Driverless cars, lawyers and the adjustable ethics switch – will we have to choose who to kill?
As car manufacturers start planning for driverless cars they want liability to reside with the owner – unlike Google which has taken a different tack. Welcome tot the world of adjustable ethics settings. That’s right, you get to choose who to kill. Patrick Linhas recently posted a piece in Wired on robot cars with adjustable ethics settings. He calls it a really bad idea. And so do I.
However, his main difficulty comes from a wrong starting point. He assumes that driverless cars will operate in the way all car companies are now designing them to operate.
So far as I know, all the traditional car companies are looking for ways to keep the driver legally in control… so the driver retains liability in the case of an accident.
They are also seeing cars remaining as a ‘personal toy’, with most people still owning the car they use… and being forced to take responsibility for setting its ‘kill’ parameters. To choose between killing yourself or the other party, or between an adult and a child.
This debate is far from settled, and is likely to never get past the ‘crazy idea’ stage. However, it does illustrate some of the major flaws in the design process being followed by the traditional car companies.
By trying to keep the driver in control, they are designing their cars to include fancy new screens and software to connect people with the car and with their environment and with all the advertisers that want to beam messages to them as they travel.
This causes great problems in designing an even more costly and complicated human machine interface (HMI).
They require systems to monitor multiple sets of inputs: one from the cars sensors that are monitoring the passive environment, one from the sensors monitoring messages from other cars, as well as from the sensors monitoring messages from fixed roadside infrastructure, and also from satellite GPS and other positional signals, as well as from all the operating components in the vehicle itself, engine, tyre slip, etc., and finally another whole set from the driver: steering, acceleration, braking, indicating.
And they have to include algorithms that choose between the different sets and then alert the driver when they are in control… all in time to respond to an impending accident… with over-riding personally set ‘kill’ parameters (if some crazy people get their way).
But it’s not going to happen, Google is going to win this particular race.
I wrote about about Google’s Dinky New Car earlier this year and the impacts it will have, not just on the car industry, but on traffic congestion, city planning and the use of rail, and most relevant to this discussion: safety.
The car companies are trying to fit computers into a traditional car.
Google thinks of it as a computer on wheels that is designed to work just like we do. The car is controlled from ‘inside’, with little need to communicate with external objects.
We use our mental models of the world and our senses to understand the passive environment around us (including moving objects), and respond accordingly. In the same way, Google uses detailed 3D models and sensors to identify objects in the external world, and respond accordingly.
As such, Google’s car has no need to communicate with other vehicles or infrastructure. It has no steering wheel, accelerator or brakes… so it has no need of a complicated HMI. Which is why it has been able to safely operate on existing roads with existing vehicles without a driver.
While not all situations are solved, the cars have now driven over 700,000 miles in all types of traffic and road conditions without accident (except two caused by humans).
What most people don’t realise is that these cars are programmed to learn from experience how to deal with real-life situations. Unlike humans however, every time one robot car learns, they all do. And the more of them on the road, the more they all learn.
There will be no need to choose who to kill… the cars will do their best to avoid an accident and they will do a much much better job than people. No one will program them to choose one scenario over another. The choice will be the result of a complex interplay between algorithms operating in an entirely unique set of circumstances that no-one can possibly foresee.
Nevertheless… the result overall will be very few actual collisions.
As the stats build, parents will very quickly choose driverless over driven for their children… and for themselves.
There will be accidents of course… just as there are in aviation, despite high levels of automation. And just as in aviation, we will put huge resources into understanding why… so it does not happen again the same way.
Despite our best efforts, planes still crash. But we would never go back to pilots being in sole control.
To emphasise the trend to automation, there is a joke around the industry that one day there will only be two crew on an aircraft: a pilot and a dog. The pilot is there to feed the dog, and the dog is there to bite the pilot if he tries to touch anything.
Today, there is no dog and pilots can still take back control in an emergency. At least in the air, there is usually a few minutes to try to recover. The reality is however, that in a major failure, this is often not achieved because the pilots no longer know what the plane is doing – so they don’t know what to do to correct the failure.
