Hard Drive Crash

Author:Keri Grieman
Position:Queen Mary University of London; The Canadian Internet Policy and Public Interest Clinic
Keri Grieman
Hard Drive Crash
An Examination of Liability for Self-Driving Vehicles
by Keri Grieman*
© 2018 Keri Grieman
Everybody may disseminate this ar ticle by electronic m eans and make it available for downloa d under the terms and
conditions of the Digital P eer Publishing Licence (DPPL). A copy of the license text may be obtain ed at http://nbn-resolving.
Recommended citation: Keri Gr ieman, Hard Drive Crash: An Examination o f Liability for Self-Driving Vehicles, 9 (2018) J IPITEC
294 para 1.
Keywords: Self-driving vehicles; self-driving cars; vehicular liability; autonomous vehicles; AVs; artificial
intelligence; public policy; AV liability; self-driving vehicle liability
Discussion then turns to how liability might be al-
tered prospectively in order to incentivize outcomes
beneficial to both consumers and creators from a
public policy perspective. This includes a proposal
of how such a proposal might be structured. Focal
points include public policy, social acceptance, and
potential incidental problems raised.
Abstract: This analysis considers the poten-
tial impacts of completely self-driving vehicles on ve-
hicular liability. This begins with examining how such
vehicles might be treated under an evolution of the
current liability system, and the potential results of
attributing liability to an operator, the vehicle itself,
different manufacturers, and a government entity.
A. Introduction
Preventative maintenance is a benecial concept
to many industries - the pre-emptive “repair” of
areas that will become problematic in the future. It
is, however, a concept that rarely impacts common
law jurisdictions, where stare decisis rules the day.
Law very seldom pro-actively regulates activities,
particularly those of emerging technologies - one
cannot regulate what does not exist. How could one
have imagined the adaptation of privacy laws before
everyone carried a recording device in their pocket?
Moreover, regulating pre-emptively can serve to
quash the very innovation they attempt to pave the
way for.
2 Yet there are exceptions to this inability to predict
change. Areas that subtly adjust the way that we
interact with our world rather than radically altering
them. These are changes that we can see coming and
can conceivably prepare for without discarding the
current system. The self-driving car is such an area:
the modern world is already equipped with roads,
stoplights, and fuel pumps. We are not attempting to
regulate in a new dimension, no ying cars have yet
emerged; but the imminent changes would benet
enormously from pre-emptive adaptation.
If frameworks of legal liability for self-driving
or autonomous vehicles (AVs) are held off, the
potential benets of the AVs will be stied. This is
not to say that they will not come, merely that they
may come agonizingly slowly, as shareholders limit
the monetary risks they are willing to take. Nor is it
suggested that the changes required are simple, but
that they are necessary. It is important to balance
proactivity with over-regulation, and the difculty of
post-ante regulation with administrative efciency.
Vehicular liability must be written to incorporate
AVs. A system that reects the underlying
differences between AVs and human drivers
encourages benecial change. In order to achieve
this change as efciently and cohesively as possible,
AV legislation should be written proactively, rather
than allowing the question of liability to bring change
incrementally and with crippling uncertainty. Such a
legislation system may be best complemented by the
creation of an independent public insurance entity.
Hard Drive Crash
B. Assumptions
5 The central tenet of vehicular laws in many, if not
all, common law countries is fault. Who is liable,
when they are liable, and why. Rules of the road
are written to reect what one can and cannot do,
resulting in fault when one fails to follow them. For
this reason, analysis will focus on fully autonomous
vehicles - those that do not require a human driver
whatsoever, and taking those countries basing
liability on fault as a starting point. There are
recognized levels of autonomy within the industry:
from an entirely human driven vehicle at 0, to an
entirely human excluded one at 5.1 Where a human
driver is required or expected to maintain full or
partial control of the vehicle, regular conceptions
of liability are imperfect, but may be sufcient.
Partially autonomous vehicle components can be
turned off, as can components of full autonomy, such
as self-parking.2 Level four autonomous vehicles
are indeed fully automated, but are not capable of
covering every driving scenario,3 and have already
been rolled out in some areas - namely Las Vegas
and Singapore city centre,5 although they are limited
* Queen Mary University of London; The Canadian Internet
Policy and Public Interest Clinic.
1 Hope Reese, ‘Updated: Autonomous driving levels 0 to 5:
Understanding the Differences’ TechRepublic (20 Janaury
accessed 12 January 2018.
2 ibid - breakdown of vehicular autonomy levels
0: The human driver is in complete control.
1: The human driver still holds the majority of control, but
a specic function, such as accelerating, may be done by the
2: “[A]t least one driver assistance system of both steering
and acceleration/deceleration using information about the
environment is automated, like cruise control and lane-
centering.” The driver may be incrementally separated
from the operation of the vehicle, but must remain ready to
re-take control in an instant.
3.’Safety-critical’ functions are taken by the vehicle. While
the driver must be able to intervene, they are not required
to re-acquire control instantaneously.
4. The vehicle is able to perform all necessary functions, but
not under all conditions.
5. The vehicle is able to perform all necessary functions
under all conditions considered safe enough for a human to
operate a vehicle.
3 Natasha Merat and others, ‘Driver behaviour when resuming
control from a highly automated vehicle’ (Institute for
Transport Studies, University of Leeds, 16 October 2014),
4 Saqib Shah, ‘Las Vegas’ self-driving bus crashes in rst
hour of service’ Engadget UK (11 November 2017)
bus-crash/> accessed 12 January 2018.
5 Andrew J Hawkins, ‘Singapore’s self-driving cars can now
be hailed with a smartphone’ The Verge (22 September 2016)
to a dened environment. While these vehicles can
be used as independent taxis, it will be assumed that
they can currently be run under transit-like liability,
particularly given that their activity is currently
conned to a dened area. The scope of this paper
will primarily be concentrated on privately-owned
vehicles. It will also be generally assumed that
society is in favour of a system that allows for the
compensation of victims in vehicular accidents.
While no specic jurisdiction will be focused on,
Canada provides a helpful, broad set of examples
as it employs different insurance systems in each
province and territory, but uniformly bases liability
on fault.
Finally, exceptionally rigorous testing will be
assumed. In order to be allowed to enter the
market, relevant regulators should conduct
stringent testing under a variety of conditions for all
different manufacturers and models. Cars are heavy
machinery, and their destructive potential should
not be underestimated. While manufacturers will
undoubtedly conduct in-depth testing themselves,
an entity independent from the company needs to
test the vehicles in question to ensure a sufcient
level of safety and driving quality.
C. Technical Aspects
7 There are several typical elements that are used by
AV manufacturers order to allow the car to function.
These include a video camera mounted on or near
the front windshield allowing for the detection of
trafc lights and moving objects; a rotating sensor
on the roof which scans the area in a large radius,
creating a three-dimensional map; distance sensors
on the bumpers to measure space between various
obstacles; smart-navigation maps updating in real
time to track accidents, speed limits, and car-to-car
communication; and the articial intelligence that
commands the control centre.6 These methods are,
as yet, imperfect - sensors struggle with inclement
weather, and the roof sensor aka LIDAR (light
detection and ranging) faces problems with bright
sunlight. 7 The technology in the marketplace has not
driving-car-nutonomy-grab-ride-hail-test> accessed 12
January 2018.
6 Muhammad Amat, Dr Clemens Schumayer, ‘Self Driving
Cars: Future has already begun’ (Institute of Transport and
Logistics, Vienna University of Economics and Business, 7
May 2015)
accessed 1 January 2018, 8.
7 ibid 5; A fatal crash occurred where a Tesla detected bright
sunlight reecting off a truck as a cloud rather than an
obstacle. While the driver was supposed to be able to regain
control, and the Autopilot feature was not intended to
Keri Grieman
yet reached level 5, though when it does, accidents
are still to be expected. While it is possible and
indeed likely that adaptations and new technologies
will emerge, the aforementioned will serve as a
minimum level of AV “competency” – that full AVs
will have at least these levels of technology available
to them.
D. Public Policy
The changes brought about by AVs will impact
society in many ways, and not all of them will be
positive. Challenges may be obscure - a decrease in
car accidents has the potential to result in an even
greater shortage of organs available for donation.8
Impacts have been noted to range as far as the airline
and hotel industries, predicting that as long-distance
automobile transit becomes more convenient and
comfortable, air travel will become less competitive.9
More directly, automation will bring about the loss
of work for many, including professional drivers.
