What If Autonomous Cars Just Never Happen?

More and more semi-autonomous cars cruise on the road every day, and yet the more time I spend testing these driver-assisted vehicles, the more I think that full autonomy may never, ever happen.

A harsh look on all of this was first laid out all the way back in the 1960s, at the start of the first revolution in regulating vehicles and making them safer. Back in the decade that launched consumer product safety with books like Silent Spring and Unsafe At Any Speed, American regulators first started talking about cars not as a technology or a means of transportation, but as a public health crisis. Cars are, from the perspective of safety, a disease. They're an outbreak run wild, killing tens of thousands of Americans every year - and the leading cause of death for teens, as noted by the Center for Disease Control.

For us to combat this disease, we need to reduce the number of fatal car accidents on the road, and there aren't a ton of possible ways of making that happen. We're certainly not better and more attentive drivers than we were in the past. It sure looks like the way to reduce deaths is to make cars safer, and the most death-proof car ought to be one that drives itself and never crashes.

Yet there are other ways to bring down traffic deaths than robot cars. There are ways that we have access to right now, not dangling perpetually ahead of us - things like making our roads safer, designing cities better, building safer conventional cars and creating smarter crash tests. (Even better driver education would go a long way, in theory, though in practice that's tough to implement too.)

This is leaving aside that most people today may not even want a fully autonomous car, were one available, as studies have shown.

Now, we've seen huge strides in getting cars closer and closer to this mark over even the past few years. A new Tesla, Volvo, Mercedes or Cadillac off the showroom floor will be able to, with varying degrees of competence, cruise down the highway without you steering the car. The vehicle will stay in its lane, hold whatever speed you desire and brake if someone stops short ahead of you. It feels very much like the car is driving itself.

But if 2017 has proven anything in the world of cars, it's that whooshing along without incident in one lane falls far short of what you can comprehensively call "driving."

Just the other week, a "fully autonomous" bus took to the road in Las Vegas and almost instantly got involved in a crash. Somebody backed up into the little autono-bus and the bus' computer brain couldn't sufficiently think to back up out of the way. The crash wasn't the driverless bus' fault! It was that of a human in a human-driven car. But it couldn't do all of the minute little jobs that a competent human should be able to manage.

As another example, Tesla's Autopilot system is smart enough to execute lane changes on its own. You signal like you're going to change lanes and the Tesla will slide one lane over. That is, it will if the Tesla doesn't see any car in its way, even one hiding in your blind spot. But the Tesla has its own blind spots.

It wasn't long ago that somebody died behind the wheel of a Tesla on Autopilot, crashing into a truck that the Tesla couldn't see. The driver was trusting on his car to keep him safe. It had saved his life before in a different scenario.

But in this over-reliance on a driver-assisted car, we saw how little we expect unexpected edge cases. We'd all be safer and happier if all cars even had current semi-autonomous tech, sure, letting cars take over in our most bored, texting, straight highway traffic moments. But the barriers to full autonomy may be altogether too high.

Autonomous cars and human-driven cars may never be able to mix. What then? Do we make dedicated lanes for self-driving cars? How distinct do they need to be? If we need them to be physically blocked off from human traffic, do we really have the money to build that kind of autonomous roadway network? We'd basically be building a new train system at that point. And how do we phase out the last of the human-driven cars? Can we ever do that?

By then I'm starting to wonder how much putting that money into repairing our current road system would be for reducing driver deaths. Fix all the bad intersections, the undivided two-lanes, and so on. A full rollout of driverless cars for all car owners is almost impossibly far away. There are plenty of cars on the road today that still don't even have airbags or ABS, let alone a full suite of cameras and sensors.

So if you can't mix driverless cars with human-driven cars, do you really ban human driving? Car companies have talked about it, sure, but it's wildly unprecedented and it puts an incredibly strong amount of pressure on those without enough money to replace their cars for something safer, as Patrick George pointed out.

There is a lot of momentum leading us towards autonomous cars, and it feels like we're on some kind of cusp with them, but I hate to say that other technologies have sat in the waiting room of reality for longer. Virtual reality has been about to happen since the early '90s. Yet it stuttered under the weight of its glowing promotions, never able to promise as much as everyone hoped it would. VR sets cost too much and offered too little. The rigs were expensive, and the content was lacking. Many applications of VR today are downright embarrassing.

Now the tech limps along with many companies afraid to invest much into it, as Kill Screen recalled in first one then another feature on the 'broken promise' of the tech.

And self-driving cars have been in this cusp for longer than you think. Delphi did an autonomous cross-country drive in 2015, but so did a pair of academics in 1995. Full autonomy is the future today, just as it was 22 years ago. At a certain point you have to wonder if this is never going to happen, like flying cars, perpetually promised to be a few years away.

