ArticlePDF Available

Autonomous Vehicles and Embedded Artificial Intelligence: The Challenges of Framing Machine Driving Decisions

Authors:

Abstract

With the advent of autonomous vehicles society will need to confront a new set of risks which, for the first time, includes the ability of socially embedded forms of artificial intelligence to make complex risk mitigation decisions: decisions that will ultimately engender tangible life and death consequences. Since AI decisionality is inherently different to human decision-making processes, questions are therefore raised regarding how AI weighs decisions, how we are to mediate these decisions, and what such decisions mean in relation to others. Therefore, society, policy, and end-users, need to fully understand such differences. While AI decisions can be contextualised to specific meanings, significant challenges remain in terms of the technology of AI decisionality, the conceptualisation of AI decisions, and the extent to which various actors understand them. This is particularly acute in terms of analysing the benefits and risks of AI decisions. Due to the potential safety benefits, autonomous vehicles are often presented as significant risk mitigation technologies. There is also a need to understand the potential new risks which autonomous vehicle driving decisions may present. Such new risks are framed as decisional limitations in that artificial driving intelligence will lack certain decisional capacities. This is most evident in the inability to annotate and categorise the driving environment in terms of human values and moral understanding. In both cases there is a need to scrutinise how autonomous vehicle decisional capacity is conceptually framed and how this, in turn, impacts a wider grasp of the technology in terms of risks and benefits. This paper interrogates the significant shortcomings in the current framing of the debate, both in terms of safety discussions and in consideration of AI as a moral actor, and offers a number of ways forward.
Full Terms & Conditions of access and use can be found at
https://www.tandfonline.com/action/journalInformation?journalCode=uaai20
Applied Artificial Intelligence
An International Journal
ISSN: 0883-9514 (Print) 1087-6545 (Online) Journal homepage: https://www.tandfonline.com/loi/uaai20
Autonomous Vehicles and Embedded Artificial
Intelligence: The Challenges of Framing Machine
Driving Decisions
Martin Cunneen, Martin Mullins & Finbarr Murphy
To cite this article: Martin Cunneen, Martin Mullins & Finbarr Murphy (2019): Autonomous
Vehicles and Embedded Artificial Intelligence: The Challenges of Framing Machine Driving
Decisions, Applied Artificial Intelligence, DOI: 10.1080/08839514.2019.1600301
To link to this article: https://doi.org/10.1080/08839514.2019.1600301
© 2019 The Author(s). Published with
license by Taylor & Francis Group, LLC.
Published online: 13 May 2019.
Submit your article to this journal
View Crossmark data
Autonomous Vehicles and Embedded Artificial Intelligence:
The Challenges of Framing Machine Driving Decisions
Martin Cunneen, Martin Mullins, and Finbarr Murphy
University of Limerick
ABSTRACT
With the advent of autonomous vehicles society will need to
confront a new set of risks which, for the first time, includes
the ability of socially embedded forms of artificial intelligence
to make complex risk mitigation decisions: decisions that will
ultimately engender tangible life and death consequences.
Since AI decisionality is inherently different to human deci-
sion-making processes, questions are therefore raised regard-
ing how AI weighs decisions, how we are to mediate these
decisions, and what such decisions mean in relation to others.
Therefore, society, policy, and end-users, need to fully under-
stand such differences. While AI decisions can be contextua-
lised to specific meanings, significant challenges remain in
terms of the technology of AI decisionality, the conceptuali-
sation of AI decisions, and the extent to which various actors
understand them. This is particularly acute in terms of analys-
ing the benefits and risks of AI decisions. Due to the potential
safety benefits, autonomous vehicles are often presented as
significant risk mitigation technologies. There is also a need
to understand the potential new risks which autonomous
vehicle driving decisions may present. Such new risks are
framed as decisional limitations in that artificial driving intel-
ligence will lack certain decisional capacities. This is most
evident in the inability to annotate and categorise the driving
environment in terms of human values and moral under-
standing. In both cases there is a need to scrutinise how
autonomous vehicle decisional capacity is conceptually
framed and how this, in turn, impacts a wider grasp of the
technology in terms of risks and benefits. This paper inter-
rogates the significant shortcomings in the current framing of
the debate, both in terms of safety discussions and in con-
sideration of AI as a moral actor, and offers a number of ways
forward.
The self-driving car raises more possibilities and more questions than perhaps any other
transportation innovationself-driving cars have become the archetype of our future
transportation. Still, important concerns emerge. Will they fully replace the human
driver? What ethical judgments will they be called upon to make? What socioeconomic
impacts flow from such a dramatic change? Will they disrupt the nature of privacy and
security?
CONTACT Martin Cunneen martin.cunneen@ul.ie
APPLIED ARTIFICIAL INTELLIGENCE
https://doi.org/10.1080/08839514.2019.1600301
© 2019 The Author(s). Published with license by Taylor & Francis Group, LLC.
This is an Open Access article distributed under the terms of the Creative Commons Attribution-NonCommercial-NoDerivatives
License (http://creativecommons.org/licenses/by-nc-nd/4.0/), which permits non-commercial re-use, distribution, and reproduction
in any medium, provided the original work is properly cited, and is not altered, transformed, or built upon in any way.
(NHTSA, 2016)
Part One: Introduction
Artificial Driving Intelligence: Context to the Significance of Autonomous
Vehicle Decisions
Autonomous vehicles (AV) offer the opportunity to harness the benefits of
the latest sensory technologies and artificial intelligence (AI) to make driving
decisions which mitigate many risks associated with human driving deci-
sions. Indeed, the focus on the AI driving of AV gives rise to two contrasting
formulations of the decisional benefits and risks of the technology which
epitomise general disorientation regarding the issues of machine decisionality
and impacts in terms of benefits and risks to society as a whole. The
combination of sensory and intelligence technologies provides
a topographical representation of the road phenomenon which support AI
to make more immediate and accurate driving decisions. As machines, AV
also eliminate decisionality problems associated with the human frailties of
fatigue, misperception, and intoxication, along with the problematic deci-
sions humans often make in the context of driving. This rendering of the
technological benefits of AV constitutes a safety argument that not only
identifies the welfare benefits of machine decisions but also endorses claims
that AV should be supported by policy. Conversely, the alternative perspec-
tive highlights these potential new risks as decision errors and limitations to
the driving AI. As such, there is a clear need to define and disseminate the
benefits of AV decisional intelligence in order to avoid underutilisation of the
technology due misplaced risk perception (Floridi, 2018).
Governance and Framing of AI Decisions
The roll-out of an emerging technology presents numerous challenges for
governance regimes, while the transitioning of such technologies from nano-
material to GMO entails the assessment and understanding of the benefits and
risks. It is therefore crucial to anticipate possible risks. AV are an topical
example of a socially embedded and potentially ubiquitous AI technology.
Here, as with other technologies, those charged with risk governance regimes
face the dichotomy of empirical objectivity and moral dilemma. While there
are clearly persuasive utilitarian AV safety arguments predicated on the reduc-
tion of deaths and injuries, other moral questions emerge which are more
problematic and emphasise the need to exercise caution in introducing AV.
Such debates address the fundamental premise of the desirability of machine
decisionality over matters of human life and death decisions and mainly obtain
to instances whereby automated vehicles confront moral choices in the course
2M. CUNNEEN ET AL.
of a journey. At one end of the spectrum, AV choices could optimise route-
planning algorithms to help avoid schools or known problematic areas, while
at the other end, such choices could influence possible road traffic accident
(RTA) decisions. Such split-second pre-collision decision windows are thus
contextualised in scenarios of unavoidable traffic accidents which may result in
death and injury. This paper therefore reflects on the balance between risk and
utility inherent in moral questions. As such, it considers both sets of arguments
and advances the case for more focus and precision in the conceptual framing
of such debates. While this paper acknowledges that AVs are likely to provide
a safer form of driving than human drivers in the long-term, it nonetheless
interrogates the shortcomings of both empirically-based safety arguments and
the various ethical nuances. To better understand the full compass of AI
decisionality, it is necessary to weigh claims that AV can mitigate human
driving risks by making safer driving decisions against claims that all AI
decisionality is circumscribed by an inherent absence of moral agency. In
both instances, a clearer elucidation of the decisional capabilities of AI in the
context of AV would offer greater clarity.
Autonomous Vehicle Literature Space
Autonomous vehicles offer many significant societal benefits: from enhancing
the mobility of those who lack it, transforming urban spaces and supporting
the environment, to radically improving safety and saving lives. However, since
the opportunities of any substantive technology also carry both embedded and
new forms of risk, any actualisation of potential AV benefits also necessitates
mitigation of the risks involved. Moreover, AV risk mitigation cannot be
undertaken by state governance regimes alone, but must rather be a multi-
stakeholder phenomenon. In this instance, traditional state government and
new governance models are simply outpaced, as is evident throughout the
current era of digital innovation (Marchant, 2011), and highlighted by the
NHTSA (2017); the speed with which HAVs are advancing, combined with the
complexity and novelty of these innovations, threatens to outpace the Agencys
conventional regulatory processes and capabilities.For these reasons, intelli-
gence technologies can only be responded to by a shared risk mitigation
process wherein numerous actors cooperate. As such, the conceptualisation
and framing of technology in terms of meaning, benefits, and risks, will
ultimately determine how stakeholders engage with the technology.
The key consideration of AV risk mitigation discussed across the literature
concerns assessment of the AV capacity to make driving decisions. As such,
any research which further illuminates the decisionality phenomenon of AV
both contributes to the multi-stakeholder risk mitigation process and promotes
access to AV societal benefits. Moreover, analysis of the scope of AV decisions
in terms of both benefits (risk mitigation) and potential limitations (new forms
APPLIED ARTIFICIAL INTELLIGENCE 3
of risk) supports the dynamics of new governance relations which consist of
both top-down and bottom-up movement. It is of further interest to consider
that while AVs arguably afford opportunities to minimise and potentially
eliminate the many risks associated with human driving, future benefits cannot
be realised unless accurate and effective anticipatory risk governance research
is undertaken today. A broad and immensely complex decisional context is
inherent to AV technologies, such as, investigating how different governance
actors and policy-writers understand the decisional capacity and societal
impact of AV decisions (Allenby, 2011; Anderson et al., 2014;Lin,2013). It
also concerns diverse ethical interpretations of AV decisions, including those
who identify the need to control ethical decisions as a predetermined config-
uration of action (Gogoll and Müller, 2016) or calculating metrics such as
values and risk (Bonnefon et al., 2015;Goodall2016;Young2016). In addition,
considerable extant research rehearses the many abstract questions surround-
ing machine ethics (Malle, 2013; Lin, 2016;Doganet al.,2015;Gerdes&
Thornton, 2016), while others consider meaning, conceptual confusions, and
limited decisional capacity (SAE, 2016;Nyholm&Smids,2016). Issues of the
technical scalability of AV decisional capacity are also of significance
(Fraichard, 2014), along with a layer of legality and governance which is
heavily reliant on fulsome understanding and anticipation of the impacts of
AV, particularly in terms of societal risk (Loh and Loh, 2017; Kasperson et al.,
2009; Hevelke & Rümelin, 2015).
