehicles (or EVs) came out?
Tesla’s all electric Roadster was the first of its kind just a decade ago. In 2012, there were only 109,000 all electric vehicles on the road globally. Today, there are just about 3 million EVs on the road, making up about .3% (yes, three tenths of a single percent) of all motor vehicles.
I understand this isn’t a completely fair parallel… at all. Replacing piston engines with batteries to save the environment isn’t the same as replacing drivers with autopilots to increase profits and human mortality — but AVs are the next evolution in personal transportation, like it or not.
As with electric vehicles, adoption and trust takes time, and even as it spikes (the number of all electric vehicles grew 10x over the last five years), it still makes up a tiny fraction of all vehicles on the road.
Sure, looking into the crystal ball, AVs are poised to revolutionize the auto industry and society as a whole. According to McKinsey:
Global revenues associated with autonomous vehicles (AVs) in urban areas could reach $1.6 trillion a year in 2030 — more than twice the combined 2017 revenues of Ford, General Motors, Toyota, and Volkswagen… In the United States alone, if autonomous vehicles were fully adopted, the benefit to the public [including things like traffic congestion, real estate pricing, and costs associated with traffic safety] would exceed $800 billion a year in 2030.
But since we’re extremely far away from mass adoption, what can we expect this year and beyond? Before we get there, some background as to why we’re not producing and using AVs en masse.
Narrow AI and Edge Cases
Quick Debrief: What is Narrow AI? For those that didn’t read my piece on AI in 2019, a quick refresher on the types of AI: you’ve got narrow, general, and super.
General AI are things (hardware, software, or likely something using both) that are just as intelligent as humans, and are able to learn like we do. Super AI are — you guessed it — things that have surpassed human intelligence. We’ll revisit this maybe never.
AVs (of today) are narrow AI, or AI that are just as good, if not better, than humans at one task. Google is better at searching the web than a human. Your Fitbit can track your heart rate better than you can. You get the idea.
Edge Cases: Humans are still better “complete” drivers
For now, I (a human) am a better “complete” driver than a Tesla rigged with enough infrared-emitting lidars to be classified as a military grade weapon (the Pentagon, by the way, have their own colossal R&D budgets for autonomous battle vehicles and weaponry).
I say “complete” for a reason. Self-driving products from Tesla, Toyota, Google, Uber, and a slew of startups and OEMs are — if our primary metrics are consistency, safety, and lawfulness — better drivers than me.
As we’ve all heard before, human error is the leading cause of accidents. There were an estimated 40,000 traffic fatalities on US roads in 2017, according to the National Safety Council, and more than 90% of them were caused by human error. So even as deaths decrease, the reason for fatality remains the same: humans.
Still, humans are better “complete” drivers in that we’re equipped to handle “edge cases” AVs can’t. This from The Atlantic:
“Even with an appropriate set of guiding principles, there are going to be a lot of perceptual challenges that are way beyond those that current developers have solved with deep learning networks… we will end up wanting our cars to be as intelligent as a human, in order to handle all the edge cases appropriately.”
Would an AV know when to honk? Or how to respond to being honked at? What if one passenger was elderly and required more time to board the vehicle? Would the driverless car behind her know not to honk? What if it gets stuck behind a mail truck? Or a garbage truck? Or a motorcade? Or a marathon?
What if a road is blocked but the AV doesn’t recognize the signage? What if it can’t see the sign because it’s foggy? Snowing? Sleeting? What if the car needs to break the law in order to prevent an accident? What if there’s an unavoidable wreck and it has to triage passengers and pedestrians? What if your Tesla auto-pilot keeps slamming into parked vehicles over and over and over again?
Peppered into all of this is culture. Don’t sleep on it. If you’re in LA, Newark, New Delhi, Berlin, or Beijing, the roads, driving habits, and norms are all different. And it evolves. Will a vehicle’s software adapt as quickly as humans to changes in cultural driving norms?
There are too many “what ifs” to risk releasing fully autonomous vehicles in 2019. Humans, for now, are better at making the right decisions in such edge cases. And while companies are already building closed courses to train their AVs on all kinds of situations, including those above, there will still be accidents that a human would have avoided. The technology will lead to a netreduction in accidents, but part of the adoption barrier stems from the lack of control.
Solving a Solved Problem
On top of this, the problem that AVs are attempting to solve (traffic congestion, travel time, emissions, expensive modes of transport, etc) is, in a sense, already solved. It costs Uber and Lyft less to pay someone to drive you around than it costs them to outfit a car with a crap ton of tech. Again, from The Atlantic:
“Consumer vehicles with all those lasers and computers on board would be prohibitively expensive. On top of that, the question of calibrating and maintaining all that equipment would be entrusted to people like me, who don’t wash their car for months at a time.”