In 2014, it was reported that more than 4.4 million
persons in the United States alone worked as
drivers.10 While it may in turn bring new jobs, the
specics of such work remain to be seen. In terms
of hired driving services, some stakeholders are
already making their investments – in 2015 then
CEO of Uber, Travis Kalnick, stated the intention
to replace human drivers with AVs.11Undoubtedly,
there will be opposition to AVs for various reasons,
and astute commentators note that unions for
drivers will likely respond to the challenge to their
replace human senses, it does demonstrate the problem
for future vehicles and their sensors. Neal Boudette,
‘Tesla’s Self-Driving System Cleared in Deadly Crash’ New
York Times (New York, 19 January 2017)
fatalcrash.html?_r=0> accessed 1 January 2018.
8 Ian Adams, Anne Hobson, ‘Self-Driving Cars Will Make
Organ Shortages Even Worse’ Future Tense (30 December
shortages.html> accessed 1 January 2018.
9 Kevin LaRoche, Robert Love, Autonomous vehicles:
Revolutionizing Our World” Borden Ladner Gervais LLP
(2016) blg.com/en/News-And-Publications/Documents/
Autonomous-Vehicles2016.pdf> accessed 18 June 2017,
citing Autonomous cars will make domestic ights run
for the money: Audi Telematics Wire (27 November 2015)
ights-run-for-themoney-audi/> accessed 18 June 2017.
10 Mark Fahey, ’Driverless cars will kill the most jobs in
select US states’ CNBC (2 September 2016)
in-select-us-states.html> accessed 4 January 2017.
11 Stephen Edelstein, Uber CEO to Tesla: Sell me half a million
autonomous electric cars in 2020 Green Car Reports (7 July
in-2020> accessed 18 June 2018.
profession by raising doubt about AV safety, and
lobbying against them.12 AVs are likely to face more
opposition than most changes, given that humans
appear to have an inherent distrust of non-human
intelligence. While evidence-based algorithms are
shown to be more accurate than humans, people
lose condence in the algorithm more quickly than
humans, and continue to prefer the human even
where the algorithm consistently outperforms the
9 AVs are indisputably on their way. Both traditional
and disruptive automakers are steering into the skid
of AVs – even by 2012, Google’s AV had completed
over 300,000 miles of accident-free self-driving.14 AVs
are already in commercial use: AV trucks transport
mining materials in the Australian outback, and self-
driving tractors are already in the eld.15 Moreover,
autopilot systems have been used by commercial jets
for many years, aiding in maneuvering, navigation,
and landing, lending “a signicant amount of
automated assistance,” and allowing planes to land
in conditions that were “previously difcult for
human pilots.”16
Accident reduction is a crucial potential benet.
While AVs are imperfect, they lack the failings
endemic to human drivers: limited scope of vision;
ability to be distracted; inability to focus on multiple
areas at once; etc. Even more importantly, they lack
the ability to be affected by the same level of choice
as humans in terms of driving; namely, humans can
and do drive when tired, ill, or impaired. Advocates
of AV note that more than 90% of trafc collisions are
caused by human error17 - while AVs are imperfect,
12 Ratan Hudda and others, ‘Self Driving Cars’ (Fung Institute
for Engineering Leadership, UC Berkeley College of
Engineering, 29 May 2013)
wordpress.com/2013/06/self_driving_cars.pdf> accessed
1 January 2018, 9.
13 Berkeley J Dietvorst, Joseph P Simmons, Cade Massey,
‘Algorithm Aversion: People Erroneously Avoid Algorithms
after Seeing Them Err’ (July 2014) Journal of Experimental
Psychology: General, forthcoming
com/sol3/papers.cfm?abstract_id=2466040> accessed
2 January 2018.
14 ibid.
15 David Robson, ‘The Truth About Driverless Vehicles’
BBC (London, 13 October 2014)
story/20141013-convoys-of-huge-zombie-truck> accessed
30 December 2017.
16 Harry Surden, Mary-Anne Williams, ‘Technological
Opacity, Predictability, and Self-Driving Cars’ (2016)
Cardozo Law Review 38:121
edu/articles/24> accessed 30 December 2018, citing Simon
Wood, ‘Flight Crew Reliance on Automation’ Federal
Aviation Administration Advanced Avionics Handbook
(UK Civil Aviation Authority, 2009)
17 Emily Chung, ‘Autonomous cars could save Canadians $65B
a year’ CBC News (Toronto, 21 January 2015)
Hard Drive Crash
they will not eliminate these accidents immediately,
but they have the potential to greatly reduce such
accidents. Currently, the World Health Organization
estimates that injuries caused by road trafc will
become the worldwide fth leading cause of death
by 2030.
In the United States, automobile accidents
are “the lead cause of death for people between the
ages of 3 and 34,” with a death every 30 seconds.19
It is estimated that in the United States alone, AVs
could save 300,000 lives per decade - 29,447 lives per
year,20 and as much as $190 billion USD per year in
health costs.21 However desirable these miraculous
predictions, they depend on a minimum level of
widespread adoption of AVs.22
Infrastructure efciency and cost will be directly
impacted. Worldwide, the cost of trafc congestion
is estimated to reach $2,200 billion USD per year.
northern North America, self-driving cars have been
predicted to save $65 billion CAD by reducing trafc
congestion, fuel costs, and “time wasted behind the
wheel.”24 In reducing the need for car ownership,
$5 billion CAD can be saved on congestion costs
Google has already built the largest trafc
jam surveillance network in the world by providing
over 500 million smart phones with an operating
system - the mapping function allows Google to
track trends over time.
Independent researchers
have modelled an algorithm that allows signicant
alleviation of trafc jams by multi-vehicle routing,
and requires only 10% of vehicles on the road to
follow the algorithm.
In other words, benets of
AVs need not reach a majority before they produce
tangible infrastructure benets - only a minimum
point of saturation.
65b-a-year-1.2926795> accessed 1 January 2018; Berkeley
J Dietvorst, Joseph P Simmons, Cade Massey, ‘Algorithm
Aversion: People Erroneously Avoid Algorithms after Seeing
Them Err’ (n 13) 6.
18 Ratan Hudda and others, ‘Self Driving Cars’ (n 12) 5-6.
19 ibid 0.
20 Adrienne Lafrance, ’Self-Driving Cars Could Save 300,000
Lives Per Decade in America’ The Atlantic (29 September 2015)
america/407956/> accessed 1 January 2018.
21 ibid.
22 ibid.
23 Hongliang Guo and others, ‘Routing Multiple Vehicles
Cooperatively: Minimizing Road Network Breakdown
Probability,’ (2017) 1/2 IEEE Transactions on Emerging Topics
in Computational Intelligence.
24 Vijay Gill and others, ’Automated Vehicles: The Coming
of the Next Disruptive Technology’ The Conference Board
of Canada (21 January 2015)
library/abstract.aspx?did=6744> accessed 1 January 2018.
25 Emily Chung, ‘Autonomous cars could save Canadians $65B
a year’ (n 18) 1.
26 Ratan Hudda and others, ‘Self Driving Cars’ (n 12).
27 Honglian Guo and others, ‘Routing Multiple Vehicles
Cooperatively’ (n 23) 121.
12 In terms of accessibility, AVs would open an entire
world to those unable to drive themselves. Many
individuals are, for reasons of age, physical ability,
or current state, unable to drive. These individuals
are dependent on either public transit, expensive
private means of transport, such as taxis, or family
and friends. It has been noted that allowing these
individuals increased access to transportation has
the potential to increase total vehicle transit by up
to 11%.28 While this obviously increases demand, it is
cause for celebration as these individuals evidently
do not have the freedom or capability to travel as
much as their able-bodied counterparts.
Ecologically, there are also several benets. Even
with the current state of technology that is expected
to improve, projections have placed the reduction
of oil consumption and related greenhouse gas
emissions at 2 to 4%.29 These predictions were
based on the use of technologies such as “adaptive
cruise control, eco-navigation, and wireless
communications.”30 The ease of incorporating AVs
with other technologies has even greater potential,
with “car to infrastructure communication”31 - one
“smart” parking system reduced time spent looking
for spaces by 21%.