This journal has previously highlighted such issues. For instance, Trappl
(2016) underscores the need to consider the important conceptual differ-
ences between human and machine contexts of moral decisionality in the
context of AV, while Bringsjord and Sen (2016) highlight the potential
confusion surrounding the differing contexts of intelligence and ethical
capacities of AV. They also point out the vital need to support actors in
reaching more accurate and informed choices in terms of AV policy and
regulation. Millar (2016) proposes the need to investigate ethical decision-
making in AV, while Goodall (2016) shifts the emphasis from ethics to
risk. Others, such as Coeklebergh (2016), attempt to elucidate the impor-
tance of changes in relations which socially embedded technologies bring
about between agents and actors (Coeklebergh, 2016). This is most evident
in the consideration of key legal and ethical concepts by way of the
changing human phenomenological relations regarding AV
(Coeckelbergh, 2016). However distinct these approaches, they are united
in their attempts to fathom the decisional relations of AI and applied
applications such as AV. This paper therefore seeks to interrogate how
AI and AV as decision-making technologies are conceptually framed and
how such framings determine the engagement and understanding of
diverse agents (Burgess and Chilvers, 2006; Coates and Coates, 2016). It
is our contention that this causal, societal chain of understanding AV
4M. CUNNEEN ET AL.
decisions rests upon accurate conceptual (re)framing of the technology as
it continues to emerge and evolve. Understanding the role of AV decision-
ality is itself a complex challenge which requires careful elucidation
(Choudury, 2007; Young, 2016), while the basic function of AV requires
the driving intelligence to make decisions affecting human welfare and life
(Lin, 2016). In fact, AV will typically make thousands of such decisions on
every trip, and global deployment will translate into millions of such
decisions per day. Accordingly, it is imperative to explore the many facets
of the AV decisional spectrum, not merely on terms of awareness of the
limitations of AV decisionality, but also with a knowledge of the key
contexts wherein different actors confuse or misunderstand the meaning
of AV decisions.
Aims
The paper holds the AV decisional phenomenon to consist of two conceptual
frameworks regarding decision capacity and risk mitigation. The term conceptual
framework refers to the way AV and AI technologies are currently presented in
terms of decisional capacity and risk mitigation. Thus, the decisional phenomenon
of the technologies is delineated in two forms which contrast to human driving
decisions. In the first instance, AI in the context of driving is held to offer superior
decisions which mitigate risks; in the second, the phenomenon of AI as a driving
intelligence presents decisional limitations which present new risks which reside
in AI capacity to make moral determinations. While many pressing questions
arise in the literature regarding how to best investigate and anticipate the societal
impact of emerging technologies in terms of risk (Asveld and Roeser, 2009),
conspicuously little attention is given to conceptual analyses, particularly in
terms of AV conceptualisation frameworks. Asveld and Roeser (2009), for exam-
ple, emphasise not only a deficiency in investigating technological risk and
morality, but also an apparent gap in applying analytical methods to the emerging
technological risk phenomenon. This is most apparent when one considers the
need for enhanced transparency and explainability necessary to enable the diverse
arrayofstakeholderstocontributetoissuespertainingtothesocietaluseofAI
technologies.Thisisalsocrucialinterms of the many actors required to make
important decisions regarding the technology. There is a need to further elucidate
the conceptual frameworks, not as two distinct conceptual framings, but as one
hybrid framework which ringfences a decisional capacity consisting of beneficial
risk mitigation measures but which also presenting new risks. Overall, there is
a need to consider the intersection of the two main framing moments: that of the
safety argument, which largely stresses the upside in term of risk management;
and ethical critiques which underscore the downside in challenging AI moral
agency and capacity to make life and death decisions.
APPLIED ARTIFICIAL INTELLIGENCE 5
Part Two: Conceptual Framing: The Challenges of AI Decisions
Conceptual framing is a process of using established concepts and their meanings
to construct a framework around phenomenon. The construction and use of
concepts creates a model of conceptual relations which bring meaning to the
phenomenon. In terms of emerging technologies, the way in which the new
phenomenon is conceptually framed has significant impact on how others engage
with and understand the technology. This is particularly important in terms of
how different actors understand technological relations, to society, and to actors
who need to make decisions regarding societal engagement. As such, conceptual
frameworks are key to how we understand the benefits and risks which the new
technological phenomenon presents. A conceptual framework aims to configure
the relations in the technological phenomenon, but this is not always
a straightforward process. The conceptual framework is a way of modelling and
communicating what a technology means during a phase of technological emer-
gence. A seminal example of how a seemingly straightforward and intuitive
instance of conceptual framing led to distortion and difficulties around precise
meaning, resides with the concept of artificial intelligence. In his Dartmouth
proposal of 1955, John McCarthy introduced artificial intelligence (AI) as the
inaugural, conceptual framing of machines which carry out intelligent tasks
(McCarthy 1955, 1956). Since then, the conceptualisation of AI has posed sig-
nificant difficulties (Searle 1980;Kaplan,2016;Dreyfus,1965,1972;Dreyfus,1986)
in terms of how different agents and actors engage with its conceptualisation
(Johnson & Verdicchio, 2017). Such challenges have important ramifications with
respect to the anticipatory research and anticipatory governance of the potential
impacts of emerging technologies (Donk, 2012;GasserandAlmeida,2017).
Conceptual frameworks are utilised in most areas of research that rely on clear
models of relational meaning pertaining to complex entities and values; from
finance (Berger and Udell, 2006), and education and models of pedagogy (Savage
and Sterry, 1990), to societal impacts of technology and risk (Bachman et al., 2015;
Kasperson et al. 1988). An important commonality throughout the use of con-
ceptual frameworks concerns the attempt to control meaning; whether to improve
it, to supplant established frameworks,ortocreatenewones.Thus,frameworks
concern the ability to conceptually construct meaning for the purposes of knowl-
edge and dissemination. Conceptual frameworks are often rehearsed as a means of
investigating and reinforcing particular models obtaining to understanding and
communicating the meaning of technological phenomenon. The utility of con-
ceptual frameworks as an elucidatory exercise has proven beneficial in modelling
and communicating new technologies, meaning, and societal relations (McGrath
and Hollingshead, 1994). Two key qualities are intrinsic to effective and accurate
conceptual framing: first, the ability to clearly present the relationship between
entities or phenomena which reside together; and second, the ability to commu-
nicate these relationships to stakeholders. As conceptual framing is an essential
6M. CUNNEEN ET AL.
piece of the cognitive jigsaw and an important mechanism to convey both mean-
ing and understanding of phenomenon (Sorokin, 2016), it is intrinsic to under-
standing and communicating innovative technology (Maxwell 2013,41).Atthe
same time, it can exert a normative impact on debates within society. Accurately
anticipating the societal, ethical, and legal (SEL) impacts of emerging technologies
is a process of investigation that is, in and of itself, contingent upon the initial
conceptual framing of the technologies (Donk et al. 2011). Conceptual frame-
works can contribute to a more informed context of conceptual meaning which
determines other downstream frameworks which, in turn, determine conceptual
accuracy, clarity, and transparency, such as governance frameworks (Renn 2008).
The conceptual framing often begins as an ad hoc process with little consideration
of the accuracy of the concepts used, and which in fact, are sometimes are
arrogated from other domains (Cook and Bakker 2012).
Framing Artificial Intelligence and Autonomous Decisions
The new and emerging technological paradigm of AV has generated some
technological disorientation, more specifically in respect of the decisional capacity
of embodied AI products. There is a progression of conceptual meaning and
conceptual framing that begins with the research phase and culminates with how
the media and society engage with the concepts relating to the technology.
However, given that development of innovation depends upon the key metrics
of governance, the media, and public perception, there is a need for closer scrutiny
of how initial framing plays out in the public arena. The literature on risk
amplification speaks to this issue and points to the need for debates which set
a positive and inclusive tone (Pidgeon et al., 2003). This is true both of the more
general phenomena of risk amplification, as well as more discrete phenomena,
such as dread risk. Risk amplification and fear of new and emerging technology is
well-documented in the literature and suggests the care needed around initial
conceptual framing (Frewer et al., 2002). This aspect is taken up by Johnson &
Verdicchio (2017) who maintain the need to argue for a reframing of AI discourse
that avoids the pitfalls of confusion about autonomy and instead frames AI research
as what it is: the design of computational artefacts that are able to achieve a goal
without having their course of action fully specified by a human programmer
(Johnson & Verdicchio, 2017). While their critical approach bears on the chal-
lenges of framing embodied AI products such as AVs, they represent a minority
who address the question. We contend that in addition to autonomy there are
further related complex challenges specific to the framing of AI and AI decision-
ality. In relation to how effective ontological domains are set out between concepts
(Franklin & Ferkin, 2006), there is a similar need to anticipate conceptual chal-
lenges in the initial framing and ontologies of concepts used in framing embodied
AI products (Cunneen et al. 2018). This is essentially a call for temporal con-
siderations to be captured in the concepts employed, as since this field is highly
APPLIED ARTIFICIAL INTELLIGENCE 7
dynamic, and in terms of the configuration of actors and their anticipated roles,
these are liable to change over time.
The safety argument and the ethical challenge both relate to anticipating the
SEL impacts of AVs regarding decisional capacity. A critical analysis of both
examples suggests that the disparity arisesfromafailuretoengageatthenecessary
meta-level or construct informed accurate conceptual frameworks of AV decisio-
nal capacity, and failure to consider the important differences between how society
and users understand human and machine decision-making in more detail. In
fact, the core question of the SEL impact of AVs is yoked to the meaning frame-
work of machine driving decisions and human driving decisions. This underlines
the necessity to interrogate the conceptual framing of AV driving decisions.
Without accurate SEL impact analysis, the challenges of uncertainty and risk
will hinder informed research, development, and societal perception (Renn
2008: xv). And without accurate metrics of the SEL impact, systems of governance
cannot provide the mechanisms which balance the need to support innovation
with the duty to assess potential risks and protect society from harms. This is
particularly emphatic in cases where innovation is maintained to be ethically
justifiable. In short, all innovation warrants a process of analysis by which to
accurately frame the legal and general principles of associated societal rights to
safety, freedom, equality, privacy, and welfare. While both the safety argument and
the ethical challenge are in agreement in framing AVs to centre on the decisional
capacity of vehicular driving intelligence, they offer very different matrices of the
range of decisions AI must carry out to safely traverse the human road network.
Each interpretation begins with the focus on decisions but frames the decision
capacity differently, and each anticipates very different accounts of the potential
SEL impacts of AV decisions and governance. Diverse perspectives and inter-
pretations are an integral aspect of developing research and knowledge contexts,
but as multiple agents and actors engage with the different frameworks, the
potential for inaccurate framing feeding into systems of governance is
a significant concern. We have two very different accounts of decisional capacity
regarding the anticipation of SEL impacts and governance of AVs. Each one
frames the decisional capacity in dramatically opposing ways: one claims it is
a superior driving decision capacity that will save lives; the other insist it presents
a risk of limited decisional capacity which could inadvertently pose significant
ethical problems (Lin, 2016).
1
Proper analysis clarifies the AV decision domain,
and if we are to judge by the two principal framing values of the safety argument
and ethical challenge, the AV decisional framework presents a technological
medium that remains conceptually obscure.
8M. CUNNEEN ET AL.
