Driverless Vehicles and the Moral Algorithm

February 24, 2017

Our cars are set to change dramatically in the coming years.
Autonomous and connected vehicles will be the future of personal travel. They
promise better mobility for all, reduced congestion, an improved environment
and, perhaps most importantly, increased safety.

The large majority of road traffic accidents are caused
when a driver makes the wrong choice. AVs may never attain a flawless safety record,
but by removing the driver from the decision-making process they will remove
the source of most errors and therefore significantly improve safety: remove
the human, remove the risk. Yet, before we begin to delegate such decision-making
to autonomous vehicles we need to ask what decisions they can reasonably make
and on what basis.

The technology within AVs will
use an algorithm to make sure damage to both machine and human is minimised. Academics
and the press have referred to this type of mathematical solution to ethical
decision-making as the ‘moral algorithm’.

As part of UK Autodrive (the
largest of three separate consortia that are currently trialling automated
vehicle technology), my firm is contributing to the thinking on societal and
legal issues around the development of autonomous vehicles by developing a
series of white papers
. This article summarises the content of the latest
in the series on ‘The Moral Algorithm’.

The reality of creating a moral algorithm

Research suggests that while
academic debates on the moral algorithm may be intellectually stimulating, they
run ahead of the technology being programmed into today’s AVs.

This sentiment is echoed by members
of the UK Autodrive consortium, who stress that not only is it technologically
impossible to programme an algorithm with an infinite number of moral values, but
that it is not something any government would ever sign up to. Unless the social
norms of society completely change, no minister is going to stand up in
Parliament and say it is better to kill one person over another.

‘The trolley problem’ and why it’s overstated

Thus far, consideration of
the moral algorithm has tended to be focused on the ‘trolley problem’. The
trolley problem is a thought experiment in ethics which says:

‘There is a runaway trolley racing down the
railway tracks. Ahead of it there are two people who are tied to the tracks and
unable to move. If nothing else happens, they will be killed. However, you are
standing in the signal box next to a lever that will switch the trolley to a
side track. On the side track is one person, who also cannot move. Will you
take responsibility for pulling the lever that will kill that one person, or do
nothing and allow the two to die?’

In the world of autonomous
vehicles, it might be rephrased in this way:

‘There is an AV driving down the road. Someone else pulls into the road
unexpectedly. The AV does not have the necessary stopping distance available to
pull up safely so has to steer to one side or another.

One one pavement is an 80-year-old woman, on the other is a group of
children. Which party does the AV choose to put at risk?

Beyond the trolley

Although ‘the trolley problem’ grabs headlines when
discussing the moral values of driverless vehicles, AVs will potentially have
much wider scope for decision-making.

As drivers operating in this complex environment, the
decisions we make on the road are reflective of each of us as individuals.
Currently, you drive your own car in your own way – driving is seen as an
extension of your independence and personality.

Therefore, it will be important to widen thinking about the
moral algorithm to encompass the many lesser decisions that we make as part of
a journey.

For example, AVs will need to decide whether to enter an
area marked with chevrons and bordered by a solid white line in order to pass a
parked car, or break the speed limit to get out of the way of an ambulance.
These sorts of actions involve breaking road traffic laws in the interests of a
greater good such as keeping traffic flowing or giving priority to emergency

The moral algorithm will also need to be capable of
dealing with countless situations which do not entail a breach of the rules of
the road, but do concern the interaction of vehicles with others using the same
congested road space.

These are currently matters of driver choice, and are
reflected in different driving styles on the road – including, for instance,
how aggressively a vehicle accelerates or brakes, how sharply it changes lanes
or performs other manoeuvres, how much room it allows to other vehicles to do
the same things, and so on.

These are moral questions because they concern where
drivers choose to place themselves on a spectrum between the pure maximisation
of self-interest (getting from A to B as quickly as possible) and an altruistic
deferral to the interests of other road users. In practice, the same individual
may well adopt a mix of styles, or drive differently on different days as mood
or circumstances dictate.

AVs will, however, need to be programmed to deal
consistently with these real world driving issues. Examples of good practice
include ‘zip-fastener’ merging of slow traffic, taking turns where traffic
lights have failed at an intersection, giving way to other vehicles emerging
from side roads, stopping at zebra crossings, slowing down for horses and
allowing room to cyclists when overtaking.

And they will need to do this in a considerably more
complex environment, as they will be required to drive on busy roads and
interact with other cars (some manual, some autonomous), road users (such as
cyclists) and pedestrians.

Levels of autonomy

There are generally recognised to be six steps to full
autonomy, all of which progressively increase safety. 

Level 0 – Driver

Driver is responsible
for the vehicle. Controls lateral and longitudinal movement at all times.

Level 1 – Driver

Driver is
responsible for the vehicle. Controls lateral and longitudinal movement at all
times. However, the system can support lateral OR longitudinal control.

Level 2 – Advanced
driver assistance

Driver is
responsible for the vehicle. Controls lateral and longitudinal movement. May
hand some control over to the system. Must actively monitor system performance
and retake full control where necessary.