Additionally, AVs do not need to
park in a space that is convenient or easily accessible
– they can park underground or remotely, and the
driver can summon the car when required. A trafc
signal synchronization program saved “31.2 million
hours of travel time, 38 million gallons of fuel and
337,000 metric tonnes of carbon dioxide per year.”33
Furthermore, most cars are unused for 95% of their
lifespan, but AVs have the potential to reduce the
amount of cars on the roads overall, as AVs can
be farmed out for others when not in use by the
owner.34 Car sharing programs have led to less car
ownership, and a reduction of emissions in cities,35
28 Todd Litman, ‘Autonomous Vehicle Implementation
Predictions: Implications for Transport Planning’ Victoria
Transport Policy Institute (22 December 2017)
www.vtpi.org/avip.pdf> accessed 1 January 2018 13; citing
Michael Sivak, Brandon Schoettle, ‘Road Safety With Self-
Driving Vehicles: General Limitations And Road Sharing
With Conventional Vehicles, Sustainable Worldwide
Transportation Program’ University of Michigan (2015)
29 Julia Pyper, ‘Self-Driving cars Could Cut Greenhouse
Gas Pollution’ (15 September 2014) Scientic American
cars-could-cut-greenhouse-gas-pollution/> accessed
1 January 2018.
30 ibid.
31 Muhammad Amat, Dr Clemens Schumayer, ‘Self Driving
Cars: Future has already begun’ (n 6) 13.
32 ibid.
33 ibid.
34 Ratan Hudda and others, ‘Self Driving Cars’ (n 12).
35 Darrell Etherington, ‘Car sharing leads to reduced car
ownership and emissions in cities, study nds’ Tech Crunch
(19 July 2016)
Keri Grieman
a benet which is likely to increase as it cuts into
the requirements for taxis and other chauffeuring
needs. Finally, there is the simplest benet of all:
not having to drive.
Whilst the advantages are numerous, the technology
remains vulnerable to smothering by the tyranny
of the immediate - the defence of the bottom line
in companies protecting themselves from liability,
and legislation in taking a “wait and see” approach.
E. Liability
For all the many benets of AVs, they are imperfect.
Accidents will still happen, particularly in the early
years. It is thus important to determine what party
is potentially liable; specically, who should pay for
any damages incurred as well as compensation to
the victim. While current liability systems will need
to be tweaked to allow for integrated AI driving; i.e.
for vehicles between levels 1 and 5, their setup still
allows for and generally requires a human driver
to take control. In aiming to fully achieve their full
societal benet, level 5 vehicles should have no
interaction from the driver. This raises the obvious
question as to who should be liable and how.
Informed commentators have suggested that
parties potentially liable for AV accidents could
include the user, the owner, the manufacturer, the
manufacturer of AV components, or a government
entity. Methods such as product liability have the
potential to cause difculties both in the expense
incurred through the legal process in determining
liability, and in determining how and why an AV
made the “decision” that it did; class action suits
are too cumbersome for something as ordinary as
auto accidents. What mechanism, therefore, should
be used to allocate liability? As will be examined,
negligence under our current conception of the
notion, has the potential to prove problematic in
allocation of liability.
I. Potential Liability Allocated
to the Operator
Given that one of the potential benets of AVs is
increased car-sharing, it is possible that the user and
owner may be different individuals. The user might
simply be someone who has independently hired the
car through, say, a taxi service app. The owner is the
person who has technical ownership of the AV. For
the purposes of legal application, one can treat the
in-cities-study-nds/> accessed 1 January 2018.
user, owner, or general occupant as one entity, as
they run into the same potential concerns. For the
sake of discussion, these entities can be rened into
one, the “operator” – the denition of which should
rely on the individual determining the destination.
In a liability context, the operator is the entity
who is most closely aligned with current fault
attribution. While each country differs slightly
in their application of the law, vehicular liability
generally relies on the individual who has control
of the vehicle. In Canada, for example, section 214
of the Canadian Criminal Code states that to operate
“means, in respect of a motor vehicle, to drive the
vehicle.”36 Crimes such as operating under the
inuence rely on this denition of operation, and
on the concept of “the care or control of a motor
vehicle… whether it is in motion or not.”
“Care and
control” has included situations such as a passenger
grabbing the steering wheel,38 sitting in the driver’s
seat “braking and steering an inoperable vehicle,”39
or using the steering wheel while being towed, as
noted by Osler PJ in R v Morton:40
when, though the means of propulsion is under the control
of the driver of a towing vehicle, there is a person in charge
of the towed vehicle who is manipulating the steering wheel
and brakes and exercising a signicant measure of control
over the direction and movement of that vehicle, I consider
that person can be said to be operating or driving the motor
19 In other words, determining liability of an operator
has centred around their intent and ability to
inuence the movement of a vehicle through
functions in the province of a driver. Many
jurisdictions are willing to nd drivers liable for
driving under the inuence of intoxicants even if
they were not in the driver’s seat, nor piloting the car,
but were in the car and had access to the keys. The
capacity to direct the car, whether or not in current
use, has been used to determine care and control,
and thus liability. This approach does not make sense
for AVs. The intent of a fully autonomous vehicle is
that the occupant will not have control, and thus will
not be able to direct the specic movements of the
vehicle. The occupant may have the ability to direct
the car generally - they are, after all, determining
the end destination of the AV. However, “care and
control” does not make this distinction.
36 ‘An Act respecting the criminal law’ (RSC 1985) c C-46,
‘Canadian Criminal Code’ section 214.
37 ibid section 2(a).
38 R v Belanger, [1970] 10 CRNS 373 (SCC).
39 R v Flemming [1980], 43 NSR (2d) 249 (NS Co Ct), cited in R v
Danji, [2005] ONCJ 70, 16 18 MVR (5th) 1.
40 R v Morton [1970], 12 CRNS 76 (BCPC).
Hard Drive Crash
20 An operator of an AV is analogous to the passenger
of a bus. They have an ultimate destination in mind.
They are capable of inuencing the vehicle’s path
by asking the driver to stop, potentially by pushing
a button or pulling a cord. If the bus were to be in
an accident, however, even if the bus is inarguably
at fault, the bus passenger is in no way responsible
for that accident or the damages resulting from it.
Operators of an AV have no less a duty of care to the
occupants of other vehicle than an ordinary human
driver does, but attempting to extend liability does
not, from a logical standpoint, make sense. The duty
of care may include not interfering with drivers,
or distracting them, but should not overextend to
include the “but for” test – i.e. “but for” the occupant
choosing to use the AV, the accident would not have
occurred. This is simply too broad to be functional.
In the case of an accident between an AV and a
human driver, the legal result would depend on
which vehicle were at fault. If the AV were at fault,
the previous issue arises: the occupant is unlikely
to have acted negligently or unreasonably. If the
human driver were at fault, all current laws are
easily applicable. If fault is mixed, the court can
apportion damage based on contribution to the
harm, as is common in many areas of law, but the
AV portion should not fall on the operator.
There are cases where traditional conceptions of
liability should apply, namely where the operator
had previous knowledge of a potential issue with the
AV. AVs have the potential to learn, and better their
“driving”. This is a desirable feature of AVs - not
only can AVs learn from their own behaviour, but
potentially the behaviour of other vehicles capable
of communicating with them. Such a system is likely
to function on an update system, similar to updates
on a computer or smart phone. This could result in
a situation where an operator, or an owner, were
confronted with a notice warning them of a defect
with the car’s programming, or a potential update.
If the operator were to ignore this warning and
continue to use the vehicle, they can and should be
found liable for an accident resulting from the lapse
in update. This may be an extreme outlier scenario
but serves to sufciently include the operator’s
23 Additionally, if an individual - whether owner, user,
operator, or unrelated party - were found to have
tampered with any programming impacting the AV’s
ability to function safely, this could produce a range
of liability. This range should run from negligence
to attempted murder, depending on what happened
and how, such as if it was intended to affect another
operator. This does limit the operator’s freedom to
adjust their vehicle’s programming as they would
like, but such a step is crucial to the uniformity and
thus predictability of AVs - a necessary requirement
for ensuring the safety potential of the vehicle.
Despite these minor exceptions, conceptions of
liability surrounding the vehicle’s operator must be
updated to reect the reality that the operator does
not, and should not, affect the “decision making” of
the vehicle. This is the societally desirable outcome
- removing the operator from the second-to-second
decision making process is what allows the AV to
drive in a way that avoids human failings. In the same
way that a taxi passenger has made a responsible
decision and thus should not be charged with driving
under the inuence, neither should an AV operator
be at the mercy of decisions which are not their own.