Part Three: The Safety Argument
Framing the Space regarding the Societal Benefits of Autonomous Driving
Decisions
The first conceptual framing of AV, referred to as the safety argument,
expounds on the decisional phenomenon of AV to offer superior driving
decisions and decisional capacity compared to human drivers. As such it
examines AV decisions in terms of the safety and risk mitigation benefits
inherent in the technology. However, the need to counter any potential
misunderstanding that may arise remains evident. AVs are widely regarded
as technologies which can outperform human driving capabilities because
although humans are statistically proficient drivers and It is possible to drive
a private car 13000km a year for fifty years with more than a 99 percent
chance of survival(von Suntum, 1984: 160), human drivers make human
errors. In fact, the NHTSA report that recognition errors, decision errors,
performance errors, and non-performance errorscontribute to 94% of the
critical reasonsof road traffic accidents (Singh 2015). Their analysis is
supported by research undertaken by Evans (1996) which also concluded
that The road user was identified as a sole or contributing factor in 94% of
crashes in the US study and in 95% of crashes in the UK study. The central
claim which supports the ethical justification of the safety argument and the
development and use of AV then, is that an AV which functions at least as
efficiently as human drivers will eliminate many of the human errors which
directly contribute to road traffic fatalities. Since this contention is far more
complex than it first appears however, this paper aims to properly elucidate
the reasoning of the argument.
As stated above, statistical driving data to date highlights an alarming
correlation between erroneous human driving decisions and road accident
fatalities. In light of this, advocates purport that AVs offer the opportunity to
dramatically reduce human decision-related driving fatalities (General
Motors 2018; LaFrance 2015; Bertoncello and Wee 2015; Litman, 2017).
Many other commentators are equally enthusiastic about the potential safety
benefits but caution that these remain purely hypothetical unless issues such
as policy (Campbell 2018) and practical implementation (Grush and Niles
2018: Schiller et al. 2018) are addressed. Nonetheless, numerous states,
including the U.S., U.K., and Germany, have already fast-tracked policy to
support AV research and testing, and both the automotive industry (GM:
2018) and private funded research (Anderson and Kalra et al. 2014) insist
that AVs afford significant safety benefits that will save countless lives.
However, Campbell (2018) points out that while there may be acceptance
of AV safety arguments in principle, there is prevailing confusion as to how
the technology is to be supported and such benefits actualised.
APPLIED ARTIFICIAL INTELLIGENCE 9
The opportunity to save lives provides a compelling ethical basis to sup-
port any innovative technology, and when directly contrasted to AVs, human
driving decisionality is characterised by bad or inferior decisions which cost
lives. The safety argument of AV technologies stresses the core safety benefits
which are built upon more accurate driving ability and supported by
advanced decisional capacity to navigate the road network. The focus on
decisional intelligence and capacity as a means to reduce RTAs and fatalities,
ringfences the problem space: namely, that the use of AVs will avoid most
types of RTAs by eliminating the opportunity for human driving decision
error. It therefore follows that the success of the technology hangs on the AV
capacity to make effective driving decisions and reduce the frequency of RTA
loss of life and limb. The crux of the argument is that the replacement of
human driver by AVs will save lives by decreasing the occurrence of distrac-
tion, intoxication, or fatigue-related RTAs. What is most interesting about
the safety argument is its reliance on the technological realisation of high-
level operational success. Thus for the safety argument to achieve its aims the
technology must accomplish a level of effective driving decisional capacity
equal to the success metrics of human driving decisional capacity. The AV
ability to do this rests on their ability and decisional capacity to make more
consistently accurate driving decisions than humans and traverse the human
road network with at least an equal statistical record of safety performance to
human drivers.
The challenge is that for AVs to accomplish the claim of reducing the
frequency of human driving decision error in the core categories which make
up 94% of driving errors, AVs must accomplish the global feat of improving
across the spectrum of all driving decisions. There are two parts to this
analysis of human driving decisions: the erroneous (bad) decisions that
make up the NHTSAs 94% claim; and the global frequency of successful
(good) decisions. Yet, the argument automatically defaults to a concentration
on technology by removing the array of bad driving decisions which arise
from key example areas. However, for AV technologies to reduce the 94%
frequency of bad human driving decisions it must first achieve the opera-
tional milestone of at least equalling good human driving decisions which
depends on a successful roll-out of the technology. While statistically such
a benefit can further support the overall anticipated global performance
metric, the primary challenge is still to reach a level of successful driving
decisions equivalent to good human driving decisions. This is largely omitted
from the ethical justification of safety arguments which hinge on statistical
evidence of the AV capacity to outperform humans in the single category of
human driving decisions. By doing so, it yokes the safety argument to
decisional benefits, which, in turn, depend upon the global decision perfor-
mance of the technology to outperform the most successful aspects of human
decisionality.The WHO (2018) has estimated 1.35 million RTA related deaths
10 M. CUNNEEN ET AL.
in 2016 and this is a figure that continues to climb. When considered in
relation to the NHTSA assessment it is not surprising that many of these
fatalaties relate to erroneous human driving decisions. However, the ratio of
good driving decisions is simply not considered. In fact, the target figure
required to justify the technology is not the ratio of decisions which account
for RTA fatalities, but rather the ratio of human miles driven safely. Thus, the
many points of tension within the safety argument include: (1) paradoxically,
and as is the case with all forms of data-dependent analysis, human bias can
compromise data accuracy and the conclusions drawn from analysis; (2),
when reconsidered in light of the above critique of good human driving
decisions, the claims of the safety argument will take considerable time to
justify AV use.
The following section argues that the safety argument as a key metric in
developing a conceptual framework of anticipatory SEL impact of AV goes
too far in claiming decisional superiority while so many unresolved chal-
lenges remain, and criticisms persist regarding the efficacy, ability, and
scalability of the technology.
The Problem with the Safety Argument
The safety argument has framed the safety claims of AV decisions to rest on
a conception of decisionality which only addresses the most problematic
decisions of human driving. It is indisputable that AV will offer a driving
decision spectrum which will preclude intoxication, distraction, fatigue, and
poor behavioural decisions such as speeding. However, this argument is more
problematic than it appears as it depends on the overall ability of the
technology to outperform the full range of human driving decisions.
Indeed, whether this is achievable remains a matter of conjecture, particularly
considering that AV will bring new accident metrics to the safety figures.
There will undoubtedly be RTAs uniquely tied to AV decisionality in the
global context, such as sensor error, programming bugs, unanticipated
objects, classification error, and hardware faults. While it is hoped these
will precipitate less damaging incidents due to speed limitations and safety
mechanisms, they nonetheless present further challenges to AV safety argu-
ments and the target of global improvement in driving decisions. Effectively,
this means that if, for example, the key categories of human driving error
could be removed from the equation of road safety statistics, the resulting
figures would point to a human driving capacity, perhaps beyond that of any
emerging driving technologies. The analysis claims that the premises of the
safety argument and the statistical RTA figures which support the claim that
AV decisions are safer than human driving decisions, cannot, at present, be
maintained. The safety argument concerns the most problematic decisions of
human driving, yet completely elides the immensely efficient spectrum of
APPLIED ARTIFICIAL INTELLIGENCE 11
decisions which human driving represents. To boost statistical gain, it builds
on the removal of problematic decisions from the driving spectrum. What
remains unclear, however, is that in doing so, it must provide a core deci-
sions spectrum which is as successful as the normal majority of human
driving. The latter point is the most contentious part of the argument,
given that the former claim depends on the success of the latter, which
will, in turn, remain unverifiable for some time. It is evident that the safety
argument fails to provide an accurate account of AV decisional capacity. For
AV to resolve this aspect they must achieve a consistent level of competency
across the entire spectrum of human driving decisions.
The Safety Argument Does Not Define the SEL Impact of AV Decisions
Given that studies such as Blanco et al. (2016) conclude that contrasting AV
driving data to human driving data cannot effectively determine the outcome,
the safety argument cannot be justified by data analysis alone. Moreover, while
many studies have produced figures quantifying human driving safety there
are numerous issues inherent in using such data (ibid). A number of studies
and sets of figures stand out in this regard. Early analysis by von Suntum
maintains that It is possible to drive a private car 13000km a year for fifty years
with more than a 99 percent chance of survival(von Suntum, 1984, 160).
2
More recent analysis by Hevelke & Rumelin (2015) concludes that in the
period 2005 to 2009, there was one accident for every 1.46 million kilometres
travelled. On this basis Goodall (2014a) calculates that with a Poisson dis-
tribution and national mileage and crash estimates, an automated vehicle would
need to drive 725,000 mi on representative roadways without incident and
without human assistance to say with 99% confidence that they crashed less
frequently than vehicles with human drivers(Goodall 2014a). While driving
statistics and analyses are an important dimension of traffic management and
safety, a report by the Virginia Tech Transportation Institute questions the
practice of using such figures as definitive support for safety analysis and
highlights the problems inherent in relying on such data (Blanco et al. 2016).
They assert that not only is the human driving data questionable due to the
omission of up to 59.7% of unreported accidents, but also because the format
and criteria of driverless vehicle crash reporting data presents a very different
challenge (ibid). At most, it can offer one aspect of a multifaceted investigation
into the SEL impact. Difficulties regarding the efficacy of contrasting driving
data are also brought to the fore by Schoettle & Sivak (2015) who draw
attention to the 1.5 million autonomous driving miles and the 3 trillion
conventional vehicle miles travelled annually in the U.S. alone. Such contrast-
ing figures cannot deliver the data required to develop an accurate SEL impact
or support the claims of the safety argument. In fact, each research group
identifies significant differences which influence the driving data; so much so,
12 M. CUNNEEN ET AL.
the task has been likened to comparing apples and oranges(Blanco et al.
2016). For example, AV driving miles is predominantly defined by open road
good weather driving which does not include driving in fog, adverse weather
events, or snow (Schoettle & Sivak, 2015). Research groups also concur that
current figures suggesting a greater frequency of crashes in autonomous
driving than human driving are probably based on inaccurate findings due
to the limited exposure of the self-driving car project to real-world driving
increases statistical uncertainty in its crash rate. That uncertainty will decrease
as it receives more on road in traffic testing(Blanco et al. 2016).
Clearly data deficiencies will persist for some time, but even when all the
data is in, certain issues concerning the accuracy of contrasting human
driving data to autonomous driving data will inevitably remain. For instance,
questions regarding environmental variables such as weather and object
classification, and which kinds of accidents qualify, will continue to be raised
(Schoettle & Sivak, 2015; Blanco et al. 2016, 40). If the safety argument or
similar claims that AVs will demonstrate safer driving and save lives are to be
verifiable, a more complex analysis of the driving capacity, specifically
regarding driving decisions, must be provided. Yet it is clear that a direct
contrast between human and autonomous driving data cannot provide defi-
nitive support to claims relating to the SEL impact or otherwise. Data
analysis may well support anticipatory scrutiny of the SEL impact, but it
cannot be definitive, since numerous issues of data accuracy and methodol-
ogies compromise such dependencies and contest such claims. As such, it is
possible that statistical data may not provide the safety argument with the
justification it seeks. On the contrary, since it is a purely hypothetical
argument based on the belief that technology can eliminate the key categories
of human driving decisional error, the matter of whether advanced AVs can
improve the safety statistics for human travel by road may remain uncertain
for some considerable time. In effect, AVs must achieve 1.7 million accident-
free, diverse weather, driving miles before figures can properly evaluate their
performance and verify the claims of the safety argument. Considering the
above criticisms, the safety argument cannot define the meaning of AV
decisionality. Nor can it claim to represent the sum of AV decisions. At
most, it can inform a partial understanding of one layer of the AV decisional
ontology. The ethical challenge builds its critique of the safety argument on
such inferences.
