As the availability of data on almost every aspect of life, and the sophistication of machine learning (ML) techniques, has increased (Lepri, Oliver, Letouzé, Pentland, & Vinck, 2018) so have the opportunities for improving both public and private life (Floridi & Taddeo, 2016). Society has greater control than it has ever had over outcomes related to: (1) who people can become; (2) what people can do; (3) what people can achieve; and (4) how people can interact with the world (Floridi et al., 2018,). However, growing concerns about the ethical challenges posed by the increased use of ML in particular, and Artificial Intelligence (AI) more generally, threaten to put a halt to the advancement of beneficial applications, unless handled properly.
Balancing the tension between supporting innovation, so that society’s right to benefit from science is protected (Knoppers & Thorogood, 2017), and limiting the potential harms associated with poorly-designed AI (and specifically ML in this context), (summarised in figure 1) is challenging. ML algorithms are powerful socio-technical constructs (Ananny & Crawford, 2018), which raise concerns that are as much (if not more) about people as they are about code (see Table1) (Crawford & Calo, 2016). Enabling the so-called dual advantage of ‘ethical ML’—so that the opportunities are capitalised on, whilst the harms are foreseen and minimised or prevented ( Floridi et al., 2018) —requires asking difficult questions about design, development, deployment, practices, uses and users, as well as the data that fuel the whole life-cycle of algorithms (Cath, Zimmer, Lomborg, & Zevenbergen, 2018). Lessig was right all along: code is both our greatest threat and our greatest promise (Lessig & Lessig, 2006).
Table 1: Ethical concerns related to algorithmic use based on the ‘map’ created by (Mittelstadt, Allo, Taddeo, Wachter, & Floridi, 2016)
Rising to the challenge of designing ‘ethical ML’ is both essential and possible. Indeed
those that claim that it is impossible are falling foul of the is-ism fallacy where they confuse the
way things are with the way things can be (Lessig & Lessig, 2006), or indeed should be. It is
possible to design an algorithmically-enhanced society pro-ethically1 (Floridi, 2016b), so that it
protects the values, principles, and ethics that society thinks are fundamental (Floridi, 2018). This is the message that social scientists, ethicists, philosophers, policymakers, technologists, and civil
society have been delivering in a collective call for the development of appropriate governance
mechanisms (D’Agostino & Durante, 2018) that will enable society to capitalise on the
opportunities, whilst ensuring that human rights are respected (Floridi & Taddeo, 2016), and fair and ethical decision-making is maintained (Lipton, 2016).
The purpose of the following pages is to highlight the part that technologists, or ML
developers, can have in this broader conversation. Specifically, section ‘Moving from Principles to Practice’ discusses how efforts to data have been too focused on the ‘what’ of ethical AI (i.e. debates about principles and codes of conduct) and not enough on the ‘how’ of applied ethics.
The ‘Methodology’ section outlines the research planned to contribute to closing this gap
between principles and practice, through the creation of an ‘applied ethical AI typology,’ and the
methodology for its creation. Section ‘Framing the results,’ provides the theoretical framework
for interpreting the results. The ‘Discussion of initial results’ section summarises what the
typology shows about the uncertain utility of the tools and methods identified as well as their
uneven distribution. The section on ‘A way forward’ argues that there is a need for a more
coordinated effort, from multi-disciplinary researchers, innovators, policymakers, citizens,
developers and designers, to create and evaluate new tools and methodologies, in order to ensure
that there is a ‘how’ for every ‘what’ at each stage of the Machine Learning pipeline. The
penultimate section lists some of the limitations of this study. Finally, the last section, concludes
that the suggested recommendations will be challenging to achieve, but it would be imprudent
not to try.
On 22nd May 2019, the Organisation for Economic Co-operation and Development (OECD)
announced that its thirty-six member countries, along with an additional six (Argentine, Brazil,
Columbia, Costa Rica, Peru, and Romania), had formally agreed to adopt, what the OECD
claims to be the first intergovernmental standard on Artificial Intelligence (AI) (OECD, 2019).
Designed to ensure AI systems are robust, safe, fair and trustworthy, the standard consists of
five complementary value-based principles, and five implementable recommendations to
policymakers.
The values and recommendations are not new. Indeed, the OECD’s Recommendation of the
Council on Artificial Intelligence (OECD, n.d.) is only the latest among a list of more than 70
documents, published in the last three years, which make recommendations about the principles
of the ethics of AI (Spielkamp et al., 2019; Winfield, 2019). This list includes documents
produced by industry (Google2, IBM3, Microsoft4, Intel5), Government (Montreal Declaration6,
Lords Select Committee7, European Commission’s High-Level Expert Group8), and academia
(Future of Life Institute9, IEEE10, AI4People11). The hope of the authors of these documents is
that the principles put forward, can, as abstractions (Anderson & Anderson, 2018), act as
normative constraints (Turilli, 2007) on the ‘do’s’ and ‘don’ts’ of algorithmic use in society.