II. Potential Liability Allocated
to the AV Itself
25 The AV itself is not a logical successor to the human
driver in terms of liability, although it may at
rst glance appear to be so. The entity that best
ts current conceptions of liability in terms of
“driving” and “care and control” of the vehicle is the
articial intelligence entity that enables the AV. For
simplicity’s sake, the AI and the AV will be treated
as a singular entity given their inseparability for the
purpose in question.
Determining whether the AV has made a “wrong”
decision may require extensive evaluation of the
way in which it makes decisions. It may require a
comprehensive understanding of how decisions
were made, and what information was available.
Requiring the AV to take on responsibility for actions
taken implies a level of responsibility. However,
there are two problems with this: rst, from a
functional standpoint, the AV has no assets except,
potentially, itself. In an accident, the victim is to be
compensated for damage to the vehicle, injury, etc.
However, without delving into an analysis of robo-
slavery, it is clear that an AV does not own anything,
whether or not it owns itself. Accordingly, whether
or not the AV owns itself, depriving the owner of the
AV is detrimental to the owner, rather than the AV.41
Second, the AV’s decision making originally depends
on how it was programmed. While it may “learn”,
its key input is given before it ever hits the road
41 Interestingly, a robot has been already been ‘arrested’ for
its actions. A robot in Switzerland was created by a group
of artists and given a bitcoin budget per week to randomly
purchase from the dark web, with the intention of
displaying the items purchased. The robot was conscated
along with its purchases, which included a passport and
ecstasy tablets, but was returned three months later with
all purchases except the Ecstasy. Arjun Kharpal, ‘Robot with
$100 bitcoin buys drugs, gets arrested’ CBC Tech Transformers
(Ottawa, 22 April 2015)
with-100-bitcoin-buys-drugs-gets-arrested.html> accessed
1 January 2018.
Keri Grieman
- it does not inherently “choose” to do something
wrong, it follows directions that it has been given.
This is not the sort of “guilty mind” or mens rea
envisioned by current legal regimes. Moreover,
this approach to liability would assume that the
AV both can and does “think” like a human, and
thus could be assessed to the same standards. Even
the ways that the AV “learns”, or what it “learns”
about, are initiated by its programming, and are not
inherently based on human thought patterns. The
AV’s “decisions” are not the same as human ones.
This evokes the question of whether the party that
originally programmed the AV should be liable for
what the AV is programmed to do.
III. Potential Liability Allocated
to the Manufacturer - parts
Manufacturing can be separated into two parts:
the main manufacturer or assembler, and parts
manufacturers. Consider rst the parts themselves.
Continuing to treat parts manufacturers under
traditional common law liability understandings
does not seem particularly problematic - main
manufacturers maintain the duty to check parts
they buy to a reasonable standard, and the parts
manufacturers maintains the duty to manufacture
them to the standard promised. Individual parts
currently account for relatively few accidents, and
there is no reason to believe the relatively low rate of
product liability suits or issues would increase. While
machinery for AVs may be more complicated, even
vehicles that are not fully autonomous are improving
at tasks like diagnosing parts or physical issues with
vehicles. While product liability suits are slow and
costly, the relatively small-scale requirements for
individual faulty parts means that this is likely still
a functional way to address the problem without a
systemic overhaul.
IV. Potential Liability Allocated to the
Manufacturer - programming
First, some denitional clarication. It has been
suggested that Google is likely to license a developed
version of its AV software to car manufacturers,
allowing for a prospective licensing industry
alongside the AV market.42 However, given that
Google has a successful AV of its own, and major
automotive manufacturers are creating their own
AVs, discussion will focus around manufacturers as
having produced and programmed their own AVs.
42 Ratan Hudda and others, ‘Self Driving Cars’ (n 12).
In the current state of the market, most
manufacturers selling vehicles that have AV features
state that the driver must be able to take control at
all times, and that any autonomous features are not
in fact self-driving; thus by using the vehicle, the
“driver” conrms that they will always be “driving”,
even if the car is able to function in any way on its
own. This appears to be an attempt to potentially
contract out of liability in favour of having the driver
agree to assume it. Whether or not this will hold up
to substantial legal challenges remains to be seen,
either in tort liability or contractual restrictions,
but it nonetheless appears to be the current method
attempted. The current state of the liability union
is divided - automotive companies and Google have
lobbied governments to absolve them of liability -
to negative effect in California, but positive effect
in Nevada.43 Volvo has already made public its
willingness to accept full liability, whereas Tesla
has stated that it will accept liability only for design
Whether liability should fall on major manufacturers
through the decisions made by their agents in
programming an AV, and on the subsequent decisions
of the AVs acting on that programming, opens an
obvious chain of questioning. While removed from
the immediacy of the road, programing largely
ts the conception of “control” over the vehicle -
how it is driven, when and where it stops, how to
react to changes in the environment. Programmers
for manufacturers, acting in their professional
capacity, could be treated as creating liability for
the manufacturer in embedding their decisions, even
if it is an initial step in a machine learning process.
This is compounded by the “black box” problem - it is
often difcult for articial intelligence to “explain”
why it did what it did - the AV in question might
have weighed many factors, and learned from many
sources, which ultimately resulted in a particular
action. Elon Musk, co-founder, CEO and Product
Architect at Tesla, used the following analogy:
Point of views on autonomous cars are much like being stuck
in an elevator in a building. Does the Otis [Elevator Company]
take responsibility for all elevators around the world, no they
This presents an interesting point. Programming the
way something works has not previously resulted
in major liability. Nor has it prevented society from
doing away with elevator attendants, or in the case
of cars, drivers. However, not only do the number of
43 Ratan Hudda and others, ‘Self Driving Cars’ (n 12) 6.
44 Danielle Muoio, ‘Elon Musk: Tesla not liable for driverless
car crashes unless it’s design related’ Business Insider
(Sydney, 20 October 2016)
accessed 1 January 2018.
45 ibid.
Hard Drive Crash
elevator accidents pale in comparison to the number
of car accidents - even proportionately - the elevator
deals with a pre-set course with no obstacles or other
players, programmed or otherwise. Cars must deal
with a great deal more and put more lives at risk than
just those inside of it, and there is an inherent level
of “decision making” involved.
32 Priorities for the AV are set in advance. This often
brings to mind philosophical debates such as the
trolley problem, wherein one must choose whether
to divert a trolley hitting three people instead of
hitting one person. However, problems like this
do not address what AVs are, or are intended to
do: AVs are not intended to make a choice of the
amount of humans tied to the train track to kill.
They are intended to stop the trolley. Treating AV
“decisions” as identical to human ones ignores the
reality that AVs can work with far more input than
humans can: 360 degrees of vision, multiple heights
and layers of sensors, and a lack of distractions.
If AVs can communicate with one another, and
there are enough to do so, they could provide
information in real time; for example, “up-coming
pothole” and “group of joggers on road shoulder”
are not particularly difcult messages to transmit.
This translates into larger concepts as well, such as
“human-sized entity darting into trafc”. The world
is not tied to two tracks and no breaks, and reducing
the decisions to be made to such a scenario fetters
our understanding of what could be.
Statutorily pinning liability on manufacturers forces
them to prioritize liability. This does not mean that
manufacturers would place it as a rst priority -
human life is likely to forever hold the primary spot,
if only because cases of deaths may kill public favour
of AVs. But it does inevitably affect priorities as a
whole. Damage to property is certainly preferable
to damage to humans, yet focusing on liability may
shift this emphasis. It is entirely possible to be both
in the right legally, yet making the wrong decision.
While measures such as the strict liability approach
of capping the amount of damages to be paid may
be reasonable stopgaps, they present their own
domino issues - potentially neither covering the full
amount of damages, nor removing the incentive to
de-prioritize physical damage in favour of safety.
Consider a situation wherein an AV is suddenly
faced with an obstacle it can either hit lightly,
causing no injury, or stop immediately and cause
the human car speeding behind it to injure either
the AV and the speeding car’s occupants. In a case
where liability is not in question, and human safety
is the highest priority, the AV hits the obstacle -
damaging the AV, but neither set of passengers.
If liability is a priority, the AV avoids liability by
coming to a stop as the human driver would be in the
wrong through speeding, and being unable to make
a safe stop without hitting the AV. However, this is
not the societally desirable outcome: car parts are
replaceable, human health is not. It is quite possible
to be correct in law but not in morality, and the
concern for liability means the prioritization of cost
and correctness over better outcomes. Mandating
liability means incentivizing the wrong priorities.