Part Four: The Ethical Challenge
The above safety argument asserts a framing of AV decisions that mitigate
the many human driving decision frailties which constitute risk scenarios
within the driving phenomenon. However, the same AV driving intelligence
technologies which support and determine its decisional capacity also present
APPLIED ARTIFICIAL INTELLIGENCE 13
when compared to human driving decision capacity in terms of decisional
limitations. In fact, AVs face difficult challenges in responding to driving
events that involve the multi-layered space obtaining human values, ethics,
and laws. For example, there are immense complexities inherent in program-
ming autonomous intelligent machines to identify, process, and carry out
decisions which conform to human values and ethics. This emphasis on the
ethical component of driving decisions is highlighted by Gerdes &
Thompson (2016) who suggest that the decisional capacity of AVs will
ultimately be judged not by statistics or test-track performance, but rather
through the ethical lensof the society in which such vehicles operate. They
further claim that the questions of machine morality both as a capacity and
as moral intelligence that supports moral analysis and decisionality, along
with societys moral analysis of the AVs decisions, will determine how society
anticipates the SEL impact of AVs decisions.
The second conceptual framework explored in this paper concerns possi-
ble AV decisional limitations. While the safety argument focuses on key
decision benefits and risk mitigation, many commentators have identified
the potentially obverse decisional limitations to the technology which may
engender new forms of driving risks. Since it is unlikely that AV will have the
capacity to make decisions which encompass human values, rights, societal
norms, and ethics, one such framing relates to the ethical limitationsof the
technology. Many view this distinction as a significant technological deficit
which, in the case of unavoidable road traffic accidents, could present new
risks to society and users. For example, certain programmable AV decisions
will intrinsically align with predefined human rules, such driving laws, driv-
ing codes, social norms, and accepted conduct. Other decisions will be
autonomous as AVs present a complex decision ontology (Cunneen et al.
2018). For AV to function they will necessarily require a wide scope of
decisionality, given that that certain decisions are devised to override erro-
neous human instructions, and some decisions may even break laws as we
know them. AV driving intelligence will consist of diverse intelligence com-
ponents designed to best support navigation of the human road network. The
ability of a machine to make decisions and traverse the human road network
without not only causing harm, but in such a way as to be safer than human
drivers, represents a new and important development in the phenomenon of
human/machine relations in that it transposes the uniquely human ability to
safely navigate through the world to a machine. This transfer of ability is
underpinned by the claim that AVs will diminish driving risks and provide
a safer driving experience. Moreover, the ability of machines to replace
human drivers in this way marks an important step towards reliance on AI
and the beginning of a risk-mitigation relationship wherein society increas-
ingly looks to machines to reduce real world risks to humans. In essence, this
14 M. CUNNEEN ET AL.
means that more of the risk phenomenon is allayed by transferring the risk
mitigation decisional context to AI.
Patrick Lin (2011) criticises the safety argument and proposes an ethical
challenge. He backs his position by utilising psychological, moral experi-
ments, and edge cases, to provision the claim that even the most advanced
AV intelligence will have limited decisional capacity. This is evident when the
AV is confronted with scenarios that necessitate decision responses relating
to human rights and morals (Lin 2013;2016). Lin maintains that in the event
of accidents, AVs could fail to respond to scenarios encompassing human
values, through an inability to identify the metrics of the human values,
ethical relations, or legal consequences of an action or inaction. As such it
questionable whether, even with safety improvements in mind, manoeuvring
intelligence alone can support the telos or goal of safe driving. Lins approach
is informed by research carried out by Wallach & Allen (2008) on the
possibility of moral machines. These authors appeal to the trolley dilemma
(TD) and argue that machine morality will be required for a robot carto
adequately respond to driving events and specifically to unavoidable RTAs.
Lin also appeals to numerous hypothetical scenarios to defend his conclusion
that the diversity of the human environment and road network will require
driverless technology to have moral intelligence:
If motor vehicles are to be truly autonomous and able to operate responsibly on our
roads, they will need to replicate-or do better than-the human decision-making process.
But some decisions are more than just a mechanical application of traffic laws and
plotting a safe path. They seem to require a sense of ethics, and this is a notoriously
difficult capability to reduce into algorithms for a computer to follow.
(Lin, 2015)
As variations in the trolley dilemma, edge cases are generally used to evaluate
emotive scenarios and choices regarding the lesser of two evils. When con-
fronted with a no-win scenario, the AI of a driverless car must an immediate
response to an unavoidable RTA. Such a dilemma challenges the AI to make
a decision that will inevitably lead to the death of at least one person. Complex,
no-win, lesser-of-two- evils type scenarios are therefore simulated using emo-
tive values such as a parent and child, elderly women, children and school
buses, or even a doctor, to interrogate social norms, moral expectations, and
individual reasoning.
3
The complex relational interplay which frame such
moral dilemmas arguably support the view that moral intelligence is prerequi-
site for AVs to make accurate, informed decisions in response to life-
threatening, but not, inescapable, driving events (Lin 2015). In short, AVs
will require the capacity to engage with real-life human dilemmas in order to
carry out the function of safely traversing the human road network.
APPLIED ARTIFICIAL INTELLIGENCE 15
The Ethical Challenge
Premise 1: The human road network is laden with human values
The human road network encompasses numerous unpredictable scenarios consisting
of variations of human agents in the form of human drivers, pedestrians, cyclists,
children, and animals. This environment will force morally loaded events and
scenarios upon the driving intelligence of an automated system.
Premise 2: AV will necessarily make decisions which involve human values
When confronted with events and scenarios which involve human values, possible
harm, or loss of life, AVs will make decisions which directly impact on human
welfare, even if they are only programmed to classify objects they perceive in the
environment, according to classifiers of size and shape.
Conclusion:
Our human road network presents a diverse and fluid environment. All vehicles
negotiating this environment will undoubtedly be confronted with unexpected events,
potential collisions, and life or death RTAs. AV responses to such moral scenarios
depend on the intelligence frameworks which determine its actions. In the absence of
moral intelligence AVs will mediate moral scenarios as non-moral scenarios via
a value spectrum which is solely predicated on relational quantifications between the
individuals and objects.
AV will respond and make decisions based on limited data and hold to identifi-
able values relating to object metrics of size, mass, and speed. Lins application
evinces the confusion as one that seeks to address human driving intelligence
(HDI) and its unique contextual meaning to be isomorphic to artificial driving
intelligence (ADI) (Cunneen et al. 2018). Lins use of the trolley dilemma applies
a human moral and psychological thought experiment to ADI. While such
a scenario can apparently support doing so, the notion of confronting an ADI
with such a decision is problematic for several reasons. The challenge is in
understanding why. It is only by considering the ontological differences in
HDI and ADI as driving intelligence supporting determined decision capacities
and spectrums of decisions, that the significant differences between them
become clear. Nonetheless,the emotive strength of the scenario has been integral
to the dominance of many negative news headlines. It has also played a role in
detracting from questions about the compatibility of human moral experiments
to AI and autonomous machines. Lins approach is supported by similar analysis
developed by Noah Goodall (2014a), who presents three arguments: that even
perfectly functioning automatedvehicles will crash, that certaincrashes require the
vehicle to make complex ethical decisions and that there is no obvious way to
encode human ethics in computers. Finally, an incremental, hybrid approach to
develop ethical automated vehicles is proposed(Goodall 2014a). Goodall appeals
to the two further predicates of Lins argument: he maintains that AVs will
themselves crash; he purports that since unavoidable RTAs will present
16 M. CUNNEEN ET AL.
scenarios obtaining to human values, AVs need moral intelligence. As Goodall
concludes, the main problem is the absence of current programming technology
which can support the ethical resolution of complex moral decisions within the
AV decisional spectrum. In other to do this, driving intelligence would need to
be augmented with moral intelligence in order to process identifiable moral
dilemma data and generate moral decision options within the decisional spec-
trum. Goodalls approach is particularly interesting in that he responds to this
difficulty by postulating options which could support moral decision processing.
He is not alone in this.
Many others have also claimed that moral theorisations can be pro-
grammed into machine intelligence.
4
Goodalls view supports the ethical
challenge by claiming that; If, however, injury cannot be avoided, the auto-
mated vehicle must decide how best to crash. This decision quickly becomes
a moral one…” (Goodall 2014a,60) This contention is echoed by Korb (2007)
who claims that all AI inhere a necessary duty to ensure that the intelligence
has cognisance of its actions being adjudged to be right or wrong. The AI
must have parameters of acceptable, unacceptable, right and wrong beha-
viour if the autonomous behaviour is achievable.
But if we build a genuine, autonomous AI, we arguably will have to have built an
artificial moral agent, an agent capable of both ethical and unethical behaviour. The
possibility of one of our artefacts behaving unethically raises moral problems for their
development that no other technology can.
(Korb, 2007)
The possibility of programming machines to make ethical decisions is highly
controversial. For instance, ensuring that the processing of ethical datasets does
not contain programming biases raises concerns. Verifying that the decisions
carried out are lawful and embody our individual and societal values also pose
difficulties, as do issues of transparency of manner and access. Such challenges
suggest programming ethics into socially embedded machines to be an insur-
mountable task: all the more so when they relate to decisions which have a direct
bearing on human harm or loss of life. There are two strands of response to this
situation. The first represents the general view of the industry and attests to
insignificant applications in respect of the issues surrounding moral decision-
ality in autonomous vehicles. As such, the practical application of moral decision
intelligence is not deemed significant enough to warrant the research, invest-
ment, and/or development of a moral decisional capacity for autonomous
vehicles. This cost/benefit view of moral decision capacity is refuted by techno-
logical ethicists such as Lin, who argue that no matter how low the frequency of
moral dilemmas confronting autonomous vehicles, it should be intrinsic to the
decision capacity of autonomous vehicle programming. Lins critiques have
fuelled research attempts to address the challenge of programming ethics into
autonomous vehicles.
APPLIED ARTIFICIAL INTELLIGENCE 17
The move to appeal to an artificial moral agent is often contextualised in
emotive life or death scenarios of decisions or actions taken by a vehicles
operating system about unavoidable RTAs. However, Goodall raises the
important point that certain crashes are unavoidable. Thus, no matter how
good the sensory and impact avoidance software is, there should always be
some calculation for damage limitation when AVs will crash. Goodall follows
the standard approach to AVs actions about some inevitable collision states
(Fraichard and Asama, 2004), as it is evident that the standard approaches to
AVs and RTAs frame actions as moral actions. Goodall maintains that an
automated vehicles decisions that preceded certain crashes had a moral
component, and the challenge concerns the reality that there was no
obvious way to encode complex human morals in software(Goodall,
2014a, 58). A similar view is put forward by Kumfer et al. (2016)in
A Human Factors Perspective on Ethical Concerns of Vehicle Automation.