Going back to EVs momentarily, the recent spike in electric car adoption is due in part to cheaper batteries. Once a nascent technology (like all of the onboard hardware and software AVs require) is developed as industry standard (much like lithium-ion batteries) you will likely see heavy investment to bring the cost down and find safe and cheap alternatives.
So, long story slightly shorter: tech will get less expensive as paying humans becomes more expensive. And AVs are set to save cities and corporations up to $400B a year. But for 2019–2020, the human problem is (mostly) solved by humans. Not automation.
And then there’s 5G
You can’t talk about driverless cars without talking about 5G.
As I wrote in my piece on 5G, for things like self-driving cars and delivery drones to exist — things that require insanely fast response times from databases, information, and algorithms that are preventing them from, you know, killing people — we need the lowest of latencies.
You can read more about latency on your own, but just know that without 5G infrastructure (aka small cells), there’s no 5G. Without 5G, there’s no driverless cars (this is debated, but I stand by it). So, it’s a very real dependency, and one that will determine where and when driverless cars are rolled out.
Floor It: What to Expect This Year and into 2020
The AV industry saw its first death (ICYMI) before its first major hack. Morbid, yes. And death is not really comparable to a data breach. Still, it underscores the same thing: there are vulnerabilities and holes in these systems, and the companies that produce them make mistakes. A hack is bound to happen soon.
After all, hacks and data breaches rise consistently year over year agnostic of what industry you’re in. So why not 2019? We need protections from careless engineers and cybercriminals before hacking vehicles becomes commonplace.
Widely adopted semi-autonomous vehicles.
Back in 1926, commercial aviation faced a real issue: the public’s perception of air travel looked a lot like this:
This is “barnstorming,” or a popular roaring 20s form of entertainment where pilots performed stunts and showed off their skills, likely while Vaudeville music played somewhere in the background. How fun!
Here’s a not-so-fun fact: barnstormers caused 66% of all fatal flying accidents in 1924.
The commercial aviation industry went to Calvin Coolidge and requested that the federal government step in and create standards of safety in an effort to increase trust. Thus, the Air Commerce Act became law in 1926. For the next 93 years, we’d see advances in technology, pilot training, and accessibility for Americans of all incomes slowly transform air travel from a death-wish to a trusted form of travel (more on AV regulation coming up).
Today, 2 million Americans travel domestically via plane every single day.
AVs will follow suit. By deploying semi-autonomous vehicles, humans will get the trial period we so badly need. We’ll ride slow buses — the mass transit vehicles that drive themselves at comically low speeds to and from pre-defined destinations.
We’ll hop in manned autonomous vehicles. In fact, if you’ve gotten into a Tesla Model S or X you’ve done this already! Toward the end of 2019 and into 2020, Nissan, Mercedes, and Toyota will likely do the same, introducing cars that can drive themselves, but still A) require a human driver and B) puts the legal onus on the human, should anything harmful happen.
Lastly, you’ll get exposure as a pedestrian. Remember the first time you saw an electric scooter? You probably rolled your eyes, scoffed, laughed, furrowed in confusion, or did all four. I certainly did. But now I — like many who live in scooter-ridden cities — ride them too.
AVs are already roaming around cities like Las Vegas, Detroit, and Austin. The mere exposure to seeing a driverless car pass by will exponentially increase comfort around AVs.
Widely deployed AVs (for testing).
Fun fact: Waymo — Google’s self-driving car arm — clocked in 8 million miles on public roads in 2018. That’s double 2017’s mileage. So, we should expect at least 16 million miles in 2019.
For reference, that’s one autonomous mile for every 163,750 non-autonomous miles. At that rate, by 2030, we’ll see one autonomous mile for every 80 non-autonomous miles.
Keep in mind, it’ll likely happen in select cities, but still. Self-driving taxi services “don’t just live in our imagination” thanks to a partnership between Waymo and Jaguar.
All that training (millions of miles on public streets and billions of miles in simulations) led to this — a closed, discrete, fully autonomous service for a small subset of people in a single location.
This is another effort to get people like you and me comfortable with AVs. If your cousin in Phoenix endorses it, maybe you’ll start to trust it more, especially when you start to see testing happening in your town.
Waymo abstracts data to create comfort for passengers.