As (Jobin, Ienca, & Vayena, 2019) and (Floridi, 2019c) point out, this intense interest
from such a broad range of stakeholders reflects not only the need for ethical guidance, but also
the desire of those different parties to shape the ‘ethical AI’ conversation around their own
priorities. This is an issue that is not unique to debates about the components of ethical ML, but
something that the international human rights community has grappled with for decades, as
disagreements over what they are, how many there are, what they are for, as well as what duties
they impose on whom, and which values of human interests they are supposed to protect
(Arvan, 2014), have never been resolved. It is significant, therefore, that there seems to be an
emerging consensus amongst the members of the ethical ML community with regards to what
exactly ethical ML should aspire to be.
A review of 84 ethical AI documents by (Jobin et al., 2019) fund that although no single
principle featured in all of them, the themes of transparency, justice and fairness, non-
maleficence, responsibility and privacy appeared in over half. Similarly, a systematic review of the
literature on ethical technology revealed that the themes of privacy, security, autonomy, justice,
human dignity, control of technology and the balance of powers, were recurrent (Royakkers,
Timmer, Kool, & van Est, 2018). As argued by (Floridi & Cowls, 2019), taken together these
themes ‘define’ ethically-aligned ML as that which is (a) beneficial to, and respectful of, people
and the environment (beneficence); (b) robust and secure (non-maleficence); (c)respectful of
5 Intel’s recommendations for public policy principles on AI: https://blogs.intel.com/policy/2017/10/18/naveen-rao-announces-intel-ai-public-
6 The Montreal Declaration for Responsible AI: https://www.montrealdeclaration-responsibleai.com/the-declaration 7 House of Lords Select Committee on Artificial Intelligence: AI in the UK: ready, willing and able?:
8 European Commission’s Ethics Guidelines for Trustworthy AI: https://ec.europa.eu/futurium/en/ai-alliance-consultation/guidelines/1 9 Future of Life’s Asilomar AI Principles: https://futureoflife.org/ai-principles/ 10 IEEE General Principles of Ethical Autonomous and Intelligent Systems: http://alanwinfield.blogspot.com/2019/04/an-updated-round-up-
human values (autonomy); (d) fair (justice); and (e) explainable, accountable and
understandable (explicability). Given this emergent consensus in the literature, it is unsurprising that these are also the themes central to the OECD standard. What is perhaps more surprising s that this agreement around the basic principles that ethical ML should meet is no longer limited
to Europe and the Western world. Just three days after the OECD publication, the Beijing
Academy of Artificial Intelligence (BAAI), an organisation backed by the Chinese Ministry of
Science and technology and the Beijing municipal government, released its fifteen AI principles for: (a) research and development; (b) use; and (c) the Governance of AI (Knight, 2019), which when read in full, bear remarkable similarity to the common framework (see Figure 2).
Table 2: Comparison of ethical principles in recent publications demonstrating the emerging consensus of ‘what’ ethical AI should aspire to be12
This fragile13 consensus means that there is now the outline of a shared foundation upon which
one can build, and that can be used as a benchmark to communicate expectations and evaluate
deliverables. Co-design in AI would be more difficult without this common framework. It is,
therefore, a necessary building block in the creation of an environment that fosters ethical,
responsible, and beneficial ML, especially as it also indicates the possibility of a time when the
distractive risk of ethics shopping14 (Floridi, 2019) will be lessened. Yet, challenges remain,
The availability of these ‘agreed’ principles supports but does not yet bring about actual
change in the design of algorithmic systems ( Floridi, 2019a). As (Hagendorff, 2019) notes, almost
all of the guidelines that have been produced to date suggest that technical solutions exist, but
very few provide technical explanations. As a result, developers are becoming frustrated by how little help is offered by highly abstract principles when it comes to the ‘day job’ (Peters & Calvo, 2019). This is reflected in the fact that 79% of tech workers report that they would like practical resources to help them with ethical considerations (Miller & Coldicott, 2019). Without this more
practical guidance, other risks such as ‘ethics bluewashing’15 and ‘ethics shirking’16 remain
(Floridi, 2019b).
Such risks, associated with a lack of practical guidance on how to produce ethical ML,
make it clear that the ethical ML community needs to embark on the second phase of AI ethics:
translating between the ‘what’ and the ‘how.’ This is likely to be hard work. The gap between
principles and practice is large, and widened by complexity, variability, subjectivity, and lack of
standardisation, including variable interpretation of the ‘components’ of each of the ethical
principles (Alshammari & Simpson, 2017). Yet, it is not impossible if the right questions are
asked (Green, 2018; Wachter, Mittelstadt, & Floridi, 2017) and closer attention is payed to how
the design process can influence (Kroll, 2018) whether an algorithm is more or less ‘ethically-
aligned.’