This paper stresses that nothing in current AV programming supports moral
decisionality and perhaps it never will. This creates a scenario where we are
assessing the moral actions of AVs, but the only moral context is the human
moral framework we are applying to it. Once again, this is an unexpected
double-bind requires further investigation. Goodalls emphasis on optimal
crash strategies demonstrates how the pre-crash window presents an example
of decisional limitations, and how such an ability introduces morality into
the decisional capacity by supporting the need to investigate how moral
values might be applied (Goodall, 2014a). This directly addresses what
Goodall denotes as a clear deficiency in governance and U.S. state law to
cater for computerized control of pre-crash or crash avoidance behaviour
(ibid). Goodall is undoubtedly correct in his assessment of the need to
investigate such behaviour, especially as it relates to what he describes as
crashes with a moral component. However, Goodall, like many others who
subscribe to this standard interpretation of approaching the operating system
governing control of AVs as artificial moral agents, has unnecessarily and
emotively weighted the analysis. As soon as a hypothesis of artificial agency is
applied to a device with interactive qualities, it expands to a far more
complex conceptualisation of artificial moral agency. As such, there is not
only a need to clarify the conception of artificial agency, but also, to address
the seemingly unavoidable expansion to the artificial moral agency and to
elucidate and qualify the application of both conceptions of agency.
an AV cannot cope with situations that were not anticipated and taken into
account by the programmer (even if it includes a learning approach). Overall, an
AVs decision making is imperfect and uncertain.
(Dogan et al, 2016)
18 M. CUNNEEN ET AL.
The Ethical Challenge
Ethical tensions align with the various ways decisions carried out by AVs reflect
moral preferences as general principles, such as to ensure our safety, and cause
no harm to others. For this reason, moral challenges in the form of possible
decisions are bound to confront the technology. Lin is undoubtedly accurate in
this respect (ibid). Nonetheless, AV morality may not comparable to human
morality, and therefore unamenable to assessment by the same metrics of
morality we use to judge people. This is one reason why the focus on decision
intelligence and capacity is essential, but it should also be framed in such a way
that can be appropriately conceptualised without anthropomorphising the deci-
sional process. While this tendency is more prevalent outside the academy,
anthropomorphises may nonetheless impact the media discourse which plays
an important role in formulating public opinion and policy. Analogous debates
around genetically modified crops are instructive in this regard (Frewer et al.,
2002; Turney, 1998).
5
Our case here is that in order to address this tendency for
a certain type of Rhizome-like framing (Deleuze and Guattari, 2009) which takes
the discourse in a certain direction it is necessary to revert to the conceptual
frame. This is all the more so when debates brush against mythic narratives
which anthropomorphise the issue. To accurately understand the moral context
of AV decisionality then, numerous layers of analysis must be ontologically
elucidated (Cunneen et al. 2018). Questions such as how the machines were
programmed to respond to a given scenario, how they classified key objects,
people and relationships, and whether the moral analysis was predetermined or
arose from accurate autonomous moral decisions as a result of machine learning
or an adaptive algorithm, must be addressed.
Yet, the unreliable advocacy of the safety arguments which compare AV and
human decision data based on statistical data analysis, along with the biases of
public perception, arguably impede any accurate anticipation of the SEL impact
of AV. The field of ethics, and particularly applied ethics, boasts a tradition of
vanguarding societal safety by investigating tensions when law and policy fail to
present the issue in a format which satisfies all interested parties, such as nuclear
fuel, nuclear arms, and child labour. However, in the context of AV, even ethics
has been criticised for its apparent failure to accurately anticipate the SEL impact
of AV. Considering recent ethical critiques of the safety argument, many
commentators question the format and use of ethical challenges in focusing
on the question of moral limitations of AV decisional capacity, while Nyholm &
Smids (2016) maintain the ethical challenge developed by Lin (2011, 2012, 2015)
is confused. Their analysis is centred on Lins use of variations of the TD to
support his hypothesis (Lin, 2011, 2015) which underscores the clear disanalo-
gies between AVs and the use of the TD (Sven and Smids, 2016). Noah Goodall,
a vocal supporter of Lins ethical challenge to AV, also flags the difficulties posed
by the use and misuse of the TD in elucidating AV decision limitations (Goodall,
APPLIED ARTIFICIAL INTELLIGENCE 19
2013). Charisi et al. (2017) similarly criticised the TD as a means of supporting
analysis of the SEL impact, maintaining that the question should rather be
focused on ensuring that the ethical programming of the AV decisional capacity
is achievable in a transparent manner (ibid). Lin himself has also come to
acknowledge the contingent nature of the TD in the context of AV, but insists
that as it cannot be completely dismissed as a potential scenario wherein AV will
be confronted with weighing moral decisions, the challenge persists (Lin, 2015).
An Ethical Lens: Societal Perception and Anticipating SEL Impact
Lins criticisms of the safety argument have important ramifications in
relation to the public use of the technology, public risk perception, and
informed consent. If we are to judge based on media headlines such as killer
cars, the challenge of machine morality has already coloured public percep-
tion and risk perception of an ability of AV to make autonomous moral
decisions.
6
If Gerdes and Thornton (2016) hypothesis is proved correct,
societys perception of AVs will be the key determinant of the SEL impact
of the technology. Their stance privileges social perception and conceives
judgement of autonomous technologies to be bound by societysethical
lens(Gerdes and Thornton, 2016). Public risk perception is clearly averse
to AV accidents that are highly publicised, even when countered by claims
that millions of miles of safe driving have been achieved. However, when one
considers the numerous news headlines and the data obtained from public
questionnaires (Bonnefon et al.,2016) regarding AV safety and user percep-
tion, societal concerns align with the ethical challenge and contrast against
the claims of the safety argument.
The importance of understanding the concept of AV decisionality to
anticipate the SEL impact of the technology has been closely discussed
and defended in this paper, despite evidence that both research and
public perception already misunderstand it. As such, there is a clear
need to provide an accurate account of autonomous decisional capacity
by elucidating the concept of AV decisionality. This will require ring-
fencing the limitations of decisional capacity as the technology evolves.
This contrasts both positions and takes the view that the crux of the SEL
impact will concern the decisions that AV can or cannot make. This
disparity in interpreting AV decisional capacity identifies an underlying
difficulty in terms of framing AV and autonomous technology decision-
making. Autonomous vehicles present an autonomous decision-making
agency immersed in one of the most fast-paced and unpredictable
aspects of modern society; namely, the human road network. This
phenomenon of driving is known to elicit emotive responses from the
most rational of people. As such, it inhabits an unpredictable emotive
human space that AVs must quantify in numerous different ways. The
20 M. CUNNEEN ET AL.
top-tier challenge therefore is to produce a technology which can satisfy
the demands and expectations of a public that will, many suggest, have
a low tolerance of AV decisional errors. Carl Nash (2017), for instance,
contends that low public tolerance for AV-related fatalities could have
unexpected effects on the stakeholders involved in developing the
technology.
7
TheultimatesuccessofAVthen,willdependupon
a combination of factors contributing to general social perception, com-
prising societys ethical lens (Gerdes & Thompson, 2016) and societys
risk perception (Litman, 2017) of the technology. If this is an accurate
assessment of the underlying drives which will ultimately determine the
SEL impact, then the focus on AV decisionality will effectively be
instrumental to and anticipate it.
Conclusion
The autonomous vehicle is but one application of artificial intelligence technol-
ogies which have an enormous bearing on contemporary and future society. As
AI implementations become more diverse, and indeed ubiquitous, there will
a greater need to understand the different contexts of decisional application.
Essentially this means that in order to accurately frame each unique decision
context the technology must to be taken at face value and a nonlinear relational
model of classification concepts created. The implication is that we should not
strive for a general one size fits allconception of artificial intelligence applica-
tion as no single framework can sufficiently account for the numerous possibi-
lities of machine decisional applications. Conceptual framing is an integral part
of an epistemological continuum which supports governance and regulation.
Given the strategic position is occupies in chains of meaning which promote
public debate, it is vital that such framing is open to interrogation. Overall, this
paper examines how the crucial process of conceptual framing has received
relatively little attention from academics in the area of anticipatory governance.
The central argument is that conceptual framing has downstream effects in
terms of the debates on the governance of automated vehicles. It is a nuanced
argument in that the idea of the unitary conceptual framework is not posited
per se. Rather an acceptance of the limitations of current debates is acknowl-
edged, along with calls for more precision in the construction of such frame-
works, and perhaps paradoxically, an exploration of the value of reflexive
approaches to conceptual framing. In fact, it is inherently self-defeating to insist
on complete and final conceptual frameworks. Instead, conceptual framing
should be a more reflexive practice with an iterative component, as it is not
just a matter of accuracy in terms of the concepts used, but rather a realisation of
the impact of such framing that counts. Since both the safety arguments and the
debates around the ethics of AVs launch discourse in a particular direction, there
is a need to revisit the initial framing and to adopt a more pluralistic outlook in
APPLIED ARTIFICIAL INTELLIGENCE 21
terms of that conceptual framing. Such is the complexity and amplification of
debates around the societal impact of AVs that there is a tendency to pare back
the debate for more general consumption. While this is normal in any field, is
particularly prevalent in the area of emerging technology. For instance, debates
on the risks posed by nanotechnology demonstrate similar weaknesses in terms
of initial conceptual framing and the results have been regulatory shortcomings
and widespread confusion amongst stakeholders. Reintroducing precision into
conceptual framing, and indeed an acceptance that conceptual frameworks
occupy an important position in any hermeneutic cycle, can help move debates
on the deployment of AVs forward. The subsequent payoff enables technologies
with positive SEL impacts to be better supported, while technologies with
potentially negative SEL impacts can be framed more accurately, and through
properly informed assessment, can be further developed.
Notes
1. As will be outlined in the paper there is a clear risk that inaccurate conceptual frame-
works can have adverse and serious ramifications for the investment, governance, and
public perception of technologies. This is immensely problematic when public perception
holds to emotive and unsubstantiated claims relating to potential adverse impacts and
risks. There are numerous examples from genetically modified foods, global warming to
vaccinations. Each presents clear examples of a negative public response to innovation
due to difficulties in how innovation is conceptually framed, and communicated.
2. For a more recent breakdown of RTA figures and related causes, see figures at: https://
crashstats.nhtsa.dot.gov/Api/Public/ViewPublication/812115.
3. MIT Moral Machine experiment .
4. Interestingly, Yilmaz et al (2016) point out that research into machine ethics now affords
a realistic scenario that will also contribute to a better understanding of human ethics.
5. https://www.theguardian.com/technology/2015/dec/23/the-problem-with-self-driving-
cars-who-controls-the-code
https://www.theguardian.com/science/2016/jun/14/statistically-self-driving-cars-are-
about-to-kill-someone-what-happens-next.
6. A concerning pattern here relates to the dramatic imagery of Phillipa FootsT
rolley
Problem as an emotive image of the challenges to machine ethics; Phillipa Foot, The
Problem of Abortion and the Doctrine of the Double Effect(1967). Chris Urmson is
critical of such philosophical discourse: https://www.washingtonpost.com/news/innova
tions/wp/2015/12/01/googles-leader-on-self-driving-cars-downplaysthe-trolley-problem/ .
7. If AVs fail to live up to expectations or become ensnared and held back due to a rush
to develop legislation to respond to identifiable safety concerns raised by public and
competing industrys critiques, as in the case of autonomous trains, the prospect of
AVs arriving in any developed scale could be hindered.
Funding
This work was supported by the Horizon 2020 [690772].