Level 3 – Advanced
driver assistance

Driver is
responsible for the vehicle. Controls lateral and longitudinal movement. Can
hand full control to the system. Must actively monitor system performance,
retaking control as necessary.

Level 4 – Highly

Driver is only
responsible, and exercises control, outside of specific use cases where the car
is able to self-drive.

Level 5 – Fully

System can control
lateral AND longitudinal movement in all use cases. Driver intervention is not

There are already
countless examples of vehicles with some element of automation already on the
road, such as self-parking or adaptive cruise control.

In spite of this,
conventional cars will also be around for many decades, so AVs will have to be
able to co-exist with them. It is possible that there could be separate roads
or lanes for AVs, but this option would be costly and in many places unrealistic.
So the ideal solution is for AVs to be able to share the road with their less
‘intelligent’ ancestors.

So, knowing that AVs will be sharing the roads with
conventional vehicles for many years ahead, on what basis should a set of moral
values be programmed into a vehicle? Should we give greater priority to
consistency, or should we reflect the individual preferences of travellers in
that vehicle, to the extent that the technology is capable of doing so?

Personal safety is

Those involved in AV development agree that it is almost
impossible to conceive of a transport system with a 100% safety record.

Yet, if the principal selling point of AVs is their
contribution to increases in safety, the level of risk we are willing to accept
may actually be quite low.

In a survey, researchers found that people wanted to drive
around in vehicles that would protect them and their family at all costs and
did not approve of regulation that would compromise the safety of passengers in
order to minimise casualties on the part of other road users.

The researchers concluded that regulation that enforced a
purely utilitarian moral algorithm would delay the uptake of AVs to such an
extent that ‘lives saved by making AVs utilitarian may be outnumbered by the
deaths caused by delaying the adoption of AVs altogether’.

Ultimately, given the safety imperative, the policy goal
must be to encourage the take-up of as many AVs as possible, as quickly as
possible. This requires co-ordination at state level, consistency, and a clear
legal framework which facilitates and encourages AV technology.

Co-operation is key

AVs not only make demands on car manufacturers’ research
and development teams, but also on the way their retail arms sell the end
product. Progress and take-up will be improved if manufacturers work together.

If all the stakeholders in the AV market really believe
that automation will equal a safer, more ethical transport system, then they
will have to find a way to work together to achieve this, while protecting and
developing their perceived unique selling points.


Although the industry may be improving co-operation, this by
itself is not enough. The safety imperative points to a need for state
regulation to facilitate and encourage the take-up of AVs by regulating the
moral algorithm rather than assuming a form of informal self-regulation by

In order to achieve the economic, mobility and safety
benefits of AVs, there needs to be a new and appropriate legal and regulatory
framework that is specifically designed for the purpose. Regulation needs to
lay a clear path so that existing and new participants in the market can safely
move forward, literally and metaphorically, by securing public trust and ensuring
that AV manufacturers are not working at legal risk. Regulation will need to
balance the potentially competing factors of adherence to the law, safety and
maximising take-up.

Our experts all believe that the current lack of bespoke
regulation leaves the industry in a state of limbo.

In a recent consultation on proposals to support advanced
driver assistance systems and automated vehicles, the government stated that it
would ‘continue to regulate in a rolling programme of reform. This will help to
facilitate the introduction of innovative new technologies in a safe, agile and
evidence-based manner for the benefit of UK consumers and business.’

Doubt delays

Research shows that waiting for a shift in public opinion
in order to at least set the boundaries for a regulatory regime could damage
the progress of AV development.

It is therefore important not to let the trolley problem
monopolise the debate regarding the moral algorithm. The trolley problem is a
distraction from the much larger category of day-to-day moral decisions and the
need to develop solutions to them.

There are numerous other hurdles to jump, including
problems with liability, cyber-security and public trust. It is particularly in
the area of public trust where the car industry and governments around the
world must start a conversation that coalesces differing moral and ethical

That is why ‘A programme of public education and
consultation’ is one of several areas for consideration suggested in our recent
white paper.

Other suggestions include:

The formulation of a coherent plan for
allocating regulatory responsibility for AVs, and in particular the moral
algorithms that govern their behaviour.

Consideration of how regulation – which
ultimately requires the oversight of both vehicle and road network regulation
in an integrated way – can best be implemented across the various tiers of UK
government including the devolved administrations.

The creation of a legal framework that is
sufficiently flexible and responsive to be able to make swift decisions about
new technology as it is developed, within a clear decision-making framework

As part of the development of the legal
framework, cooperation with the relevant international bodies to assist
in the development of international standards for AVs.

Consideration of how co-operation between sector
participants can best be facilitated (or required where necessary) to ensure
that AV development takes place as rapidly as possible and that CVs and AVs can
interact with each other.

The formulation of a coherent short, medium and
long term plan regarding the deployment of AVs on public roads in the UK.

Development of a policy regarding how the moral
algorithm will operate in terms of major safety situations. 

Given the technical challenges of
connected and autonomous vehicles, it is important that, before all of these
recommendations are implemented, the first step is to realise that regulatory
decisions need to be taken, and they need to be taken soon.

Stuart Young is Head of Automotive at Gowling WLG.