The sooner we start doing this, the better. If we do not take on the challenge and
develop usable, interpretable and efficacious mechanisms (Abdul, Vermeulen, Wang, Lim, &
Kankanhalli, 2018) for closing this gap, the lack of guidance may (a) result in the costs of ethical
mistakes outweighing the benefits of ethical success (even a single critical ‘AI’ scandal could
stifle innovation): (b) undermine public acceptance of algorithmic systems; (c) reduce adoption
of algorithmic systems; and (d) ultimately create a scenario in which society incurs significant
opportunity costs (Cookson, 2018). Thus, the aim of this research project is to identify the
methods and tools already available to help developers, engineers, and designers of ML reflect
on and apply ‘ethics’ (Adamson, Havens, & Chatila, 2019) so that they may know not only what to do or not to do, but also how to do it, or avoid doing it (Alshammari & Simpson, 2017). We hope that the results of this research may be easily applicable to other branches of AI.
With the aim of identifying the methods and tools available to help developers, engineers and
designers of ML reflect on and apply ‘ethics’ in mind, the first task was to design a typology, for the very practically minded ML community (Holzinger, 2018), that would ‘match’ the tools and methods identified to the ethical principles outlined in table 2 (summarised as beneficence, non-maleficence, autonomy, justice, and explicability).
To create this typology, and inspired by (Saltz & Dewar, 2019) who produced a
framework that is meant to help data scientists consider ethical issues at each stage of a project, the ethical principles were combined with the stages of algorithmic development outlined in the
overview of the Information Commissioner’s Office (ICO) auditing framework for artificial
intelligence and its core components17, as shown in table 3. The intention is that this encourages ML developers to go between decision and ethical principles regularly.
Table 3: ‘Applied AI Ethics’ Typology comprising ethical principles and the stages of algorithmic development
The second task was to identify the tools and methods, and the companies or individuals
researching and producing them, to fill the typology. There were a number of different ways this
could have been done. For example, (Vakkuri, Kemell, Kultanen, Siponen, & Abrahamsson,
2019) sought to answer the question ‘what practices, tools or methods, if any, do industry
professionals utilise to implement ethics in to AI design and development?’ by conducting
interviews at five companies that develop AI systems in different fields. However, whilst analysis of the interviews revealed that the developers were aware of the potential importance of ethics in
AI, the companies seemed to provide them with no tools or methods for implementing ethics.
Based on a hypothesis that these findings did not imply the non-existence of applied-ethics tools and methods, but rather a lack of progress in the translation of available tools and methods from
academic literature or early-stage development and research, to real-life use, this study used the
traditional approach of providing an overarching assessment of a research topic, namely a
literature review (Abdul et al., 2018).
Scopus18, arXiv19 and PhilPapers20, as well as Google search were searched. The Scopus,
arXiv and Google Search searches were conducted using the terms outlined in table 4. The
PhilPapers search was unstructured, given the nature of the platform, and instead the categories
also shown in table 4 were reviewed. The original searches were run in February 2019, but
weekly alerts were set for all searches and reviewed up until mid-July 2019. Every result (of
which there were originally over 1,000) was checked for relevance – either in terms of theoretical
framing or in terms of the use of the tool – actionability by ML developers, and generalisability
across industry sectors. In total, 425 sources21 were reviewed. They provide a practical or
theoretical contribution to the answer of the question: ‘how to develop an ethical algorithmic
system.22’
Table 4 showing the search terms used to search Scopes, arXiv and Google and the categories reviewed on PhilPapers
Scopus is the largest abstract and citation database of peer-reviewed literature: scientific journals, books and conference proceedings: https://www.scopus.com/home.uri
The third, and final task, was to review the recommendations, theories, methodologies,
and tools outlined in the reviewed sources, and identify where they may fit in the typology. To
do this, each of the high-level principles (beneficence, non-maleficence, autonomy, justice and
explicability) were translated into tangible system requirements that reflect the meaning of the
principles. This is the approach taken by the EU’s High Level Ethics Group for AI and outlined in Chapter II of Ethics Guidelines for Trustworthy AI: Realising Trustworthy AI which “offers guidance on the implementation and realisation of Trustworthy AI, via a list of (seven) requirements that should be met, building on the principles” (p.35 European Commission, 2019).
This approach is also used in the disciplinary ethical guidance produce for internet-
mediated researchers by the Belmont Report (Anabo, Elexpuru-Albizuri, & Villardón-Gallego,
2019), and by (La Fors, Custers, & Keymolen, 2019) who sought to integrate existing design-
based ethical approaches for new technologies by matching lists of values the practical
abstraction from mid-level ethics (principles) to what (Hagendorff, 2019) calls ‘microethics.’ This
translation is a process that gradually reduces the indeterminacy of abstract norms to produce
desiderata for a ‘minimum-viable-ethical-(ML)product’ (MVEP) that can be used by people who
have various disciplinary backgrounds, interests and priorities (Jacobs & Huldtgren, 2018). The
outcome of this translation process is shown in Figure 5.