Google isn’t just testing AVs on the road. They’re also testinguser interfaces, conversational assistants, and even ambient music to make AVs more approachable.
So expect AVs to creep into your life, slowly but surely. These companies need data — petabytes and petabytes of data — before they can confidently take these from “testing” to “Waymo One.” Get ready.
Stoplight: The Case for Regulation
Airplanes as precedent:
Have you ever boarded an airplane and thought to yourself, “how is this allowed? Seriously. When did it become okay to stroll into a pressurized cabin shaped like a bird that traps you 7 miles in the air with 50 strangers? Is no one going to say anything?! WHY AREN’T I SAYING ANYTHING?”
Maybe it’s just me.
People put themselves in the hands of pilots (and auto-pilots), risk personal safety, and sacrifice the ability to protect themselves in the name of efficiency and comfort. One reason we’re willing to be so vulnerable is because we’ve designed a set of rules and standards that airliners must follow to ensure passenger safety. This is also known as commercial aviation regulation (which — in light of some very real errors from companies like Boeing and American aviation regulators — needs its own updating… I digress).
AVs will provide the same cost-benefit — we’ll sacrifice agency and personal safety to auto-pilots in the name of efficiency, comfort, and (likely) cost. Thus, before we’re cool with being in a vehicle that may harm us (and others), AVs need regulation. A lot of regulation. Aviation-level regulation.
Federal legislation is coming (once Trump is gone and trust is up).
71% of Americans still don’t trust driverless cars. Obviously, regulation isn’t a cure-all. We’ll need intelligent design (current cars are designed to make drivers feel in control) and lots of time (companies like Toyota and Google have enough capital to wait for consumers to gradually find trust in an AV’s).
If there’s time to kill, will we realistically see anything before 2020?
In October of 2018, the U.S. Department of Transportation announced a plan to rewrite safety rules that would allow for major changes in the way cars are manufactured, including updating antiquated rules that assumed a human driver. The rules, for example, may not require steering wheels. A bill regulating AVs stalled in Congress last year, and it’s set to be reintroduced in 2019.
That said, because the Trump administration takes a very hands-off approach when it comes to personal safety and transportation regulation (who’s shocked?), policies at the federal level will be proposed, but likely won’t become law until we see a new president.
State and local policies are already here (and laying groundwork)
DMV regulations in Pennsylvania and Arizona have allowed AV manufacturers to test on public roads since 2017. In April 2019, California followed suit, expanding specialized access from Waymo to any AV manufacturer, as long as the vehicle is under 10,000 pounds (vans and smaller).
This makes sense. Again, cities could save some billions and billions by adopting AVs. The sooner they get citizens to trust AVs, the sooner governments can reap the benefits.
We should start to see some AV deployment incentives (a la EV tax credits) at the state and local level as a catalyst for trust. If these governments save money and constituent lives with the deployment of AVs, we’ll start to see a real spike in cities adopting AVs.
Money’s good. Safety’s great. But what about job loss?
There are over 10,000,000 people in the transportation industry in America, 2,000,000 of which are sitting behind the wheel of a truck. For reference, coal miners aren’t listed on the US Department of Labor’s summary of extraction jobs page (that’s how few of them there are).
Oh, and two out of the three states that are allowing AVs to test on public roads have the highest employment of truck drivers:
According to the Bureau of Labor Statistics.
We are staring straight in the face of a job displacement epidemic; that is, if we don’t see some meaningful steps taken. This is an opportunity for local, state, and federal policymakers to mitigate risks to safety and the economy, yes, but also unemployment.
2020 here we come.
Conclusion: A world without drunk driving, day care, or driver’s ed.
The impacts of AV mass adoption (which, to reiterate, is years and years away) are radical, and go well beyond the obvious technical and economic effects. It’s easy to focus on lasers and lowered traffic accidents; job loss and federal regulations, but let’s consider society for a moment, and how AVs will impact every member and every generation.
Imagine a world where you no longer have to drive your elderly relatives around. Your kids can take themselves to school or daycare or a sleepover. Drunk driving becomes a relic of the past, and teenagers (who will surely continue drinking) no longer understand the concept of “drunk driving.”
This is the same world with excessive urban sprawl, new forms of guerilla terrorism, and a number of unseen risks to our way of life.
The changes to how we behave, both on and off the road, will be profound. It’s never too early to consider the impacts of such a revolutionary technology.
S/o to my dear friends Alex Barbato (ISL, Software Engineer) and Robert Golden (McKinsey & Co, Business Analyst) for sharing their wisdom.