Since 2013, car companies have argued that fully autonomous cars would be in showrooms by the end of the decade, but in the past few years (when we've seen autonomous car testing in full swing) those promises have looked shakier and shakier. Ford, for instance, told Forbes in 2015 to expect full autonomy by 2020, but it couched it by saying these cars would probably be restricted to defined areas. A year later in 2016, Ford announced that it would have fully-autonomous cars by 2021, but it would just for geo-fenced ride-sharing or ride-hailing.

It's time to seriously think about a future where complete autonomy never comes to us as we imagine it, where you don't have roads solely occupied by self-driving cars talking to each other, whizzing through intersections at 80km an hour.

The best case scenario I can imagine has large cities banning cars from their inner downtowns with limited-application self-driving cabs whisking people around instead. I feel like we're only a few more terrorist attacks away from urban car bans as it is. The worst case scenario I can imagine has that flopping, too, if enough people hack into these cars and hijack them, too, as I wrote back in 2014. Beyond that, you can have cars with glorified cruise control as we do now, only across the board, not that that alleviates much of anybody's more grim hacking fears.

I'm not saying that promoting driverless cars and promoting safer roads today are mutually exclusive, but there is a lot that we can do at the moment that we should be taking more seriously. We're probably not going to become better drivers through schooling. (We've been trying to make drivers pay attention to the road ever since people started driving.) But we can force ourselves into become safer. We can mandate more and better crash tests. We can stop regulating every individual aspect of a car for safety and go back to a holistic approach. That's what's holding back, for instance, car headlights from being much better than terrible?

How much do we keep waiting on this dream and how much do we keep supporting making our infrastructure better, and our drivers safer?



    What If Autonomous Cars Just Never Happen?Yeah, driverless vehicles will happen and sooner than we might expect. Especially for heavy transport, but definitely for the daily driver.

      I'd say some of the things that's holding them up are the ethical questions eg. if something happens that the car can't avoid who does the car kill: the driver or the hapless pedestrian?

        I feel like this should be a fairly easy one to answer. The pedestrian doesn’t have any protection, they are also completely set apart from the road system. If the car has to choose between protecting the driver or the pedestrian, it should choose to protect the pedestrian.

        The driver at least has a chance of survival with a hard shell and other safety features. More of a chance than a pedestrian being hit by a couple of tons of steel

          There was a survey not long ago asking whether autonomous cars should save the people in the car or save the pedestrian. Most people answered the pedestrian. Those same people were then asked Would they buy a autonomous car that saved the pedestrian or a car that would save the passengers, and the answer was a car that would save the passengers.

            Yeah, that certainly sounds like humans.

    Driverless transportation has the capacity to be faster, safer and more convenient. It's too beneficial to fall by the wayside. That's why I think it'll be the future.

    However, I think autonomous vehicle incidents need to be treated in a similar fashion to air crashes. Thorough crash investigations, regular updates to protocols and standards, and severe penalties for non-compliance would be critical.

      I expect that to happen regardless. If automated cars make things worse, it puts the car companies out of business almost overnight, just from legal action. They wont want that.

      End of the day, most accidents are caused by a common element - humans. They either make a mistake, or react badly, and someone is injured or killed.

      You may not be able to eliminate every one, but if you take out the human element at least you're reducing the mistakes causing the accidents. A car pulling out, or speeding, running a red light, or drink/drug driving can all be avoided if the human isn't making the decision to risk their luck.

      That alone would drop the death rate if the autonomous vehicles can get the basics right. But they need the real world data to get that right first, and that needs to start somewhere.

      A controlled environment is great, but as the Vegas situation showed, the real world throws those X factors out there that frankly, they need to happen so they can work out how to avoid it.

    I think it will happen, but probably not in my lifetime. I think we will see owning a car becoming less of a thing and a fleet of autonomous taxis constantly buzzing around.

      AI is growing in intelligence, not just fast, not even at Moore's law speed, but exponentially. I would not be surprised to see AI being very close to human intelligence within ten years.

        I'm a "little" skeptical of that, but hopeful. Simply because the physics of 19th century electronics, isn't going to increase in power exponentially, we have hit the wall with semiconductors and the humble electron. The AI is coming, I don't disagree, but we need a new revolution from the underlying electronics, like the day we invented the transistor. This is where quantum computing will change the face of the computer. I would say that once we start creating our first quantum (photon) based cpu's and memory...it will be a rocket ride! There are a lot of people working on this right now...and for good reason. We just had a great break though using Topological insulators here in aus (https://www.gizmodo.com.au/2017/11/australian-scientists-are-really-killing-it-with-this-quantum-computing-thing/).

          Theres a lot more than just that quantum computing story happening as well. There was a story a while back about how researchers were able to control things at an atomic level, and control which direction electrons were spinning.