Table 5 showing the connection between high-level ethical principles and tangible system requirements as adapted from the methodology outlined in Chapter II of the European Commission’s “Ethics Guidelines for Trustworthy AI”
The full typology is available here [ redacted for anonymity ]. The purpose of presenting
it is not to imply that it is ‘complete,’ nor that the tools and methodologies highlighted are the
best, or indeed the only, means of ‘solving’ each of the individual ethical problems. How to
apply ethics to the development of ML is an open question that can be solved in a multitude of
different ways at different scales and in different contexts (Floridi, 2019a). It would, for example, be entirely possible to complete the process using a different set of principles and requirements. Instead, the goal is to provide a synthesis of what tools are currently available to ML developers to encourage the progression of ethical AI from principles to practice and to signal clearly, to the ‘ethical AI’ community at large, where further work is needed.
Additionally, the purpose of presenting the typology is not to give the impression that
the tools act as means of translating the principles into definitive ‘rules’ that technology
developers should adhere to, or that developers must always complete one ‘task’ from each of
the boxed. This only promotes ethics by ‘tick-box’ (Hagendorff, 2019). Instead, the typology is
intended to eventually be an online searchable database so that developers can look for the
appropriate tools and methodologies for their given context, and use them to enable a shift from a prescriptive ‘ethics-by-design’ approach to a dialogic, pro-ethical design approach (Anabo et al., 2019; Floridi, 2019a).
In this sense, the tools and methodologies represent a pragmatic version of Habermas’s discourse ethics23 (Mingers & Walsham, 2010). In his theory, Habermas (1983, 1991) argues that
morals and norms are not ‘set’ in a top-down fashion but emerge from a process where those
with opposing views, engage in a process where they rationally consider each other’s arguments,
give reasons for their position and, based upon the greater understanding that results, reassess
their position until all parties involved reach a universally agreeable decision (Buhmann,
Paßmann, & Fieseler, 2019). This is an approach commonly used in both business and
operational research ethics, where questions of ‘what should we do?’ (as opposed to what can we do?) arise (Buhmann et al., 2019; Mingers, 2011). This is a rationalisation process that involves a fair consideration of the practical, the good and the just, and normally relies heavily on language (discussion), for both the emergence of agreed upon norms or standards, and their reproduction.
In the present scenario of developers rationalising ML design decisions to ensure that they are
ethically-optimised, the tools and methods in the typology replace the role of language and act as
the medium for identifying, checking, creating and re-examining ideas and giving fair
consideration to differing interests, values and norms (Heath, 2014; Yetim, 2019). For example,
the data nutrition tool (Holland, Hosny, Newman, Joseph, & Chmielinski, 2018) provides a
means of prompting a discussion and re-evaluation of the ethical implications of using a specific
dataset for an ML development project, and the audit methodologies of (Diakopoulos, 2015)
ensure that external voices, who may have an opposing view as to whether or not an ML-system in use is ethically-aligned, have a mechanism for questioning the rational of design decisions and
requesting their change if necessary. It is within this frame that we present an overview of our
findings in the next section.
Interpretation of the results of the literature review and the resulting typology are likely to be
context specific. Those with different disciplinary backgrounds (engineering, moral philosophy, sociology etc.) will see different patterns, and different meanings in these patterns. This kind of multidisciplinary reflection on what the presence or absence of different tools and methods, and their function, might mean is to be encouraged. To start the conversation, this section highlights the following three headings:
1. an overreliance on ‘explicability’;
2. a focus on the need to ‘protect’ the individual over the collective; and
They are interrelated, but for the sake of simplicity, let us analyse each separately.
Explicability as the all-encompassing principle
To start with the most obvious observation: the availability of tools and methods is not evenly
distributed across the typology, either in terms of the ethical principles or in terms of the stages
of development. For example, whilst a developer looking to ensure their ML algorithm is ‘non-
maleficent’ has a section of tools available to them for each development stage – as highlighted
in table 6 – the tools and methods designed to enable developers to meet the principle or
‘beneficence’ are almost all intended to be used during the initial planning stages of development
(i.e. business and use-case development design phases). However, the most noticeable ‘skew’ is
towards post-hoc ‘explanations;’ with those seeking to meet the principle of explicability during the testing phase having the greatest range of tools and methods from which to choose.
There are likely to be several reasons for this, but two stand out. The first and simpler is
that the ‘problem’ of ‘interpreting’ an algorithmic decision seems tractable from a mathematical
standpoint, so the principle of explicability has come to be seen as the most suitable for a
technical fix (Hagendorff, 2019). The second is that ‘explicability’ is not, from a moral
philosophy perspective, a moral principle like the other four principles. Instead, it can be seen as
a second order principle, that has come to be of vital importance in the ethical-ML community
because, to a certain extent, it is linked with all the other four principles24. Indeed, it is argued
that if a system is explicable (explainable and interpretable) it is inherently more transparent and
therefore more accountable in terms of its decision-making properties and the extent to which
they include human oversight and are fair, robust and justifiable (Binns et al., 2018; Cath, 2018;
Lipton, 2016).