As for the trolley problem, we want the AV to stop
the trolley, even if it means breaking said trolley.
While instances such as negligent or malevolent
programming should still be considered, from a
public policy perspective, governments should
encourage manufacturers to take safety of all parties
as the highest consideration. As AVs reach a point of
saturation, these priorities will have an increasing
impact and importance. Statutorily mandated
liability on manufacturers does not make vehicles
safer in and of themselves - it reinforces the priority
of doing the legally correct action, rather than the
socially benecial one. Allocation of liability for non-
human damage simply does not produce the best
incentivized outcome for social priorities.
Furthermore, if liability is focused on manufacturers,
risk is concentrated onto a concerningly small
number of entities, who will simply increase product
prices to cover the risk at an even greater rate
considering the unknown cost to the manufacturer
themselves. The current system of liability and
spreading liability cost, transfers the price to a
later point in the transaction, but allows for greater
predictability and a greater sharing of the smaller,
more predictable cost.
V. Potential Liability Allocated
to a Government Entity
Ultimately, insurance will still be necessary for AVs.
There will be accidents, and thus accident victims. An
insurance infrastructure will ensure compensation
for these victims and help to establish the viability
of AVs as an institution. As previously discussed,
naming one or a combination of the previous actors
and stakeholders presents many problems. Liability
needs to be apportioned without a concept of “blame”
- damage has occurred, and the damage needs to be
xed or compensated for. A strict liability regime is
a functional way to accomplish this and legislating it
pre-emptively for level ve AVs has the signicant
benet of predictability.
AV manufacturers are understandably concerned
with the extent to which they will be liable, and in
what ways. Companies have been easing slowly into
full automation by using automated features, being
careful to mandate that the driver must still be in
control - thus avoiding liability. A “wait and see”
Keri Grieman
approach to legislation means that manufacturers are
understandably hesitant to be the rst to throw their
hat into the ring with commercial, fully automated
vehicles. It also means that smaller companies are
struck from the automation race completely, as they
lack the war chest to fund costly litigation when
an accident occurs. Providing assurances allows
manufacturers to bring an actual product to market
- the societally desirable, completely hands-off, AVs.
How, then, should this system be structured? Ideally,
at least initially, as a government-run, AV-mandated
single-pool insurance entity through which all AVs
must be insured. First, such an entity has the initial
benet of actually providing insurance rather than
waiting for the private sector to enter the market.
Second, time and proteering can be avoided by
circumventing the private insurance sector. Third,
it allows for a specialized entity to deal with the
information created by accidents; specically,
assessing it, and passing it along to the necessary
parties, such as the manufacturer, when there is a
clear problem with the AV system. Fourth, it allows
for the reduction of administrative work - no time
and effort is spent resolving damages between AV
insurance providers; rather the costs are simply paid
and the accident can be analyzed from a systemic
perspective, i.e. could the AV have made a better
“decision”? While non-AV insurance providers
will still have dealings between themselves, they
too benet from a single-system for AVs, such as
a standardized system that specializes in how AVs
function, and can thus concentrate on, for example,
provision of crash footage in the case of a combined
AV/human accident. This is not to say that AVs
should suddenly become liable for all accidents they
are involved in, but rather those where a human
driver would similarly be found at fault.
1. Avoiding the private insurance sector
Single pool compensation has been employed in
other areas to good effect. One example is New
Zealand’s ACC, a crown-corporation accidental injury
insurance board. The fund is paid into by everyone in
New Zealand who “works and owns a business,” and
through levies on vehicle licensing payment and car
fuel lling.46 The levies provide a fund that pays out
in cases involving accidental injury. This coverage
applies to everyone in New Zealand, regardless of
age or employment status, and even includes visitors
to New Zealand.47 While there are various incentives
implemented, such as a slight discount on levies for
46 Accidental Compensation Corporation ‘What we do’ (2018)
do/> accessed 1 January 2018.
47 ibid.
companies with excellent workplace injury rates, the
overall structure is a no-fault, community approach
to accidents.
Outside of accident insurance, single-pool or
single-payer insurance has been most visible in
the healthcare sector. The United States is a noted
hold-out against such a system, and spends “more
than twice as much on health care as the average of
other developed nations, all of which boast universal
coverage … [while] more than 41 million Americans
have no health insurance [and] [m]any more are
underinsured.”48 In 2003 experts estimated49 that
converting the United States would “save at least
$200 billion [USD] annually (more than enough
to cover all of the uninsured) by eliminating the
high overhead and prots of the private, investor-
owned insurance industry and reducing spending
for marketing and satellite services.”50 From a purely
logical perspective, this makes sense - an industry
run for prot is intended to make a prot, and must
do so by either over-charging or under-providing. It
is not intended to be a zero-sum game that provides
the greatest amount of care at the lowest cost, it is
intended to create a gap between what is paid by the
insured, and what is paid to the provider. Without
this gap, there is no prot. In addition to this, money
is spent on advertising for the insurance company,
ghting claims both from providers and the insured,
and “avoiding unprotable patients.”
While it is
often argued that a private insurance market allows
individuals to suit coverage to suit their needs, this
inherently provides a problem for “unprotable
Returning to automotive insurance, Canada provides
an interesting comparison as some provinces have
mandated government insurance, whilst others have
not. British Columbia, Manitoba, and Saskatchewan
all have a “one-stop shop” approach to insurance,
but differ in their exact coverage, and Quebec
drivers all have personal injury insured through
the government, while private insurers cover the
Direct cost comparisons are difcult, as the
provinces have different challenges; for example,
more extreme weather in central Canada, and a
48 S Woolhandler and others, ‘Proposal of the Physicians’
Working Group for Single-Payer National Insurance
(1 August 2003) Journal of the American Medical Association
290/6 798
49 ie, before both the roll-out and subsequent roll-in of
50 S Woolhandler and others, ‘Proposal of the Physicians’
Working Group for Single-Payer National Insurance’ (n 48).
51 ibid.
52 ibid.
53 Karen Aho, ‘When the government sells car insurance’
Nasdaq (25 March 2013)
accessed 12 January 2018.
Hard Drive Crash
greater amount of drivers in British Columbia,
Ontario, and Quebec. However, one study compared
the same city - one which straddled a government
insurance and a private insured province - and found
that those with government coverage paid less.54
Additionally, net income from public insurance, at
least in British Columbia, goes into reserves, rather
than exiting the system as a shareholder dividend.55
It has also been suggested that high costs are the
reason for the difference in percentage of uninsured.
For example, in 2002 Ontario (the province with the
highest average insurance rates) had an estimated
10-20% of drivers uninsured, whereas British
Columbia had 0.26% uninsured drivers.56
43 These arguments are not intended to frame the free
market as inherently negative or bad. What this
system aims to accomplish is to set aside, at least
temporarily, the prot of the insurance sector to
pave the way for the AV sector, for the simple reason
that AVs offer more direct societal benets.
2. No-fault
No-fault insurance is not a new concept to the
automotive world. Policy-holders and passengers
are reimbursed for accidents and damage through
their own insurer, rather than tort insurance, where
fault is assigned to a party. No-fault insurance
usually only covers up to a particular sum and
precludes individuals from pursuing the other party
in court. Unfortunately, it does not typically mean
the absence of attribution of fault, rather that the
insurance company or companies will determine
between them which party is at fault, and potentially
increase that party’s future insurance rates.57 Fault
can be attributed by percentage, wherein both
parties may see future increases in rates.58
As previously discussed, the attribution of actual
fault is difcult in scenarios solely involving AVs,
given the difculty of deconstructing the decision-
making process. What the future of AVs require is to
give up the concept of fault in actuality rather than
in name. This is not an easy thing to do - not only are
the rules of the road set up to determine fault, but
humans like blame, and we do not trust intelligences
we don’t understand. This is true even if the non-
54 ibid.
55 Lucy Lazarony, ‘Public vs. private auto insurance’ Bankrate
(22 July 2002)
private-auto-insurance/> accessed 12 January 2018.
56 ibid.
57 ‘No-Fault Insurance: What it Really Means to You’ Insurance
Hotline (11 November 2011)
accessed 1 January 2018.