          I remember it because a good part of it was done here at Wollongong Uni. No idea where its at these days.

          But it creates binary data storage at an atomic level, or lets you define pathways an atom wide, or any number of other possibilities. Nano tech becomes probable, and quantum computing a reality.

          Not saying it will happen using that, but it means there are already avenues that can make it happen. I think we're closer than most people realise.

          The transistor was the greatest invention of the 20th century, but like vacuum tubes before it, just remains too big to remain cutting edge in the future. The alternatives are out there though.

            "control which direction electrons were spinning"
            Yes... very true, and that actually has a bit to do with quantum machines also, as the way they were controlling this was via topological insulators, basically they could infer the spin direction onto light, to help view electron properties on the surface of the insulators, its called spintronics (all tech must have cool names ;). By controlling the spin direction of electrons, we gain massive current efficiencies (less heat etc). There's a good reason why the guys that done a lot of great theoretical work on this material (topological insulators) won a noble prize for it in 2016.
            But its the quantum world we need to get to Grunt. We need to get control of the qubit, and then we can exploit quantum parallelism. My hero Dick Feynman done the first abstract models of quantum computing, the guy was not form our planet ;) Quantum bits, or qubits actually have three states, 0, 1, and yes...both at the same time (gotta love that quantum freakyness), this is called superposition (look up Schrödinger's cat). With new algorithms designed to take advantage of quantum parallelism, we can do some of the hardest known tasks for a computer, lets say prime number factorisation, in seconds. This is in comparison to a semiconductor based machine, which would take billions of years to do the same thing...no shit, the math has been proved, now we just need to build the machine...

              Sounds familiar. A sibling lectures in Wollongong Uni's School of Computing and IT, and while their research doesn't get to that level of things, I think it does deal with some 0, 1, Yes state things. Not my area of expertise though, so don't quote me on that. They explained it to me once while we both happened to be in Vegas, but even as a math guy it was mostly over my head.

              But them working there does mean I get the occasional tidbit of info on this sort of thing, and it always sounds interesting.

    "the minute little jobs that a competent human should be able to manage." but don't.
    Every driver responsible for a crash from lack of attention or alcohol/drug abuse should be required to move up to an autonomous car or lose their licence.
    Those who don't want a car, will still benefit from autonomous taxis.
    To modify every, less than perfect, bit of roadway would be far more expensive than autonomous cars.

    The most Luddite article I've ever read on Gizmodo. Anyone who is following the advances in GPU technology applied to robotic vision can see the writing on the wall. It's not an "If" it's a "when".

    "We tend to overestimate the effect of a technology in the short run and underestimate the effect in the long run," - Roy Amara.

    How about we just ditch the car altogether. Even with both electric tech and self driving tech, the majority of problems created by the car still exist. For example public health problems from road dust and sitting still too long, environmental costs of land use, manufacturing and disposal, the social costs of urban sprawl and the enconomic costs of traffic. Plus the crashes that will still happen

    One of the problems with autonomous as i see it , is that it does not matter how many sensors you can chuck on a car , or a truck ......... it is essentially still a dumb machine that does not think like a human .
    You just cannot program in experiences that humans learn via trial and error .
    and i would not trust a machine that has been programed for a skill by blokes in lab coats .
    The road is a dangerous place with conditions that change constantly , also humans and animals are unpredictable , if you have been driving for some time you learn to read other drivers and road conditions .
    certainly computers and the autonoumas gear have a part to play , but fully autonomous like a cat among the pigeons sounds very scary to me .

      Sorry mate but you can program experience and with regular updates every car will learn from every other autonomous car(within their respective brands). So basically their experience will continually evolve and at a rate far greater then our own. No doubt you would of seen a lot of people behind the wheel that should not be there.
      But the big question is if a accident does happen and it's the cars fault whose going to take responsibility.

      Trying looking up AI, the different types of AI and how they work. Because you are exactly wrong about it. It is not a dumb machine, and in fact, good AI is made to learn from experiences and learn from trial and error, this is one of the main if not THE main powerful features of AI.
      When the technology has reached a certain point, the opposite of your worry will become apparent; the idea that dumb humans with all their flaws in judgement and poor understanding of risk were ever allowed the autonomy to drive these huge chunks of metal at 10s-100s of KM an hour around each other at their own will might seem insane.

        Exactly, in the future history teacher's will be telling children the chilling tale about how we drove our own cars and 100s of thousands of people a year died in accidents, it will be viewed like we view the wild west today.

    It would be possible for cars to talk to each other even if driven by humans.
    They could say where they are in relation to each other and avoid each other.
    I already drive a car with adaptive cruise control and it’s fascinating to use it in stop start traffic. (I have to start it from a stop}.

Join the discussion!

Trending Stories Right Now