Assuming temporarily that this is indeed the case25, and that by dint of being explicable
an ML system can more easily meet the principles of beneficence, non-maleficence, autonomy
and justice, then the fact that the ethical ML community has focused so extensively on
developing tools for ‘explanations’ may not seem problematic. However, as the majority of tools
and methods that sit in the concentration at the intersection of explicability and testing are
primarily statistical in nature, this would be a very mechanistic view because such ‘solutions’ -
e.g. LIME (Ribeiro, Singh, & Guestrin, 2016), SHAP (Lundberg & Lee, 2017), Sensitivity
Analysis (Oxborough et al., 2018) – do not really succeed in helping developers provide
meaningful explanations (Edwards & Veale, 2018) that give individuals greater control over what
is being inferred about them from their data. As such, the existence of these tools is at most
necessary but not sufficient.
From a more humanistic, and realistic perspective, in order to satisfy all the five
principles a system needs to be designed from the very beginning to be a transparent
sociotechnical system (Ananny & Crawford, 2018). To achieve this level of transparency,
accountability or explicability, it is essential that those analysing a system are able to “understand
what it was designed to do, how it was designed to do that, and why it was designed in that
particular way instead of some other way” (Kroll, 2018, p4). This kind of scrutiny will only be
possible through a combination of tools or processes that facilitate auditing, transparent
development, education of the public, and social awareness of developers (Burrell, 2016). As
such, there should ideally be tools and methods available for each of the boxes in the typology,
accepting that there may be areas of the typology which are more significant for ML
practitioners than others.
Furthermore, available of tools and methods in a variety of typology areas is also
important in the context of culturally and contextually specific ML ethics. Not all of the
principles will be of equal importance in all contexts. For example, in the case of national
security systems non-maleficence may be of considerably higher importance than explicability. If the community prioritises the development of tools and methods for one of the principles over the others, it will be denying itself the opportunity for such flexibility.
Table 6: Applied AI ethics typology with illustrative non-maleficence example. A developer looking to ensure their ML solutions meets the principle of non-maleficence can start with the foundational principles of privacy by design (Cavoukian et al., 2010) to guide ideation appropriately, use techniques such as data minimisation (Antignac et al., 2016), training for adversarial robustness (Kolter & Madry, 2018), and decision-making verification (Dennis et al., 2016) in the train-build-test phases, and end by
An individual focus
The next observation of note is that few of the available tools surveyed provide meaningful ways to assess, and respond to, the impact that the data-processing involved in their ML algorithm has
on an individual, and even less on the impact on society as a whole (Poursabzi-Sangdeh,
Goldstein, Hofman, Vaughan, & Wallach, 2018). This is evident from the very sparsely
populated ‘deployment’ column of the typology. Its emptiness implies that the need for pro-
ethically designed human-computer interaction (at an individual level) or networks of ML
systems (at a group level) has been paid little heed. This is likely because it is difficult to translate complex human behaviour into design tools that are simple to use and generalisable.
This might not seem particularly importance, but the impact this has on the overall
acceptance of AI in society could be significant. For example, it is unlikely that counterfactual
explanations26 (i.e. if input variable x had been different, the output variable y would have been
different as well) will do anything to improve the interpretability of recommendations made by
black-box systems for the average member of the public or the technical community. If such
methods become the de facto means of providing ‘explanations,’ the extent to which the
‘algorithmic society’ is interpretable to the general public will be very limited. And counterfactual
explanations could easily be embraced by actors uninterested in providing factual explanations,
because the counterfactual ones provide a vast menu of options, which may easily decrease the
level of responsibility of the actor choosing it. For example, if a mortgage provider does not
offer a mortgage, the factual reasons may be a bias, for example the gender of the applicant, but the provider could choose from a vast menu of innocuous, counterfactual explanations – if some
variable x had been different the mortgage might have been provided – e.g., a much higher
income, more collaterals, lower amount, and so forth, without ever mentioning the factual cause,
i.e. the gender of the applicant. All this could considerably limit the level of trust people are
willing to place in such systems.
This potential threat to trust is further heightened by the fact that the lack of attention
paid to impact means that ML developers are currently hampered in their ability to develop
systems that promote user’s (individual or group’s) autonomy. For example, currently there is an assumption that prediction = decision, and little research has been done (in the context of ML)
on how people translate predictions into actionable decisions. As such, tools that, for example,
help developers pro-ethically design solutions that do not overly restrict the user’s options in
acting on this prediction (i.e. tools that promote the user’s autonomy) are in short supply
(Kleinberg, Lakkaraju, Leskovec, Ludwig, & Mullainathan, 2017). F users feel as though their
decisions are being curtailed and controlled by systems that they do not understand, it is very
unlikely that these systems will meet the condition of social acceptability, never mind the
condition of social preferability which should be the aim for truly ethically designed ML (Floridi & Taddeo, 2016).
A lack of usability
Finally, the tools and methods included in the typology are positioned as discourse aids, designed
to facilitate and document rational decisions about trade-offs in the design process that may
make an ML system more or less ethically-aligned. It is possible to see the potential for the tools
identified to play this role. For example, at the “beneficence use-case
design” intersection, there are a number of tools highlighted to help elicit social values. These include the responsible
research and innovation methodology employed by the European Commission’s Human Brain
Project (Stahl & Wright, 2018), the field guide to human-centred design (idea.org) and Involve
and DeepMind’s guidance on stimulating effective public engagement on the ethics of artificial
intelligence (Involve & DeepMind). Such tools and methods could be used to help designers
pro-ethically deal with value pluralism (i.e. variation in values across different population
groups). However, the vast majority of these tools and methods are not actionable as they offer
little help on how to use them in practice (Vakkuri et al., 2019). Even when there are open-
source code libraries available, documentation is often limited, and the skill-level required for use is high.