58 ibid.
human intelligence is demonstrably better at the
task at hand.59 In essence, giving up fault is a leap of
faith: it requires letting go of the idea of an “intuitive
and automatic” desire to conceive of blame.60 It
is, however, a necessary step to improvement -
the move to acknowledgement of an undesirable
consequence rather than the attribution of the
individual entity responsible. There will be a period
of time where fault will still be apportioned, for
example, where accidents have occurred between
AVs and human drivers. This is necessary in order
to allow for incremental integration of AVs, rather
than wholesale substitution. However, AVs will
inherently make fault determination between AV
and non-AVs easier to determine, as AVs can provide
their own surveillance footage.
46 Moreover, as Reed et al. point out, common law is
not unfamiliar with strict liability for inherently
dangerous activities, such as the keeping of
dangerous animals, or ownership and use of
aircraft.61 These difculties have not quashed
either activity but serve to account for the dangers
inherent to them. Strict liability tends to invoke the
opposite conception of no-fault, as it incurs fault no
matter how careful or reasonable the activities of
the individual in question - “the person responsible
is required to indemnify the remainder of society.”
However, the result and acknowledgement are the
same: accidents do occur, and must be accounted for,
no matter the reasonability of the actors in question.
3. Structure
Even given a singular pool insurance provider, there
are many potential iterations of how insurance may
be structured. It is not unreasonable to leave the
consumer with a regular insurance cost that covers
damage - there is no indication that the cost involved
will be higher than a human driver, particularly
59 Berkeley J Dietvorst, Joseph P Simmons, Cade Massey,
‘Algorithm Aversion: People Erroneously Avoid Algorithms
after Seeing Them Err’ (n 13). Rather than debating what
truly constitutes intelligence, non-human intelligence
should be understood here as the computation behind the
AV’s decision making.
60 Janice Nadler, Mary-Hunter McDonnell ‘Moral Character,
Motive, and the Psychology of Blame’ (2012) 97/255 Cornell
Law Review
dir=1&article=3290&context=clr> accessed 11 January 2018
61 Chris Reed, Elizabeth Kennedy, Sara Noguiera Silva
‘Responsibility, Autonomy and Accountability: legal
liability for machine learning’ (Third Annual Symposium
for the Microsoft Cloud Computing Research Centre, 8-9
September 2016)
accessed 11 January 2018, 5.
62 ibid.
Keri Grieman
given that Google cars drove 1.3 million miles in
seven years before causing an accident.63 Reed et
al. suggest that the identiable party to insure is,
pragmatically, the keeper of the vehicle, and that this
allocation follows the precedent of aircraft owners,
where it has invoked no serious problems.64 This has
the additional benet of not disrupting the current
vehicular liability requirements, as vehicles must
already be insured by their owners.65 The amount to
be paid for insurance can initially reect the average
rates for their human driver counterparts, but
should not involve typical factoring characteristics
such as the driving record, where the owner lives,
driving experience, age, gender, or vehicle type. The
aim of the AV is to make these irrelevant, and to
exclude bias when pricing the coverage.66 It should
also provide an initial overhead for damage coverage
as the potential damage-reduction possibilities of
the AV bear fruit.
However, there should be another sector of
contribution to the singular fund - a per-car entry
cost from the manufacturer. While the initial amount
will be arbitrary, what the amount should eventually
reect is injury and related costs compensation for
AV accidents. This amount will require buffering
before a minimum level of saturation for AVs, as if
there is only one AV on the road which causes an
accident worth an accident pay-out of three million
dollars, this is unreasonably punitive. However,
as more and more AVs are put onto the road, the
injury pay-out amounts should be split between
their manufacturers on, for example, a year-to-
year determination basis. This means that when
manufacturer A causes an accident that produces
injury, that cost is split communally amongst all
63 Alex Davies, ‘Google’s Self-Driving Car Caused Its First
Crash’ Get Wired Magazine (29 February 2016)
crash/> 2 January 2018.
64 Chris Reed, Elizabeth Kennedy, Sara Noguiera Silva
‘Responsibility, Autonomy and Accountability: legal liability
for machine learning’ (n 61) 29.
65 ibid, and further noting that “This approach is supported by
the Draft Report with recommendations to the Commission
on Civil Law Rules on Robotics” (2015/2103(INL), European
Parliament Committee on Legal Affairs 31 May 2016) paras
66 These are commonly factored in features of auto insurance
- ‘Compare car insurance quotes to get the lowest rates
in Saskatchewan’ (LowestRates.ca)
insurance/auto/saskatchewan> citing David Marshall,
‘Fair Benets Fairly Delivered: A Review of the Auto
Insurance System in Ontario’ (Ontario Ministry of Finance,
html> accessed 1 January 2018; ‘Compare Auto Insurance
Quotes in Ontario’ (LowestRates.ca)
insurance/auto/ontario> accessed 1 January 2018.
This should not be seen as a shift or allocation of
blame, nor changing the insuring party. It is analogous
to collecting levies from, for example, blank CD sales.
Rather than requiring the time or public resources
to go after individual problematic activity, it is the
acknowledgment that an undesirable result occurs,
and is made possible by the manufacturer. The levy
is neither a punishment, nor an allocation of liability,
but a recognition that the end result is enabled by
the party in question. For the music industry, this
is the assumption that blank CDs are used to enable
industry-undesirable sharing. For AVs, this is the
assumption that no matter how well-designed AVs
are, accidents will, at least initially, occur and cause
injury. In both cases, costs are ultimately borne by
consumers, whether or not the purchase in question
actually enabled an undesirable result. The industry
simply passes costs along to its purchasers. While
damage may be sufciently and reasonably covered
through traditional insurance by AV owners, the levy
serves both a social purpose, of acknowledging the
enabling of these types of accidents, and a monetary
one, through compensation for injury caused. Even if
manufacturers are not held liable, it is benecial that
the consequences of design be acknowledged. While
collecting societies may have acquired a negative
reputation, the levy in itself is not necessarily a
negative way to address this problem - particularly
where manufacturers have both the motivation and
the capability of reducing this amount by decreasing
50 This may seem an arbitrary approach that punishes
manufacturers who produce vehicles that do not
cause injury. However, it incentivizes manufacturers
in societally benecial ways. First, it places injury
reduction as the ultimate cost-saving priority to
manufacturers; specically, they can reduce costs by
placing it at the top of the decision-making process
for the car, rather than avoiding liability. Second,
it promotes co-operation and standardization
between companies. Every manufacturer gains
when they collectively reduce injury costs.
Standardized reactions from AVs not only allow
for predictability for human drivers who share the
road with AVs but foster better interaction between
different manufacturers. It also encourages car to
car communication - rather than building an intra-
company network of communication, manufacturers
are incentivized to communicate cross-brand. The
success of one company is the success of all companies.
Third, it reduces the potential for manufacturers to
hold monopolies over AVs. Requiring an entry cost
to enter the market would mean that a company
must be of a certain worth to even attempt to
compete. When scal giants like technology
companies and traditional auto manufacturers are
involved, this is likely to be an unassailable moat.
Placing the cost per-car means that the success of
then-current market players reduces the potential
Hard Drive Crash
cost per entry for new manufacturers, lowering the
entry to effective competition. Fourth, it means
that companies can fold in the one-time cost per
car into the purchase price, rather than being liable
in perpetuity for an unpredictable cost. Fifth, it does
not remove the benets of branding from individual
companies as car buyers “frequently cite safety as
the most important factor in selecting a car.”
is no reason to believe that this would change and
is in fact likely to be reinforced as drivers hand
control over to an AV. Overall, there should still be
the potential for pursuing a particular company in
extreme cases, such as egregious negligence. For
example, if it can be demonstrated that a company
had knowledge of a dangerous vulnerability and
ignored it - such as a design aw that made any
crash likely to ignite the vehicle - they should bear
the full cost for that oversight. While this may seem
like an unclear standard, the law has dealt with such
standards before, given that tort law is built on the
concept of a “reasonable person”.
51 It is possible that many of these incentives could be
achieved by allowing the insurance fund, or other
parties, to pursue manufacturers for negligence.
Even co-operation could be encouraged by allowing
manufacturers to be pursued as a single entity.
However, this places a greater burden on either
the consumer or insurance entity to undergo the
necessary litigation, or at least legwork, to show
the manufacturer’s negligence. One of the problems
unique to machine learning is that the decision-
making process of articial intelligence can be
particularly opaque - consumers may nd it difcult
if not impossible to understand “black box” decision-
making.68 It may be that the consumer attempts
to recover before having proper knowledge of
whether the AV was in fact negligent. Additionally,
litigation puts further strain on the court system.