22 M. CUNNEEN ET AL.
References
Alic, J. (1994). The dual use of technology: Concepts and policies. Technology in Society 16
(2):15572. doi:10.1016/0160-791X(94)90027-2
Allenby, B. R. (2011). Governance and technology systems: The challenge of emerging technol-
ogies.InThe Growing Gap Between Emerging Technologies and Legal-Ethical Oversight (pp.
3-18). Springer, Dordrecht
Anderson, James M., Nidhi Kalra Stanley, K. D., Sorensen, P., Samaras, C., & Oluwatola, O.
A. (2014). Autonomous vehicle technology: A guide for policymakers. Santa Monica, CA:
Rand Corporation
Asveld, L., and S. Roeser. 2009.The ethics of technological risk. London: Routledge.
Bachmann, R., N. Gillespie, and R. Priem. (2015). Repairing trust in organizations and
institutions: Toward a conceptual framework. 36 (9):112342
Berger,A.N.,andG.F.Udell.2006. A more complete conceptual framework for SME
finance. Journal of Banking & Finance 30 (Issue):11. doi:10.1016/j.
jbankfin.2006.05.008.
Bertoncello, M., and D. Wee 2015. Ten ways autonomous driving could redene the auto-
motive world. Accessed: http://www.mckinsey.com/industries/automotive-and-assembly
/our-insights/ten-ways-autonomous-driving-couldredene-the-automotive-world
Blanco, M., J. Atwood, S. Russell, T. Trimble, J. McClafferty, and M. Perez. (2016).
Automated vehicle crash rate comparison using naturalistic data. Virginia Tech
Transportation Institute Report. http://www.vtti.vt.edu/PDFs/Automated%20Vehicle%
20Crash%20Rate%20Comparison%20Using%20Naturalistic%20Data_Final%20Report_
20160107.pdf (Accessed December 19, 2017).
Bonnefon, J. F., Shariff, A., & Rahwan, I. (2015)Autonomous vehicles need experimental
ethics: Are we ready for utilitarian cars?. (Accessed November 2017)Available at: http://
arxiv.org/pdf/1510.03346v1.pdf
Bonnefon, J. F., Shariff, A., & Rahwan, I.. 2016. The social dilemma of autonomous vehicles.
Science 352 (6293):157376. doi:10.1126/science.aaf2654
Bringsjord, S., and A. Sen. 2016. On creative self-driving cars: hire the computational
logicians, fast. Applied Artificial Intelligence 30 (8):75886. doi:10.1080/
08839514.2016.1229906
Burgess, J., and J. Chilvers. 2006. Upping the ante: A conceptual framework for designing and
evaluating participatory technology assessments. Science and Public Policy 33 (10):71328.
doi:10.3152/147154306781778551.
Campbell, H. 2018. Who will own and have propriety over our automated future?
Considering governance of ownership to maximize access, efficiency, and equity in cities.
Transportation Research Record, 2672(7), 14-23
Charisi, V., Dennis, L., Fisher, M., Lieck, R., Matthias, A., Slavkovik, M., ... & Yampolskiy, R.
(2017). Towards moral autonomous systems. arXiv preprint arXiv:1703.04741
Choudury, C. (2007) Modelling driving decisions with latent plans. Ph.D. thesis,
Massachusetts Institute of Technology, Cambridge. doi: 10.1094/PDIS-91-4-0467B
Coates, J. F., and V. T. Coates. 2016. Next stages in technology assessment: Topics and tools.
Technological Forecasting & Social Change 113:11214. doi:10.1016/j.techfore.2016.10.039.
Coeckelbergh, M. 2016.Responsibility and the moral phenomenology of using self-driving
cars.Applied Artificial Intelligence 30 (8):74857. doi:10.1080/08839514.2016.1229759.
Cook, C., and K. Bakker. 2012. Water security: Debating an emerging paradigm. Global
Environmental Change 22 (1):94102. doi:10.1016/j.gloenvcha.2011.10.011.
APPLIED ARTIFICIAL INTELLIGENCE 23
Cunneen, M., Mullins, M., Murphy, F., & Gaines, S. (2019). Artificial Driving Intelligence and
Moral Agency: Examining the Decision Ontology of Unavoidable Road Traffic Accidents
through the Prism of the Trolley Dilemma. Applied Artificial Intelligence, 33(3), 267-293
Deleuze, G., and F. Guattari. 2009.A thousand plateaus. Berkeley, CA: Venus Pencils.
Dogan, et al.(2016). Ethics in the design of automated vehicles: The AVEthics project.
(accessed March 21st, 2017). Available at: http://ceur-ws.org/Vol-1668/paper2.pdf.
Donk, A., J. Metag, M. Kohring, and F. Marcinkowski 2011. Framing emerging technologies:
Risk Perceptions of nanotechnology in the German press. Science Communication 34
(1):529. doi: 10.1177/1075547011417892.
Dreyfus, H. 1979.What computers cantdo. New York: MIT Press.
Dreyfus, H. 1986.Mind over machine: The power of human intuition and expertise in the era
of the computer. Oxford, U.K.: Blackwell
Floridi, L., J. Cowls, M. Beltrametti, R. Chatila, P. Chazerand, V. Dignum, and B. Schafer. 2018.
AI4 People: An ethical framework for a good ai society: opportunities, risks, principles, and
recommendations. Minds and Machines 28 (4):689707. doi:10.1007/s11023-018-9482-5
Fraichard, T., and H. Asama. 2004. Inevitable collision states: A step towards safer robots?
Advanced Robotics 18 (10):100124. doi:10.1163/1568553042674662
Franklin, S., & Ferkin, M. (2006). An ontology for comparative cognition: A functional
approach. Comparative Cognition & Behavior Reviews, 1.
Frewer, L. J., S. Miles, and R. Marsh. 2002. The media and genetically modified foods:
evidence in support of social amplification of risk. 22 (4):701-711.
Gasser, U., and V. A. F. Almeida. November 2017. A layered model for AI Governance. IEEE
Internet Computing 21 (6):5862. doi: 10.1109/MIC.2017.4180835.
General Motors. 2018. Self-Driving Safety Report. Detroit: General Motors.
Gogoll, J., and J. Müller. 2016. Autonomous cars: In favor of a mandatory ethics setting.
Science and Engineering Ethics 23 (3):681700. doi:10.1007/s11948-016-9806-x.
Goodall, N. 2014a. Ethical decision making during automated vehicle crashes. Transportation
Research Record: Journal of the Transportation Research Board 2424:5865. doi:10.3141/
2424-07.
Goodall, N. J. 2014b. Machine ethics and automated vehicles. In Road vehicle automation, ed.
S. Beiker and G. Meyer, 93102. Switzerland: Springer.
Goodall, N. J. 2016. Away from trolley problems and toward risk management. Applied
Artificial Intelligence 30 (8). doi:10.1080/08839514.2016.1229922.
Grush, B., and J. Niles, 2018. The end of driving: transportation systems and public policy
planning for autonomous vehicles.
Hevelke, A., and J. Nida-Rümelin. 2015. Responsibility for crashes of autonomous vehicles: An
ethical analysis. Science and Engineering Ethics 21 (3):61930. doi:10.1007/s11948-014-9565-5.
International, S. A. E. 2016. Taxonomy and definitions for terms related to driving automation
systems for on-road motor vehicles. Warrendale, PA: SAE International.
Kaplan, J. (2016). Artificial Intelligence: What everyone needs to know. Oxford University
Press
Johnson, D.G. & Verdicchio, M. Minds & Machines. (2017). 27: 575. https://doi.org/10.1007/
s11023-017-9417-6
Kasperson, R. Ε.(2009). "Coping with Deep Uncertainty" in Bammer, G and Smithson M
(ed.), Uncertainty and Rük: Multidisciplinary Perspectives. London: Earthscan.
Kasperson, R. E., O. Renn, P. Slovic, H. S. Brown, J. Emel, R. Goble, J. X. Kasperson, and
S. Ratick. 1988. The social amplification of risk: A conceptual framework. Risk Analysis
8:17787. doi:10.1111/j.1539-6924.1988.tb01168.x.
Korb, K. B. 2008.Encyclopedia of Information Ethics and Security. ed. M. Quigley, 27984.
Hershey PA USA: Information Science Publishing.
24 M. CUNNEEN ET AL.
Kumfer, W. J., S. J. Levulis, M. D. Olson, and R. A. Burgess 2016, September. A human
factors perspective on ethical concerns of vehicle automation. In Proceedings of the Human
Factors and Ergonomics Society Annual Meeting (Vol. 60, No. 1, pp. 184448). Sage CA:
Los Angeles, CA: SAGE Publications.
Kyriakidis, M., J. de Winter, N. Stanton, T. Bellet, B. van Arem, K. Brookhuis, M. Martens,
K. Bengler, J. Andersson, N. Merat, et al. 2017. A human factors perspective on automated
driving. Theoretical Issues In Ergonomics Science, pp.127
LaFrance, A. Self-driving cars could save 300,000 lives per decade in America. 2015. Available
at: http://www.theatlantic.com/technology/archive/2015/09/self-driving-cars-could-save
300000-lives-per-decade-in-america/407956.
Lin, P. (2013). The ethics of saving lives with autonomous cars are far murkier than you think.
Available at: http://www.wired.com/2013/07/the-surprising-ethics-of-robot-cars. (Accessed
January 15, 2017).
Lin, P. (2015). Autonomous driving: technical, legal and social aspects.Implementable Ethics
for Autonomous Vehicles.Autonomes Fahren, edited by Markus Maurer, J. et al.
Available at: http://link.springer.com/10.1007/978-3-662-45854-9_5_OnlinePDF. Berlin:
Springer Open.
Lin, P. 2016. Why ethics matters for autonomous cars. (pp. 69-85), In Autonomes Fahren, ed.
M. Maurer, J. Gerdes, B. Lenz, and H. Winner. Berlin, Heidelberg: Springer Vieweg
Litman, T. 2017. Autonomous vehicle implementation predictions (p. In 28). Victoria,
Canada: Victoria Transport Policy Institute.
Litman, T. 2017. Autonomous vehicle implementation predictions; implications for transport
planning. Victoria Transport Policy Institute accessed: 8 September 2017
Loh, W., and J. Loh. 2017. Autonomy and responsibility in hybrid systems: The example of
autonomous cars. In Robot ethics 2.0. From autonomous cars to artificial intelligence, ed.
P. Lin, K. Abney, and R. Jenkins, 3550. New York: Oxford University Press
Malle, B. 2014. Moral competence in Robots. Frontiers in Artificial Intelligence and
Applications 273:18998.
Marchant, G. E. 2011. The growing gap between emerging technologies and the law. In The
growing gap between emerging technologies and legal-ethical oversight: The pacing problem,
ed. G. E. Marchant, B. R. Allenby, and J. R. Heckert. Springer. (pp. 19-33). Springer,
Dordrecht
Maxwell, J. A. 2013.Qualitative research design: An interactive approach. Thousand Oaks,
Calif: SAGE Publications.
McCarthy, J. (1955). Dartmouth proposal, available at: http://wwwformal.stanford.edu/jmc/
history/dartmouth/dartmouth.html
McCarthy, J. (1996). What has AI in common with philosophy? Available at: http://wwwfor
mal.stanford.edu/jmc/aiphil.pdf (Accessed December 15th, 2017).
McGrath, J. E., and A. B. Hollingshead. 1994.Groups interacting with technology: Ideas,
evidence, issues, and an agenda. Sage library of social research, 194. Thousand Oaks, CA,
US: Sage Publications, Inc.