This overarching lack of usability of the tools and methods highlighted in the typology
means that, although they are promising, they require more work before being ‘production-
ready.’ As a result, applying ethics still requires considerable amounts of effort on behalf of the
ML developers undermining one of the main aims of developing and using technologically-based
‘tools’: to remove friction from applied ethics. Furthermore, until these tools are embedded in
practice and tested in the ‘real world,’ it is extremely unclear what impact they will have on the
overall ‘governability’ of the algorithmic ecosystem. For example, (Binns, 2018a) asks how an
accountable system actually will be held accountable for an ‘unfair’ decision in a way that is
acceptable to all. This makes it almost impossible to measure the impact, ‘define success’, and
document the performance (Mitchell et al., 2019) of a new design methodology or tool. As a
result, tehre is no clear problem statement (and therefore no clear business case) that the ML
community can use to justify time and financial investment in developing much-needed tools
and techniques that truly enable pro-ethical design. Consequently, there is no guaranatee that the
so-called discursive devices do anything other than help the groups in society who already have
the loudest voices embed and protect their values in design tools, and then into the resultant ML systems.
Social scientists (Matzner, 2014) and political philosophers (from Rousseau and Kant, to Rawls
and Habermas) (Binns, 2018b), are used to dealing with the kind of plurality and subjectivity
informing the entire ethical ML field (Bibal & Frénay, 2016). Answering questions such as, what
happens when individual level and group level ‘ethics’ interact, and what key terms such as
‘fairness,’ ‘accountability,’ ‘transparency’ and ‘interpretability’ actually mean when there are
currently a myriad definitions (Ananny & Crawford, 2018; Bibal & Frénay, 2016; Doshi-Velez & Kim, 2017; Friedler, Scheidegger, & Venkatasubramanian, 2016; Guidotti et al., 2018; Kleinberg, Mullainathan, & Raghavan, 2016; Overdorf, Kulynych, Balsa, Troncoso, & Gürses, 2018; Turilli & Floridi, 2009) is standard fare for individuals with social science, economy, philosophy or legal
training. This is why (Nissenbaum, 2004) argues for a contextual account of privacy, one that
recognises the varying nature of informational norms (Matzner, 2014) and (Kemper & Kolkman, 2018) state that transparency is only meaningful in the context of a defined critical audience.
The ML developer community, in contrast, may be less used to dealing with this kind of difficulty, and more used to scenarios where there is at lest a seemingly quantifiable relationship
between input and output. As a result, the existing approaches to designing and programming
ethical ML fail to resolve what (Arvan, 2018) terms the moral-semantic trilemma, as almost all
tools and methods highlighted in the typology are either too semantically strict, too semantically flexible, or overly unpredictable (Arvan, 2018).
Bridging together multi-disciplinary researchers into the development process of pro-
ethical design tools and methodologies will be essential. A multi-disciplinary approach will help
the ethical ML community overcome obstacles concerning social complexity, embrace
algorithms can have unfair consequences; (4) algorithmic predictions can be hard to interpret
(Vaughan & Wallach, 2016); (5) trade-offs are usually inevitable; and (6) positive, ethical features
are open to progressive increase, that is an algorithm can be increasingly fair, and fairer than
another algorithm or a previous version, but makes no sense to say that it is fair or unfair n
absolute terms (compare this to the case of speed: it makes sense to say that an object is moving
quickly, or that it is fast or faster than another, but not that it is fast). The resulting
collaborations are likely to be highly beneficial for the development of applied ethical tools and methodologies for at least three reasons.
First, it will help ensure that the tools and methods developed do not only protect value-
pluralism in silico (i.e. the pluralistic values of developers) but also in society. Embracing
uncertainty and disciplinary diversity will naturally encourage ML experts to develop tools that
facilitate more probing and open (i.e. philosophical) questions (Floridi, 2019b) that will lead to
more nuanced and reasoned answers and hence decisions about why and when certain trade-
offs, for example, between accuracy and interpretability (Goodman & Flaxman, 2017), are
justified, based on factors such as proportionality to risk (Holm, 2019).
Second, it will encourage a more flexible and reflexive approach to applied ethics that is
more in-keeping with the way ML systems are actually developed: it is not think and then code,
but rather think and code. In other words, it will accelerate the move away from the ‘move fast
and break things’ approach towards an approach of ‘make haste slowly’ (festina lente) (Floridi,
2019a).
Finally, it would also mitigate a significant risk – posed by the current sporadic
application of ethical-design tools and/or methods during different development stages – of the ethical principles having been written into the business and use-case, but coded out by the time a system gets to deployment.