Allowing for the levy to provide these incentives -
except in extreme cases - means that there is a strict
liability approach to a no-fault problem, namely, the
acknowledgment of blameless enablement, but the
ultimate injury caused.
4. Implementation:
When allowing manufacturers to side-step strict
liability, it is naturally important to hold high
standards to entry. This is not to say that the
entry requirements should have monetary value,
as previously mentioned, but should include such
areas as rigorous testing. Strict requirements can
67 Ratan Hudda and others, ‘Self Driving Cars’ (n 12) 7.
68 Chris Reed, Elizabeth Kennedy, Sara Noguiera Silva
‘Responsibility, Autonomy and Accountability: legal liability
for machine learning’ (n 61) 13-14.
reasonably be placed on manufactures as the AV
is still a multi-tonne machine that will be piloted
amongst unarmored pedestrians. The possibility
for co-ordination is also a positive one between
manufacturer and government, since co-ordination
such as car to infrastructure, or car to transit,
have the potential to benet both parties. Car to
infrastructure communication, such as trafc lights,
or road closures, have the ability to make the AV more
efcient, and to alleviate strains on infrastructure
such as trafc jams. Even more crucially, requiring
predictable procedures for emergency vehicles could
result in reduced emergency response times, as AVs
part like the Red Sea as required.
Car to transit communication can not only help avoid
collision, but also allows for better co-ordination in
timing, particularly when AVs are used to ll a gap
in transportation rather than replace an individually
owned vehicle. Implementation should also allow
communication between the government insurer
and manufacturers - where damage is tracked to a
particular problem, the government entity has the
ability to convey this to the manufacturer, and the
power to demand a solution. It is unlikely to reach
this level, as manufacturers are incentivized to
better their vehicles regardless, but it nonetheless
allows for a two-factor system of tracking issues with
the AVs.
A further requirement could also be standardized
signaling to third parties. One particularly prescient
analysis notes that while AVs are technically more
predictable than their human counterparts, this
does not mean that they are more predictable to
third parties – i.e., those who have not programmed
Pedestrians have indications as to whether a
human driver has noticed them. This can include eye
contact, a hand-wave, or, in extreme cases, a rude
gesture. This sort of communication has not yet been
indicated by AV manufacturers, but could grow to be
included in the “price to entry” in order to qualify to
enter the market. This could be as simple as unique
blinking indicators in the pedestrian’s direction, or
as complex as screens on various parts of the car,
but overall serves to show that there needs to be
a consistent dialogue between the regulator and
F. Public Policy Part Two
What a public policy approach to AVs aims to
achieve is incentivizing better questions. Rather
than demanding manufacturers wait on the answer
to “how liable will the company be?”, a proactive
69 Harry Surden, Mary-Anne Williams, ‘Technological Opacity,
Predictability, and Self-Driving Cars’ (n 16).
Keri Grieman
public policy approach, such as the one described,
forces companies to instead ask “what is the best
possible way to reduce injury?”.
Many billions have been put into researching
and developing fully autonomous vehicles, not to
mention the many stages of partial-autonomy along
the way. The industry growth rate is currently 16%
and is expected to be over $1 trillion by 2025.70 One
policy benet of the proposal thus far discussed
is that rather than stockpiling capital against the
eventuality of a lawsuit, companies can focus on
putting funds towards other areas such as increasing
fuel efciency, reducing vehicle cost, and improving
accessibility. This has the potential for positive
economic impact since research and development
is encouraged, rather than stied or put on hold
to wait for potential legal impacts. While there
is still an indeterminate amount of time to wait
before manufacturers are ready to put consumer-
model AVs on the road, the reluctance to assume
responsibility is palpable as all consumer available
automated features require that there be a licensed
human in the driver seat in order to take control
the instant it becomes necessary - and preferably
even before.
57 This paper’s proposal encourages the introduction
of AVs, while interfering minimally with the current
regime of road rules and liability. It does not require
the scrapping of an entirely workable system,
and simultaneously allows for the incremental
introduction of AVs on the road with a majority of
human drivers. While current automated features on
cars do still require a human driver, it is unnecessary
to allow for a change in liability where the human
driver must still be able to step in.
A Public Prosecution Service of Canada working
group has produced a report on the Future of
Automated Vehicles in Canada.
While the report
is naturally focused on implementation of semi and
full AVs in Canada specically, it provides a helpful
list on “The Role of Governments”:
Regulate vehicle safety;
Harmonize standards [between countries];
Encourage innovation;
Protect privacy of individual vehicle users;
Educate the public;
70 Muhammad Amat, Dr Clemens Schumayer, ‘Self Driving
Cars: Future has already begun’ (n 6) 18.
71 Public Prosecution Services of Canada ‘The Future
of Automated Vehicles in Canada’ (29 January 2018)
23 June 2018.
Build data expertise and capacity;
Develop and enforce trafc laws;
Oversee insurance and liability;
Ensure a safe and smooth transition;
Build and upgrade transportation
While many of these areas have been discussed in this
paper, it is a helpful reminder that a government’s
role is not simply to mandate legal change from
a removed perspective, but to aid transition in a
variety of areas and elements. Insurance and liability
are naturally important, but if laws are not enforced,
or the public remain unconvinced, then the potential
benets will not be realized in full.
60 Public policy is an important tool to achieve social
acceptance. Transparency and clarity of legislation
will be key to sufcient initial condence in
consumers to start building positive interaction
– personal experience being the ultimate key to
social acceptance, both by the individual themselves
and word of mouth. If the policy is to achieve the
aforementioned benets of AVs, it must have
the public on board. Changes inherently bring
opposition, but this has not stopped legislating
in favour of change in the past; for example, high
occupancy vehicle lanes encourage car-sharing,
tax incentives for electric and hybrid vehicles
incentivize greener purchases, and seat-belt and
airbags have forced societal change directly.73
Testing and safety are priority concerns. Social
acceptance will never be achieved unless there is
a belief in the safety of AVs. Consumers have good
reasons to be skeptical of the automotive industry,
and safety records in particular, especially given
the Ford Pinto’s transmission problems, Firestone
tire blowouts, the Takata airbag recall, and the
Volkswagen emission scandal, which all suggest
that prot may have been prioritized over safety.
AVs cannot afford this type of prot post-mortem.
Testing must be particularly stringent, and indeed
better than the average driver to overcome the
concerns over non-human drivers. The adoption
of the aforementioned levy approach is benecial
as consumers could not only avoid liability, but it
would ensure that companies are serious enough
about the vehicle’s safety capacity passengers to
“put their money where their mouth is” in terms
of human safety.
72 ibid 14.
73 Ratan Hudda and others, ‘Self Driving Cars’ (n 12).
Hard Drive Crash
Two potential ways to foster social acceptance
are publicizing existing uses and creating pilot
programs. The public already interacts with AI
transportation, such as Masdar and Heathrow
airport shuttles, the Milan driverless metro, and
driverless trucks in Australian and Chilean mines.74
A simple step is to make the public more aware that
these transportation methods are already in use,
safely, efciently, and successfully. Pilot programs
to provide AVs to impoverished communities or
those underserved by current transit initiatives
can be a way to allow for optional adoption and
demonstrable benet, though particular care should
be taken to show that this is not a testing ground.
Initiatives for the visually impaired, for example,
would demonstrate that unlike AI levels below ve,
fully autonomous AVs make car travel accessible
to all. Both publicization and pilot programs have
signicant potential in terms of building positive
personal experiences, promoting both personal
acceptance and word of mouth recognition.
Social acceptance of AVs through public policy
methods faces unique challenges. Seatbelt adoption,
for example, used a variety of methods in the United
States: policies and mandates such as laws regarding
use; incentives and rewards based on use; signs
politely reminding seatbelt use; and feedback on
community performance.75 These methods are not
easily transportable to AV adoption. While laws
regarding use are naturally important in terms
of regulation, AVs present unique challenges; for
example, although wearing a seatbelt or not is a
distinct choice, it is still possible to drive without
one. If one is in an AV, the choice is not whether
or not to drive, since by the time an individual has
made the choice to use an AV, they have accepted
the overarching function of the AV, rather than
deciding whether or not to wear a seatbelt while
still using the car in a way they are familiar. The
role of public policy in the case of AVs is to remove
uncertainties which might disincentivize use, rather
than attempt to force a particular choice. Public
policy should not be focused on forcing the adoption
of AVs, but on removing the barriers to those in the
position to adopt their use, such as uncertainties
like liability. No car owner wants to be unsure of
whether or not they will be liable for an accident
over which they had no control, even if they were
74 Alain L Kornhauser, Smart Driving Cars: Where Are We
Going? Why Are We Going? Where Are We Now? What Is In It
for Whom? How Might We Get There? Where Might We End
Up? (2013) (Princeton University TransAction Conference,
18 April 2013)
SmartDrivingCars_041113.pdf> accessed 22 June 2018 24.
