Blog

Autonomous car or philosophical car

• 2020-07-28
• Maciej Szczerba
• Business

Every 23 seconds a person dies on the ground in an accident involving a car. That is 4 000 people a day and almost a million and a half a year. When I write this entry (end of July 2020), worldometers show that since the beginning of the Covid-19 pandemic, 650 000 people have died from this disease. Therefore, unless the second wave of Covid-19 comes, we can cautiously assume that in 2020, globally, more people will lose their lives from traffic accidents than from coronavirus infection.

Is the solution therefore computer-controlled and not man-made cars? Unlike humans, the application does not feel tired, is not able to faint over the steering wheel, is not able to drive after alcohol.
The application also has no emotion. In a stressful situation it calculates what maneuver to make, what decision to make.

The concept of an "autonomous" or "unmanned" car is, from a technical point of view, a combination of sensors (sensors) with a camera system connected to an application based on an algorithm using neural networks.
The idea of the autonomous car is already around 15 years old, and has been accelerating strongly for about 10 years. 

What does 'autonomus driving' actually mean?

The American Road Safety Agency (NHTSA), which, among other things, is responsible for certifying NCAP crash tests, has developed a five-stage scale for the level of "autonomy" of vehicles:

0 - The car has no driver assistance systems.

1 - The car has specific driver assistance systems for driving, such as cruise control and blind spot warning when changing lanes.

2 - The car has systems that make decisions for the driver in specific situations. An example is the automatic braking system that Volvo first introduced, and today can be optional for most manufacturers.

3 - Conditional automation. The car moves independently, but only under certain driving conditions (weather conditions, certain road type, e.g. motorway, at a certain speed). The manufacturer who was the first to announce the entry to the third level of autonomy was Audi. In 2019 it showed the Traffic Jam Pilot system, which was to be installed in the current Audi A8. It allowed starting, driving and braking in the same lane on the motorway at speeds of up to 60 km/h, so it was a fully autonomous assistant for driving in traffic jams. However, Audi withdrew from installing the system this year, on the grounds that the current A8 is already halfway through production time. It seems that regulatory issues were also important here.

4 - High automation. The vehicle can be fully autonomous, but in certain situations, for example on a specific road network.

5- Full automation. The vehicle is fully autonomous, in all road and geographical conditions.

The two main players on the market are Tesla and Google (under the "Waymo" brand). These are also two completely different approaches to the creation of AI algorithms responsible for "driving autonomy".

Tesla installs a very advanced cruise control in its cars under the name Autopilot (nota bene it is classified as level 2 automation). Below is a video from Polish YouTube showing the Autopilot in practice:

https://www.youtube.com/watch?v=9kvcIdmA22E

The machine learning algorithms developed by Tesla learn from cases where the driver takes the lead from the Autopilot, assuming the car has not behaved correctly. Tesla thus collects data from 300,000 cars sold scattered around the world.

Tesla's main competitor, Waymo, has adopted a different strategy for building the algorithm. A fleet of about 100 vehicles equipped with Waymo systems (camera sensors), previously learned on the track, travels on specific roads and collects data about the car's behaviour. This is the so-called "bottom up approach". Perhaps much more accurate, but much more expensive.

Waymo boasted of the 10 million miles driven by its autonomous cars last year. Autonomous cars are allowed to drive in several American states, of which California can drive without a driver and carry people.

One of the most outstanding futurologists of the present time, Yuval Noah Harari (I will refer to him frequently on this blog) in his book "21 Lessons for the 21st Century", I warmly recommend:

http://www.publio.pl/21-lekcji-na-xxi-wiek-yuval-noah-harari,p182737.html

He cites an example when autonomous cars will have to solve serious ethical dilemmas. Let us assume that two children are running behind the hood of an autonomous car on a country road behind an escaping ball. The real time algorithm calculates that it has a 30% chance of slowing down and saving children. If it makes an instant left-hand turn, it will collide with the TIR coming from the opposite lane for 70%. Inside the TIRA, there is a 60-year-old driver. The driver is one, the children are two - I assume that the algorithm will not recognise potential victims for a long time to come, but will recognise the number of victims. Perhaps a human being, and not an algorithm, would try to slow down as much as possible by entering the fence/building and trying to save all three human lives ...

Yuval Harari cites the 2015 study in his book. The respondents were asked if the passenger (driver ?) of an autonomous car, in a situation of a collision with several pedestrians, should sacrifice himself and the car - the vast majority were in favour of it. However, when the question was changed and the respondents were asked how they would behave in such a situation, the answer was clear - we would save ourselves ...

In practice, as Harari emphasises with a bit of a black humour, we can imagine the future configuration of the car for a specific type of algorithm (just as today we configure additional options such as navigation or automatic transmission): "altruistic car" or "selfish car". The former will try to limit the consequences of an accident to third parties, the latter will only save passengers and even the car itself.

That is why there is so much controversy over the regulation of autonomous cars.Yuval Harari called this type of car, because of the above dilemmas, a philosophical car.