For those who work in the auto sector or auto insurance, the jargon in this blog should be relatively easy to follow. To help others, I’m going to start with a brief overview on autonomous cars. After that, I’m going to explore how different data types will transform how we make decisions on the road, and how I believe companies can set themselves up for success.
Driverless cars are coming. As an individual object with limited other road users, driverless cars and their sophisticated software are able to navigate our roads with somewhat ease. They can follow traffic lights, turn left and right, and even navigate complex systems such as magic roundabouts – think Hemel Hempstead and Swindon.
The biggest challenges for the widespread adoption of driverless cars is getting them to integrate with the current system of human driver, human passengers, pedestrians, cyclists, foxes, and cats, and the public relations risks that any accident poses.
The view of the road from the perspective of a driverless car
To help push forward driverless car adoption and understanding, a system of autonomous levels can be used to describe how sophisticated a car is.
Level 0 has no sophistication and needs to be driven by a human who is in control at all times. Level 5 is called full autonomy, so all the human passenger needs to do is tell the car where to go. Between Level 0 and Level 5 we have a sliding scale where human interaction reduces as automation increases.
Driverless cars – the 5 levels of automation
This approach raises a number of valid questions that will delay the adoption of driverless cars. Level 4 still requires humans to be alert in order to take over from the autonomous vehicle. And research has shown that our reaction times, particularly if we are over 55, may be too slow if we need to take over in case of an emergency.
The human interface with machines is widely referred to as human-in-the-loop. Machines would work much faster and more effectively without us pesky humans in the way. But we are.
CAVs: Connected Autonomous Vehicles and data
There is still some debate on what we should be calling driverless cars, but the majority of the industry is using the CAV acronym, which stands for Connected and Autonomous Vehicles. While autonomy is an important long-term part of this technology, the connected car element is equally valuable.
As we move away from pure human driving, more data is required to help machines make decisions. Connecting cars with systems becomes a pivotal point. To many, this has been viewed as the collection of telematics data. Devices installed during manufacture and/or added by third parties such as insurers can pull a large amount of car data that can be used in a variety of processes from traffic modelling and planning to pricing an insurance policy.
For Eloy, this is where we feel a number of errors start to creep in, just as the problem of the human-in-the-loop creeps in for driverless cars. Raw data is very valuable but without wider human context, human behavioural data, and human outcomes, the raw data has diminished value.
An analogy can be drawn from a famous case from the Second World War. When damaged planes returned from missions, the US military examined the bullet holes to work out the best way to prevent further losses of planes and lives. As the amount of reinforcement was limited due to weight restrictions, their recommendation was to reinforce the places on the aircraft which had the most bullet holes.
The power of looking for what isn’t there
Until a statistician called Abraham Wald pointed out to them that they should be looking at the areas without bullet holes. His logic was that planes that had been hit in those places hadn’t returned and the bullet holes in the aircrafts they were examining represented areas that could sustain damage yet not render the plane unable to safely return to base.
Wald said that the reinforcement needed to go not where the bullet holes were, but where the bullet holes weren’t, in this case the engine. Without the human input the raw data was telling the US military to reinforce the planes in the wrong place.
Applying human behaviour to the roads and feedback loops
At Eloy we’ve been thinking about the arrival of CAVs from a specific niche area; if more data is being collected, how will that data be presented back to drivers? The base assumption that machines will process the data and it will appear in Level 5 autonomy is not going to happen. We need to maximise the power of data so that we can evolve through the various levels of autonomy.
In short, we need human feedback loops. Drivers will need to learn how to absorb additional road information and take action from it. This may be behavioural changes such as traffic light protocols, or it may be methods to reduce traffic around school drop-off and collection which is coordinated by a new machine learning traffic moderator.
These are huge connected car digital services that will emerge and will be vital to the evolution of autonomous driving.
Why Tesla is in prime position
If there is one company that has a huge advantage in driverless cars over any other it is Tesla. It is hard to read into autonomous car maker’s claims because so much of the work and shared tests have been done on deserted roads.
Tesla, however, stands out because it already has a car on the road that is at Level 2 but has all capability to go all the way to Level 5.
As a consequence, Tesla will be able to create human feedback loops to help them transition upwards towards Level 5. It requires no massively expensive and lengthy trials in Arizona that give only half the picture. The cars are already out on the roads collecting real-world data as their human drivers move about.
The next stage for Tesla will be to introduce digital services that help collect richer human data and move up the autonomy ladder. Tesla insurance will collect robust data on accidents and claims, and Tesla can incorporate this with the vital goal of reducing accidents and proving why accidents do or don’t happen with Teslas.
What about the rest?
At Eloy we’re agnostic about Tesla and other automakers. Tesla is a good distance ahead, but capital will go into other automakers. Automakers need to build a consortium that is capable of understanding the data transition, as long as they understand that the human factor is vitally important.
Apple’s iCar, potentially in partnership with existing manufacturers (my bet is on Kia or Hyundai as they were the first cars to incorporate Apple CarPlay) is an obvious place to look next.
At Eloy, we are building connected car services that can exist in all cars and will help with the upward transition to autonomy. And we have some exciting news coming soon about a few things I’ve mentioned in this article. Watch this space!