To enable developers to embrace this vulnerable uncertainty, it will be important to
promote the development of tools, like DotEveryone’s agile consequence scanning event
(DotEveryone, 2019), and the Responsible Double Diamond ‘R2D2’ (Peters & Calvo, 2019) that
prompt developers to reflect on the impacts (both direct and indirect) of the solutions they are
developing on the ‘end user’, and on how these impacts can be altered by seemingly minor
design decisions at each stage of development. In other words, ML developers should regularly:
now? (as encouraged by Wellcome Data Lab’s agile methodology (Mikhailov, 2019); and
continue to do so? And how? By using foresight methodologies (Floridi & Strait,
Forthcoming; Taddeo & Floridi, 2018), such as AI Now’s Algorithmic Impact
Assessment Framework (Reisman, Schultz, Crawford, & Whittaker, 2018).
Taking this approach recognises that, in a digital context, ethical principles are not simply either applied or not, but regularly re-applied or applied differently, or better, or ignored as algorithmic systems are developed, deployed, configured (Ananny & Crawford, 2018) tested, revised and retuned (Arnold & Scheutz, 2018).
This approach to applied ML ethics of regular reflection and application will heavily rely
on (i) the creation of more tools – especially to fill the white spaces of the typology (for the
reasons discussed in the previous section) and (ii) acceleration of tools maturity level from
research labs into production environments. To achieve (i)-(ii), society needs to come together in communities comprised of multi-disciplinary researchers (Cath, Wachter, Mittelstadt, Taddeo, & Floridi, 2017), including innovators, policymakers, citizens, developers and designers (Taddeo & Floridi, 2018), to foster the development of: (1) common knowledge and understanding; and (2) a common goal to be achieved from the development of tools and methodologies for applied AI
ethics (Durante, 2010). These outputs will provide a reason, a mechanism, and a consensus to
coordinate the efforts behind tool development. Ultimately, this will produce better results than the current approach, which allows a ‘thousand flowers to bloom’ but fails to create tools that fill
int eh gaps (this is a typical ‘intellectual market’ failure), and may encourage competition to
produce preferable options. The opportunity that this presents is too great to be delayed, the ML research community should start collaborating now with a specific focus on:
1. the development of a common language;
2. the creation of tools that ensure people, as individuals, groups and societies, are given an
equal and meaningful opportunity to participate in the design of algorithmic solutions at
each stage of development;
improved, and what needs to be developed can be identified;
solutions (e.g. software), also in view of satisfying (2) and supporting (3);
principles at each stage of the development and how consistency was maintained
throughout the use of different tools’
balance the costs and rewards of investing in ethical AI across society, also in view of
supporting (2)-(4).
All research projects have their limitations and this one is no exception. The first is that the
research question ‘what tools and methods are available for ML developers to ‘apply’ ethics to
each stage of the ML system design’ is very broad. This lack of specificity meant that the
available literature was excessive and growing all the time, making compromises from the
perspective of practically essential. It is certain that such compromises, for example which
databases to search and the decision to restrict the tools reviewed to those that were not industry sector-specific, have resulted in us missing a large number of tools and methods that are publicly
available. Building on this, it is again, very likely that there are a number of proprietary applied
ethics tools and methods being developed by private companies for internal or consulting
purposes that we will have missed, for example the ‘suite of customisable frameworks, tools and processes’ that make up consulting firm PWC’s “Responsible AI Toolkit” (PWC, 2019).
The second limitation is related to the design of the typology itself. As (La Fors et al.,
2019) attest, the “neat theoretical distinction between different stages of technological
innovation does not always exist in practice, especially not in the development of big data
technologies.” This implies that by categorising the tools by stage of development, we might be reducing their usability as developers in different contexts might follow a different pattern or feel as though it is ‘too late’ to, for example, engage in stakeholder engagement if they have reached the ‘build’ phase of their project, whereas the reality it is never too late.
Finally, the last limitations was already mentioned and concerns the lack of clarity
regarding how the tools and methods that have been identified will improve the governability of
algorithmic systems. Exactly how to govern ML remains an open question, although it appears
that there is a growing acceptance among tech workers (in the UK at least) that government
regulation will be necessary (Miller & Coldicott, 2019). The typology can at least be seen as a
mechanism for facilitating co-regulation. Governments are increasingly setting standards and
system requirements for ethical ML, but delegating the means for meeting these to the
developers themselves (Clarke, 2019) – the tools and methods of the typology can be seen as the
means of providing evidence of compliance. In this way, the typology (and the tools and
methods it contains within) help developers take responsibility for embedding ethics in the part
of the development, deployment, and use of ML solutions that they control (Coeckelbergh,
2012). The extent to which this makes a difference is yet to be determined.