75 E Scott Geller, Timothy D Ludwig, A conceptual
framework for developing and evaluating behavior
change interventions for injury control (Health Education
Research, 1990) DOI: 10.1093/her/5.2.125.
aware that the probability of an accident occurring
was much lower.
G. Challenges – Legal and Technical
There are many challenges to be faced in introducing
AVs. There are uncountable minor changes that
must be introduced - everything from regulations
requiring hands on the wheel, to how vehicles are
fueled. There are much more impactful challenges
to be faced, however. Manufacturers must be
discouraged from attempting to allow their car to
game the system and offering consumers a vehicle
that disadvantages either other AVs or human
65 Where AVs can communicate between themselves
and infrastructure, the ability of third parties to
hack the system for their own potential malicious
ends is a concern, particularly in a nexus with
personal privacy. Personal privacy has already
become a crucial battle in the 21st century, and AVs
will accelerate the race between laws protecting
privacy of data, and companies using that data for
their own means. AV data can not only identify a
person and their current whereabouts, but likely a
great deal of telling information about their habits,
friends, and lifestyle. Beyond hacking, connections
between vehicles and with infrastructure and the
manufacturer could still be used to collect and
transmit personal data. Unless forced to do so,
manufacturers are unlikely to allow consumers
to opt out of data transmission since a great deal
of the data will likely be used for positive means,
such as optimizing function and driving patterns.
However, there is still the danger that information
released could identify an individual. Collection
has signicant benets, and the problem must be
addressed by controlling use and disclosure. This
is done through data protection law. The question
remains whether existing data protection law is
sufcient. While some jurisdictions have unied their
approach to data protection, such as the European
Union’s General Data Protection Regulation, there
is no global unity on issues such as what constitutes
personal data; who can use what, and how; what
protection should be in place; or how to properly
anonymize that data. Common data protection issues
and proffered solutions can be seen in other areas
such as medical data; data is crucial for research,
but there is a signicant threat to privacy if data is
insufciently anonymized or used in ways that were
not foreseen at collection. Addressing such issues for
AVs might follow practices similar to medical data
collection or may be found to require a customized
regime that can be updated faster than traditional
data protection law.
Keri Grieman
While the system suggested should, on the whole,
be able to integrate with current systems, there
may be unforeseen challenges. For example, it has
been suggested that both the Geneva and Vienna
Conventions may not allow for a vehicle that does
not permit a human driver to resume control.76
Individual jurisdictions, not to mention countries,
may have legislation or precedents that negatively
impact, or currently do not allow for, the integration
of AVs.
Functionally, AVs still have hurdles to overcome.
They are expensive, perhaps prohibitively so as, the
extra equipment that allows the AV to drive itself
are not cheap, and their cost is in addition to the
vehicle itself. Extensive testing is also expensive and
is a cost that is likely to remain. AVs still struggle
with weather, and while testing is being carried
out to overcome this,77 accidents have occurred
on the basis of weather conditions.
the lack of opacity is a barrier to trust. While AVs
have much to offer, it is a legitimate complaint
that the “decisions” made by AVs can be difcult
to understand, particularly from a lay-person’s
perspective. This lack of clarity can carry through
to lawsuits and will challenge the technical expertise
of those who may be ill-equipped to evaluate such
Even with the suggested changes, there are will
be systemic issues to be addressed. While co-
operation between companies in terms of life-saving
measures, predictability, and integration is positive,
it inherently raises concerns about competition
and collusion. Companies may be motivated to,
for example, nd a system that works well enough
between them and keep to it, rather than striving
for better, safer, or more efcient advancements.
H. Conclusion
69 There is a world of potential to be unlocked by AVs.
On a purely ethical basis, it would be very difcult to
ignore their lifesaving potential. Beyond this, there
are countless other, if lesser, benets. A car is an
expensive investment that sits unused an estimated
95% of its life.79 Currently, 40% of fuel is used nding
76 Jonathan Margolis, ‘Self-driving cars still face multiple
roadblocks’ Financial Times (New York, 11 January 2017)
680c49b4b4c0> accessed 2 January 2018.
77 snow - Ford Media, ‘Ford’s Industry rst autonomous
vehicle tests in snow’ (YouTube, 11 January 2016)
youtube.com/watch?v=vShi-xx6ze8> accessed 2 January
78 Neal Boudette, ‘Tesla’s Self-Driving System Cleared in
Deadly Crash’ (n 7).
79 Muhammad Amat, Dr Clemens Schumayer, ‘Self Driving
a parking space in urban areas.80 Time, energy,
and stress are expended on commutes that could
be spent in better, or at least more relaxing, ways.
Even better use of land is a possibility, as concepts
such as a “park and ride” for airports need no longer
take up space.81
AVs have the potential to remove every human
failing from the province of transportation. This
has an impact beyond human choices, such as
driving while intoxicated or tired. Vehicles can see
further than human eyes and communicate on many
more levels. A car that needs no human driver can
avoid a traditional vehicle’s security liabilities -
with no need for human eyes, there is no need for
a vulnerable glass portal at the front of the car. AVs
have the potential to become metaphorical tanks,
as they need not account for a driver’s ability to see
from various angles.
Current liability conceptions are deeply problematic
for AVs. Not only are they uncertain in terms of
introducing AVs, but the current jurisprudence
provides no promising answers as to where liability
may fall. Pinning liability on parties who have no
control, or on parties who will make it a primary
priority over more important concerns, is likely have
the effect of chilling the market before it can really
begin. Ignoring liability questions and assuming
that the market will develop and ourish when left
alone is optimistic at best, and at worst enables a
monopolistic and limited-benets system.
It is important that public policies regarding AVs
are scalable. It needs to be capable of addressing
a slow trickle of AVs as they enter the market,
and an increasing majority as they become more
affordable and marketable. The regime needs to
ensure that victims are not left out in the cold, and
manufacturers not incentivized to prioritize scal
vulnerability ahead of human safety.
It is crucial that we incentivize better questions -
how to achieve a perfect no-injury record, rather
than where liability should fall on a scale of
priorities. How to improve access for individuals
with mobility issues, rather than how to inch forward
without invoking liability. Regulation should aim to
encourage one particular future: where accidents
are unusual, and vehicular deaths non-existent. But
this needs to start somewhere and needs law reform
action to put the wheels in motion.
Ultimately, liability conceptions need to evolve in
order to fully realize the potential benets of AVs
on a societal level. This is best achieved by letting
Cars: Future has already begun’ (n 6) 11.
80 ibid.
81 ibid 18.
Hard Drive Crash
go of traditional liability conceptions and blame.
There needs to be strict liability as damage needs
to be reimbursed, and no-one should face nancial
hardship for decisions beyond their control. This
strict liability needs to be placed without fault.
Attempting to place fault and blame results in
inevitable time, money, and litigation spent, when
such energies are better focused on remedying the
problem, compensating the victim, and improving
the AVs. It also sidesteps the problem of incentivizing
avoidance of liability rather than the prevention of
harm. Compulsory insurance is already required in
most if not all countries currently developing AVs,
and this insurance setup can and should be expanded
to cover AV accidents. Doing so from a single pool
allows for streamlined claims, a direct dialogue
between claim evaluators and manufacturers, and co-
operation regarding AV issues. Such a system could
be realized through an independent government
entity and augmented by a manufacturer levy.
75 “May you live in exciting times” is often cited to be
a curse. Yet these are indeed exciting times – we
are at a crossroads of design, manufacturing, and
vision. We have the unique opportunity to foresee
innovation, and to level the eld in preparation of its
arrival. We have a distinct moment to celebrate one
of humanity’s greatest qualities, the prerequisite of
all innovation: drive. Let’s put the pedal to the metal.

To continue reading

Request your trial