Millar, J. (2016). An Ethics Evaluation Tool for Automating Ethical Decision-Making in
Robots and Self-Driving Cars, Applied Artificial Intelligence,30(8):787-809, DOI:10.1080/
08839514.2016.1229919
Nash, C. (2017). Self-driving road vehicles and transportation planning, (accessed December
14th, 2017) https://www.law.gwu.edu/sites/g/files/zaxdzs2351/f/downloads/Self-Driving-
Road-Vehicles-andTransportation-Planning.pdf
National Highway Traffic Safety Administration. (2016). Federal automated vehicles policy:
accelerating the next revolution in roadway safety. US Department of Transportation.
APPLIED ARTIFICIAL INTELLIGENCE 25
Pidgeon, N., Kasperson, R. E., & Slovic, P. (Eds.). (2003). The social amplification of risk.
Cambridge University Press.
Renn, O. 2008. Risk governance. London: Routledge, doi:10.4324/9781849772440
Savage, E., and L. Sterry. 1990.A conceptual framework for technology education. Reston, VA:
International Technology Education Association.
Schiller, P. L., J. R. Kenworthy, N. Aarsæther, T. Nyseth, A. Røiseland, P. McLaverty,
R. N. Abers, M. Douglass, J. Friedmann, R. N. Abers, et al. 2018. When Teslas autopilot
goes wrong. In An introduction to sustainable transportation: policy, planning and imple-
mentation, vol. 1, 3. 110, Aldershot, UK: Lutterworth Press, Cambridge, MA, and Chelsea
Green Publishing Company
Schoettle, B., and M. Sivak. 2015.A preliminary analysis of real-world crashes involving self-
driving vehicles. University of Michigan Transportation Research Institute.
Searle, J. 1980. Minds, brains, and programs. Behavioral and Brain Sciences 3:417.
doi:10.1017/S0140525X00005756.
Singh, S. (2015, February). Critical reasons for crashes investigated in the national motor
vehicle crash causation survey. (Traffic Safety Facts CrashStats. Report No. DOT HS 812
115). Washington, DC: National Highway Traffic Safety Administration.
Sven, N., and J. Smids. 2016. The ethics of accident-algorithms for self-driving cars: An
applied trolley problem? Ethical Theory and Moral Practice 19 (5):127589. doi:10.1007/
s10677-016-9745-2.
Trappl, R. 2016. Ethical systems for self-driving cars: An introduction. Applied Artificial
Intelligence 30 (8):74547. doi:10.1080/08839514.2016.1229737.
Turney, J. 1998.Frankensteins footsteps: Science, genetics and popular culture. New Haven,
London: Yale University Press.
Von Suntum, U. 1984. Methodische probleme der volkswirt-schaftlichen bewertung von
verkehrsunfallen (methodological problems connected with the economic evaluation of
traffic accidents). In: Zeitschrift Fur Verkehrswissenschaft 55:153-167.
Wallach, W., and C. Allen. 2009.Moral machines: Teaching robots right from wrong. Oxford:
Oxford University Press.
Weizenbaum, J. 1976. Computer power and human reason: From judgment to calculation. San
Francisco, CA: W. H. Freeman
Wiener, G., and B. Walker Smith (2013). Automated driving: Legislative and regulatory
action. [Retrieved November 14, 2016] Available from http://cyberlaw.stanford.edu
Wiener, N. 1960. Some Moral and Technical Consequences of Automation, Science. New
Series 131 (3410):135558. May 6, 1960. Published by: American Association for the
Advancement of Science Stable http://www.jstor.org/stable/1705998
World Health Organziation. (2018). Global Status Report on Road Safety 2018. Geneva,
Switzerland.
Yilmaz, L., Franco-Watkins, A., & Kroecker, T. S. (2017). Computational models of ethical
decision-making: A coherence-driven reflective equilibrium model. Cognitive Systems
Research, 46, 61-74.
Young, S. (2016). The moral algorithm: how to set the moral compass for autonomous
vehicles moral decisions by autonomous vehicles and the need for regulation. (Accessed
May 9, 2017). Available at: https://gowlingwlg.com/getmedia/0eb5a71b-37fb-4ea9-a2c5-
065fbc0d1e10/161205the_moral_algorithm
26 M. CUNNEEN ET AL.
... Here, the critical challenges to artificial intelligence for autonomous applications such as (i) sensor integration and performance issues to artificial intelligence and autonomous systems, (ii) complexities and uncertainties to autonomous and associated complex systems and recent developments, (iii) finetuning and optimization approaches, (iv) hardware concerns, and (v) artificial intelligence-integrated opportunities and future research directions are discussed. Cunneen et al. [37] surveyed to elaborate the use of artificial intelligence in various systems of autonomous vehicles. Here, the primary focus is drawn towards using artificial intelligence-integrated conceptual framing that supports governance and regulation. ...
... is section explores the recent research challenges to autonomous vehicles. Details are presented as follows [21][22][23][24][25][37][38][39][40][41]. ...
... Autonomous vehicles offer better driving decisional spectrum that avoids intoxication, distraction, fatigue, and nability to make timely decisions. All of these factors are associated with the ability of the technologies to outperform the human driving decisions abilities [37]. us, advancements in technology to avoid errors and give real-time responses are significant challenges for AI-integrated autonomous vehicles. ...
... Here, the critical challenges to artificial intelligence for autonomous applications such as (i) sensor integration and performance issues to artificial intelligence and autonomous systems, (ii) complexities and uncertainties to autonomous and associated complex systems and recent developments, (iii) finetuning and optimization approaches, (iv) hardware concerns, and (v) artificial intelligence-integrated opportunities and future research directions are discussed. Cunneen et al. [37] surveyed to elaborate the use of artificial intelligence in various systems of autonomous vehicles. Here, the primary focus is drawn towards using artificial intelligence-integrated conceptual framing that supports governance and regulation. ...
... is section explores the recent research challenges to autonomous vehicles. Details are presented as follows [21][22][23][24][25][37][38][39][40][41]. ...
... Autonomous vehicles offer better driving decisional spectrum that avoids intoxication, distraction, fatigue, and nability to make timely decisions. All of these factors are associated with the ability of the technologies to outperform the human driving decisions abilities [37]. us, advancements in technology to avoid errors and give real-time responses are significant challenges for AI-integrated autonomous vehicles. ...
Article
Full-text available
Intelligent Automation (IA) in automobiles combines robotic process automation and artificial intelligence, allowing digital transformation in autonomous vehicles. IA can completely replace humans with automation with better safety and intelligent movement of vehicles. This work surveys those recent methodologies and their comparative analysis, which use artificial intelligence, machine learning, and IoT in autonomous vehicles. With the shift from manual to automation, there is a need to understand risk mitigation technologies. Thus, this work surveys the safety standards and challenges associated with autonomous vehicles in context of object detection, cybersecurity, and V2X privacy. Additionally, the conceptual autonomous technology risks and benefits are listed to study the consideration of artificial intelligence as an essential factor in handling futuristic vehicles. Researchers and organizations are innovating efficient tools and frameworks for autonomous vehicles. In this survey, in-depth analysis of design techniques of intelligent tools and frameworks for AI and IoT-based autonomous vehicles was conducted. Furthermore, autonomous electric vehicle functionality is also covered with its applications. The real-life applications of autonomous truck, bus, car, shuttle, helicopter, rover, and underground vehicles in various countries and organizations are elaborated. Furthermore, the applications of autonomous vehicles in the supply chain management and manufacturing industry are included in this survey. The advancements in autonomous vehicles technology using machine learning, deep learning, reinforcement learning, statistical techniques, and IoT are presented with comparative analysis. The important future directions are offered in order to indicate areas of potential study that may be carried out in order to enhance autonomous cars in the future.
... During their operation, interventions of the vehicle might include warnings as well as automated braking and/or steering. In challenging and critical driving scenarios, intelligent vehicles are likely to make decisions that are confusing to end-users [1], [2], e.g., unexpectedly initiating a lane change. As a way to assist end-users, and to establish trust, explanation provisions have been put forward [3], [4], [5]. ...
... As a way to assist end-users, and to establish trust, explanation provisions have been put forward [3], [4], [5]. While explanations are considered helpful, we argue that they would not be effective in achieving the aforementioned goals if they are not provided in intelligible forms as obligated by the General Data Protection Right (GDPR) Article 12 1 . ...
... Intelligible explanations provision in assisted and automated driving is crucial as it is also a useful approach for 1 upholding accountability. Intelligent vehicles should have explanation mechanisms where the causes and effects of actions can be communicated to the relevant stakeholders in intelligible ways. ...
Preprint
Full-text available
Commentary driving is a technique in which drivers verbalise their observations, assessments and intentions. By speaking out their thoughts, both learning and expert drivers are able to create a better understanding and awareness of their surroundings. In the intelligent vehicle context, automated driving commentary can provide intelligible explanations about driving actions, and thereby assist a driver or an end-user during driving operations in challenging and safety-critical scenarios. In this paper, we conducted a field study in which we deployed a research vehicle in an urban environment to obtain data. While collecting sensor data of the vehicle's surroundings, we obtained driving commentary from a driving instructor using the think-aloud protocol. We analysed the driving commentary and uncovered an explanation style; the driver first announces his observations, announces his plans, and then makes general remarks. He also made counterfactual comments. We successfully demonstrated how factual and counterfactual natural language explanations that follow this style could be automatically generated using a simple tree-based approach. Generated explanations for longitudinal actions (e.g., stop and move) were deemed more intelligible and plausible by human judges compared to lateral actions, such as lane changes. We discussed how our approach can be built on in the future to realise more robust and effective explainability for driver assistance as well as partial and conditional automation of driving functions.
... While bus lane enforcement helps to mitigate risks such as those associated with the environment, social inequality and congestion, it can simultaneously create new risks. These are complex socio-technical risks that cross several socio-economic contexts and can be classified into technical, governance, public perception and legal categories [16,17]. ...
... According to Cunneen et al. [16,17,23,24], the deployment of an emerging technology creates many complex challenges for governance regimes. Governance risk is exacerbated by a lack of clarity about what the best forms of governance are for AI applications, such as automated bus lane enforcement. ...
Article
Full-text available
There is an explosion of camera surveillance in our cities today. As a result, the risks of privacy infringement and erosion are growing, as is the need for ethical solutions to minimise the risks. This research aims to frame the challenges and ethics of using data surveillance technologies in a qualitative social context. A use case is presented which examines the ethical data required to automatically enforce bus lanes using camera surveillance and proposes ways of minimising the risks of privacy infringement and erosion in that scenario. What we seek to illustrate is that there is a challenge in using technologies in positive, socially responsible ways. To do that, we have to better understand the use case and not just the present, but also the downstream risks, and the downstream ethical questions. There is a gap in the literature in this aspect as well as a gap in the actual thinking of researchers in terms of understanding and responding to it. A literature review and detailed risk analysis of automated bus lane enforcement is conducted. Based on this, an ethical design framework is proposed and applied to the use case. Several potential solutions are created and described. The final chosen solution may also be broadly applicable to other use cases. We show how it is possible to provide an ethical AI solution for detecting infringements that incorporates privacy-by-design principles, while being fair to potential transgressors. By introducing positive, pragmatic and adaptable methods to support and uphold privacy, we support access to innovation that can help us mitigate current emerging risks.
