The Ethics of Driverless Cars
Cars that are able to drive without any human input – without anyone in the driver’s seat and without anyone in control in the background – will soon be on our roads. Although such driverless cars are likely to offer huge environmental and social benefits, as well as making the roads overall much safer, they also raise difficult ethical issues. It is inevitable that driverless cars will have to make life and death decisions. Suppose a driverless car could only avoid killing five pedestrians by swerving into one. Suppose it can only avoid killing a pedestrian by putting the occupants of the car in danger.
My work on the ethics of driverless cars has two strands:
The New Trolley Problem: Driverless Cars and Deontological Distinctions
This paper has been published at the Journal of Applied Philosophy. The paper is available open access and can be found here.
Abstract
Discussion of the ethics of driverless cars has often focused on supposed real-life versions of the famous trolley problem. In these cases, a driverless car is in a position where crashing is unavoidable and all possible crashes risk harm: for example, it can either continue on its current path and crash into five pedestrians or swerve and crash into one pedestrian. There are significant disanalogies between the human versions of the trolley problem and situations faced by driverless cars which affect the application and moral significance of key deontological distinctions, such as the distinction between doing and allowing harm. The application and moral significance of the doing/allowing distinction in the context of the behaviour of driverless cars depend on (a) our conception of the behaviour of driverless cars; (b) the forms of driverless cars that are developed and used; (c) the background expectations of programmers/manufacturers/owners of driverless cars and the conditions of being able to put those cars on the roads. These are as yet unsettled – and may even be undetermined. Nonetheless, trolley problems may still be useful in thinking about the ethics of driverless cars.
Driverless Cars and the Need for Judgment
Discussion of the ethics of driverless cars is often set up by questioning which moral system it would be appropriate to program into the car, and then how such programming might be achieved. The key question is often seen to be what principles should the driverless car use to decide how to act in such life and death situations? Yet almost all contemporary moral theories recognise an ineliminable role for judgment. Correct moral judgments cannot be reached simply by applying an algorithm that could be programmed in advance from a human ethical theory.
The most obvious alternative is to use an artificial neural network. Artificial neural networks are increasing used in intelligent systems programming. They are trained rather than programmed in advance. For example, we teach a computer to recognise pictures of cats, not by giving it a set of conditions that make something count as a cat, but by 'showing' it lots of pictures of cats.
My work on the ethics of driverless cars has two strands:
The New Trolley Problem: Driverless Cars and Deontological Distinctions
This paper has been published at the Journal of Applied Philosophy. The paper is available open access and can be found here.
Abstract
Discussion of the ethics of driverless cars has often focused on supposed real-life versions of the famous trolley problem. In these cases, a driverless car is in a position where crashing is unavoidable and all possible crashes risk harm: for example, it can either continue on its current path and crash into five pedestrians or swerve and crash into one pedestrian. There are significant disanalogies between the human versions of the trolley problem and situations faced by driverless cars which affect the application and moral significance of key deontological distinctions, such as the distinction between doing and allowing harm. The application and moral significance of the doing/allowing distinction in the context of the behaviour of driverless cars depend on (a) our conception of the behaviour of driverless cars; (b) the forms of driverless cars that are developed and used; (c) the background expectations of programmers/manufacturers/owners of driverless cars and the conditions of being able to put those cars on the roads. These are as yet unsettled – and may even be undetermined. Nonetheless, trolley problems may still be useful in thinking about the ethics of driverless cars.
Driverless Cars and the Need for Judgment
Discussion of the ethics of driverless cars is often set up by questioning which moral system it would be appropriate to program into the car, and then how such programming might be achieved. The key question is often seen to be what principles should the driverless car use to decide how to act in such life and death situations? Yet almost all contemporary moral theories recognise an ineliminable role for judgment. Correct moral judgments cannot be reached simply by applying an algorithm that could be programmed in advance from a human ethical theory.
The most obvious alternative is to use an artificial neural network. Artificial neural networks are increasing used in intelligent systems programming. They are trained rather than programmed in advance. For example, we teach a computer to recognise pictures of cats, not by giving it a set of conditions that make something count as a cat, but by 'showing' it lots of pictures of cats.
But standard artificial neural networks don't provide an acceptable way to regulate the ethical decisions of driverless cars. Standard artificial neural networks are computational black boxes: we would be unable to control, or even to know, the basis for their decisions. This violates a key requirement of transparency: we need to understand the basis on which driverless cars - and other systems that shape our lives - make their decisions.
In this research project, I work with Will McNeill to explore the philosophical and ethical issues that we face when trying to resolve this dilemma.
In this research project, I work with Will McNeill to explore the philosophical and ethical issues that we face when trying to resolve this dilemma.