On the ground, decisions will be split second. Either the car will get it right… or you will have an accident.
But the chances of avoiding an accident will be greater, simply because the cars will be able to see through fog and in the dark much better than people, and they will always be on the alert, never distracted… and with some infrastructure even ‘see’ around corners.
Even more importantly, the cars will be programmed to maintain the centre of lanes and speed and spacing… dramatically reducing the chance of an accident.
And if there is an accident, very few people will be involved… because each car will have its own avoidance system. Even if one fails, no others will fail – so each will have the ability to take its own avoidance response. That is, there can be no ‘network failure’ which could occur in systems reliant on external messaging.
As for liability, yes, Google is taking the punt they will be liable… just like any product manufacturer, including aircraft manufacturers. But overall, their insurance will be much much cheaper than for any driven cars.
Imagine being able to order a car and have it arrive within two to five mins and take you safely in comfort wherever you want to go, without congestion: it will be ‘public transport’ that runs to your own schedule, point to point. Simply sit back and read a book, check your emails, write a paper, chat with a friend… or just stare out the window at all the other poor sods clinging to their wheels, focussed on the traffic around them!
And the vehicle will be directly suited to your immediate needs: a small town car for a trip into the city. A ute to cart rubbish on the weekend. A people mover for a day out with mates. An SUV for a holiday in the outback. And with ‘on-demand’ use, range ceases to to be an issue. For those few journeys that extend beyond your immediate city… just hire a vehicle with extended range capabilities. These will be more expensive, but you won’t be paying for their cost when most of the time those capabilities are hardly used.
And it will cost you a fraction of owning one type of car. Electric vehicles are much less complicated than internal combustion vehicles and hence cheaper to build once the whole supply chain is geared to volume. They are also easier to maintain and without an HMI or any fancy in-car equipment, they will be very very much cheaper still.
Anyone using Google’s car will already have their own smart device for all communications… no need to build it into the car.
The Google car is a really simple cheap device that just gets you from A to B quickly, comfortably and more safely than any human driver. And they will only get better and better in time. The current car is just the ‘Model-T’ of a whole new form of transport.
This week I spoke with some of the Federated Automotive Products Manufacturer’s in Australia and with key people from the AutoCRC. Many of these firms are facing annihilation with the imminent closure of Toyota, GM and Ford manufacturing facilities. This gives us a great opportunity to re-focus our industry on a 21st. century mode of transport: on-demand driverless vehicles.
This requires a totally different business model.
Under the current paradigm, profit is made from selling more cars with more features and more spare parts. It is incredibly wasteful.
Under the new paradigm, profit is made from selling ‘on-demand use’ of the ‘right vehicle’, at the ‘right place’, at the ‘right time’ – for ‘least cost’. It means having the least number of cars available with the least features to provide the service at least cost. It means assembling and hiring them out here in Australia, and more importantly keeping them clean, well maintained and refurbished (replacing interiors and seats as they become worn, and reconditioning motors, etc) and finally decommissioning them and re-cycling them. This business is essentially immune from import competition because it has to be delivered where it is consumed. (As a side note, it seems Renault is now making more profit from re-conditioned engines than they make on new ones).
Robotics and augmented reality (used to guide workers in real-time) will help reduce costs in processing a wide range of vehicles with relatively small volumes across a wide range of different processes (assembling, cleaning, maintaining, refurbishing, decommissioning and re-cycling).
By the time we get our act together to actually set up the capability, the vehicle designs and software will be available for licencing… but given the speed of change in technology and the length of time it takes to set up a whole new industry (with appropriate regulation), we need to get cracking – 2020 should be our target.
Fortunately, we are not starting from scratch. Intelligent Transport Systems capability in Australia is world leading on a number of fronts. And we still have highly capable engineers who are about to lose their jobs.
It will be interesting to see if we are up to the challenge.