... The developed model can assist self-driving cars find the best way. As stated in Wigley et al [7] connected and autonomous cars (CAVs) are assured to convert mobility, makes transportation available to all -even those incapable of driving due age, affordability, or disability. An analysis of limited time CAV visualizations as well as the ramping up of technologies from modest tests to large roll-out is presented in this research. ...
Conference Paper
Full-text available
Autonomous driving vehicles are too known as driver-less cars which is one of the foremost astounding advances of the twenty-first century, anticipated to be driver-less, effective, and crash dodging ideal urban cars of the future. Autonomous cars actually sense the environment, navigate and fulfill human transportation capabilities without any human inclusion. Cameras, radar, lidar, GPS, and navigational pathways help this type of vehicle detect its surroundings. Even when the conditions alter, advanced control systems interpret sensory data to maintain their locations. Autonomous vehicles are on their way to completely replacing the world’s transportation system. To reach this goal, automobile industries have begun working in this zone to realize the potential and unravel the challenges as of now. A few companies have also started their trail. It will aid in reducing traffic, reducing pollution, avoiding maximum accidents, saving time, conserving energy, and improving human safety. As a result, with the aim and vision of eradicating these challenges from our country, we are focusing on an independent car that will assist us in saving ourselves from the daily revelations we generally confront on the road. Besides, it is high time we began working in Bangladesh on a driver-less vehicle
... Even if the programming and development of these machines come to fruition, there are a lot of concerns that oppose it. If development of such an algorithm or AI is made, how can we be sure that there will be no biases in the people developing it [14]? An example of this is the "Trolley Problem" as seen in Fig. 3. ...
Conference Paper
Autonomous vehicles (AV) are technologies that are continuously developing in the past years. Systems are currently under development to make driverless cars a possibility. This paper presents the technologies that are adapted in developing autonomous vehicles as well as the hindrances and challenges of this innovation. The electronic design of AVs is focused on evolving the technological advancements on automated driving systems (ADS) which centralizes in the navigation system, path decision, surrounding perception, and controlling system. With this, the progression of AVs has drastically improved from human interacted vehicles to conditional automation though the technology is still far from achieving fully autonomous driving. This paper discusses the trends and application of analog electronics as well as the various challenges that hinder achieving full automation of the AVs. Furthermore, specific solutions are proposed to aid the mentioned problems. Only the academic studies from 2017 and up were explored in gathering information for this literature review.
... In recent decades, models based on artificial intelligence techniques have performed impressive predictions in different knowledge fields. For example, in exact sciences (Sauceda et al., 2021;Hajibabaei and Kim, 2021;Cerioti et al., 2021;Bahlke et al., 2020;Chmiela et al., 2020;Saar et al., 2021;Artrith and Urban, 2016;Ch'ng et al., 2017;Shallue and Vanderburg, 2018;Sadowski et al., 2016), in social sciences (Ng et al., 2020;Lattner and Grachten, 2019), in technology (Bae et al., 2021;Cunneen et al., 2019;Feldt et al., 2018;Huang et al., 2014), and health sciences (Bashyam et al., 2020;Zhou et al., 2019;Themistocleous et al., 2021;Lagree et al., 2021). ...
Article
Full-text available
We describe the use of artificial intelligence techniques in heterogeneous catalysis. This description is intended to give readers some clues for the use of these techniques in their research or industrial processes related to hydrodesulfurization. Since the description corresponds to supervised learning, first of all, we give a brief introduction to this type of learning, emphasizing the variables X and Y that define it. For each description, there is a particular emphasis on highlighting these variables. This emphasis will help define them when one works on a new application. The descriptions that we present relate to the construction of learning machines that infer adsorption energies, surface areas, adsorption isotherms of nanoporous materials, novel catalysts, and the sulfur content after hydrodesulfurization. These learning machines can predict adsorption energies with mean absolute errors of 0.15 eV for a diverse chemical space. They predict more precise surface areas of porous materials than the BET technique and can calculate their isotherms much faster than the Monte Carlo method. These machines can also predict new catalysts by learning from the catalytic behavior of materials generated through atomic substitutions. When the machines learn from the variables associated with a hydrodesulfurization process, they can predict the sulfur content in the final product.
Article
Presently, autonomous systems have gained considerable attention in several fields such as transportation, healthcare, autonomous driving, logistics, etc. It is highly needed to ensure the safe operations of the autonomous system before launching it to the general public. Since the design of a completely autonomous system is a challenging process, perception and decision-making act as vital parts. The effective detection of objects on the road under varying scenarios can considerably enhance the safety of autonomous driving. The recently developed computational intelligence (CI) and deep learning models help to effectively design the object detection algorithms for environment perception depending upon the camera system that exists in the autonomous driving systems. With this motivation, this study designed a novel computational intelligence with a wild horse optimization-based object recognition and classification (CIWHO-ORC) model for autonomous driving systems. The proposed CIWHO-ORC technique intends to effectively identify the presence of multiple static and dynamic objects such as vehicles, pedestrians, signboards, etc. Additionally, the CIWHO-ORC technique involves the design of a krill herd (KH) algorithm with a multi-scale Faster RCNN model for the detection of objects. In addition, a wild horse optimizer (WHO) with an online sequential ridge regression (OSRR) model was applied for the classification of recognized objects. The experimental analysis of the CIWHO-ORC technique is validated using benchmark datasets, and the obtained results demonstrate the promising outcome of the CIWHO-ORC technique in terms of several measures.
Conference Paper
Automated vehicles are merging more and more progressively in this recent era with greater safety. Beyond cars, there are tractors, trucks and trains that drive us instead of we drive them. Exploring the autonomous bus transportation with the capability to drive back and forth, or in two, opposite directions and side to side with the fail-safe operation for passenger transportation. Here Artificial intelligence is used to assist in hindrance detection, self-localization, and route planning. Also evaluating the information from a passenger’s perspective, which includes tokening and emergencies. It is decked out into the safest technology with the combination of multiple radar sensors and cameras in each corner, which virtually abolishing flaws as well as providing idleness in case a sensor fails. They could also contribute to trouble-free congestion and lessen the demand for new roads.
Article
Full-text available
This article reports the findings of AI4People, an Atomium—EISMD initiative designed to lay the foundations for a “Good AI Society”. We introduce the core opportunities and risks of AI for society; present a synthesis of five ethical principles that should undergird its development and adoption; and offer 20 concrete recommendations—to assess, to develop, to incentivise, and to support good AI—which in some cases may be undertaken directly by national or supranational policy makers, while in others may be led by other stakeholders. If adopted, these recommendations would serve as a firm foundation for the establishment of a Good AI Society.
Chapter
In this chapter, we give a brief overview of the traditional notion of responsibility and introduce a concept of distributed responsibility within a responsibility network of engineers, driver, and autonomous driving system. In order to evaluate this concept, we explore the notion of man-machine hybrid systems with regard to self-driving cars and conclude that the unit comprising the car and the operator/driver consists of such a hybrid system that can assume a shared responsibility different from the responsibility of other actors in the responsibility network. Discussing certain moral dilemma situations that are structured much like trolley cases, we deduce that as long as there is something like a driver in autonomous cars as part of the hybrid system, she will have to bear the responsibility for making the morally relevant decisions that are not covered by traffic rules.
Article
The question of the capacity of artificial intelligence to make moral decisions has been a key focus of investigation in robotics for decades. This question has now become pertinent to automated vehicle technologies, as a question of understanding the capacity of artificial driving intelligence to respond to unavoidable road traffic accidents. Artificial driving intelligence will make a calculated decision that could equate to deciding who lives and who dies. In calculating such important decisions, does the driving intelligence require moral intelligence and a capacity to make informed moral decisions? Artificial driving intelligence will be determined by at very least, state laws, driving codes, and codes of conduct relating to driving behaviour and safety. Does it also need to be informed by ethical theories, human values, and human rights frameworks? If so, how can this be achieved and how can we ensure there are no moral biases in the moral decision-making algorithms? The question of moral capacity is complex and has become the ethical focal point of this technology. Research has centred on applying Philippa Foot’s famous trolley dilemma. We claim that before applications attempt to focus on moral theories, there is a necessary precedent to utilise the trolley dilemma as an ontological experiment. The trolley dilemma is succinct in identifying important ontological differences between human driving intelligence and artificial driving intelligence. In this paper, we argue that when the trolley dilemma is focused upon ontology, it has the potential to become an important elucidatory tool. It can act as a prism through which one can perceive different ontological aspects of driving intelligence and assess response decisions to unavoidable road traffic accidents. The identification of the ontological differences is integral to understanding the underlying variances that support human and artificial driving decisions. Ontologically differentiating between these two contexts allows for a more complete interrogation of the moral decision-making capacity of the artificial driving intelligence.
Article
Automated vehicle technology presents an opportunity to remake urban mobility in a way that maximizes access, efficiency, and equity. One of the roles for policymakers is to ensure that future governance of automated vehicles (AVs) achieves this. When considering governance, the current literature centers on issues related to the safe operation and deployment of AVs but has not fully considered the implications of AV ownership and ridesourcing platform data propriety on achieving the most desirable urban mobility outcomes. Specifically, the literature has not considered: a future scenario where individually owned AVs are shared when not in use; and the implications of ridesourcing platform data remaining proprietary in future. This paper analyzes why: the future of AV ownership may not be a binary choice between owning an AV/not sharing and sharing an unowned fleet, which is the current consensus in the literature; the incentives for consumers to simultaneously own an AV and share it when they are not using it could be high; the way ridesourcing platform data is collected, used, and shared could be a very influential factor for urban mobility outcomes, but its implications have not been robustly analyzed in the literature; and future scenario-building and modeling should consider the implications of widespread sharing of individually owned AVs, as well as the implications of ridesourcing platform data propriety on urban mobility outcomes. Developing a foundation for future good governance of AV ownership and ridesourcing platform data propriety should be an immediate priority for researchers, policymakers, and practitioners.
Book
The End of Driving: Transportation Systems and Public Policy Planning for Autonomous Vehicles explores both the potential of vehicle automation technology and the barriers it faces when considering coherent urban deployment. The book evaluates the case for deliberate development of automated public transportation and mobility-as-a-service as paths towards sustainable mobility, describing critical approaches to the planning and management of vehicle automation technology. It serves as a reference for understanding the full life cycle of the multi-year transportation systems planning processes, including novel regulation, planning, and acquisition tools for regional transportation.
Article
AI-based systems are “black boxes,” resulting in massive information asymmetries between the developers of such systems and consumers and policymakers. In order to bridge this information gap, this article proposes a conceptual framework for thinking about governance for AI.
Article
There are scientific and technical challenges that must be addressed in developing systems that interact with humans and work along with other agents in complex, dynamic, and uncertain environments where ethical concerns may arise. In such systems relationships between users and autonomous components will be driven as much by issues such as trust, responsibility, and acceptability, as technical ones such as planning and coordination. This paper provides a comprehensive review and classification of existing methods in machine ethics, resulting in delineation of specific challenges and issues. To address the identified challenges, we introduce a method that leverages the method of reflective equilibrium and the multi-coherence theory as a unifying constraint satisfaction framework to simultaneously assess multiple ethical principles and manage ethical conflicts in a context-sensitive manner.