The realisation that there is a need to embed ethical considerations into the design of
computational, specifically algorithmic, artefacts is not new. Both Alan Turing and Norbert
Weiner were vocal about this in the 1940s and 1960s (Turilli, 2008). However, as the complexity
of algorithmic systems and our reliance on them increases (Cath et al., 2017), so too does the
need to be critical (Floridi, 2016a) AI governance (Cath, 2018) and design solutions. It is possible
to design things to be better (Floridi, 2017), but this will require more coordinated and
sophisticated approaches (Allen, Varner, & Zinser, 2000) to translating ethical principles into
design protocols (Turilli, 2007).
This call for increased coordination is necessary. The research has shown that there is an uneven distribution of effort across the ‘Applied AI Ethics’ typology. Furthermore, many of the
tools included are relatively immature. This makes it difficult to assess the scope of their use
(resulting in Arvan’s 2018 ‘moral-semantic trilemma’) and consequently hard to encourage heir
adoption by the practically-minded ML developers, especially when the competitive advantage of
more ethically-aligned AI is not yet clear. Taking the time to complete any of the ‘exercises’
suggested by the methods reviewed, and investing in the development of new tools or methods
that ‘complete the pipeline’, add additional work and costs to the research and development
process. Such overheads may directly conflict with short-term, commercial incentives. Indeed, a
full ethical approach to AI design, development, deployment, and use may represent a
competitive disadvantage for any single ‘first mover’. The threat that this short-termism poses to
the development of truly ethical ML is significant. Unless a longer-term and sector-wide
perspective in terms of return on investment can be encouraged – so that mechanisms are
developed to close the gap between what and how – the lack of guidance may (a) result in the
costs of ethical mistakes outweighing the benefits of ethical successes; (b) undermine public
acceptance of algorithmic systems, even to the point of a backlash (Cookson, 2018); and (c)
reduce adoption of algorithmic systems. Such a resultant lack of adoption could then turn into a
loss of confidence from investors and research funders, and undermine AI research. Lack of
incentives to develop AI ethically could turn into lack of interest in developing AI tout court. This
would not be unprecedented. One only needs to recall the dramatic reduction in funding
available for AI research following the 1973 publication of Artificial Intelligence: A General Survey
(Lighthill, 1973) and its criticism of the fact that AI research had not lived up to its over-hyped
expectations.
It this were to happen today, the opportunity costs that would be incurred by society
would be significant (Cookson, 2018). The need for ‘AI Ethics’ has arisen from the fact that
poorly designed AI systems can cause very significant harm. For example, predictive policing
tools may lead to more people of colour being arrested, jailed or physically harmed by policy
(Selbst, 2017). Likewise, the potential benefits of pro-ethically designed AI systems are
considerable. This is especially true in the field of AI for Social Good where various AI
applications are making possible socially good outcomes that were once less easily achievable,
unfeasible, or unaffordable (Cowls, King, Taddeo, & Floridi, 2019). So, there is an urgent need
to progress research in this area.
Constructive patience needs to be exercised, by society and by the ethical AI community,
because such progress on the question of ‘how’ to meet the ‘what’ will not be quick, and there
will definitely be mistakes along the way. The ML research community will have to accept this,
trust that everyone is trying to meet the same end-goal, but also accept that it is unacceptable to
delay any full commitment, when it is known how serious the consequences of doing nothing
are. Only by accepting this can society by positive about the opportunities presented by AI to be seized, whilst remaining mindful of the potential costs to be avoided (Floridi et al., 2018).
Abdul, A., Vermeulen, J., Wang, D., Lim, B. Y., & Kankanhalli, M. (2018). Trends and Trajectories for Explainable, Accountable and Intelligible Systems: An HCI Research Agenda. Proceedings of the 2018 CHI Conference on Human Factors in Computing Systems - CHI ’18, 1–18. https://doi.org/10.1145/3173574.3174156
Adamson, G., Havens, J. C., & Chatila, R. (2019). Designing a Value-Driven Future for Ethical Autonomous and Intelligent Systems. Proceedings of the IEEE, 107(3), 518–525. https://doi.org/10.1109/JPROC.2018.2884923
AI Now Institute Algorithmic Accountability Policy Toolkit. (n.d.). Retrieved from https://ainowinstitute.org/aap-toolkit.pdf
Allen, C., Varner, G., & Zinser, J. (2000). Prolegomena to any future artificial moral agent. Journal of Experimental & Theoretical Artificial Intelligence, 12(3), 251–261. https://doi.org/10.1080/09528130050111428
Alshammari, M., & Simpson, A. (2017). Towards a Principled Approach for Engineering Privacy by Design. In E. Schweighofer, H. Leitold, A. Mitrakas, & K. Rannenberg (Eds.), Privacy Technologies and Policy (Vol. 10518, pp. 161–177). https://doi.org/10.1007/978-3-319-67280-9_9
Anabo, I. F., Elexpuru-Albizuri, I., & Villardón-Gallego, L. (2019). Revisiting the Belmont Report’s ethical principles in internet-mediated research: Perspectives from disciplinary associations in the social sciences. Ethics and Information Technology, 21(2), 137–149. https://doi.org/10.1007/s10676-018-9495-z
Ananny, M., & Crawford, K. (2018). Seeing without knowing: Limitations of the transparency ideal and its application to algorithmic accountability. New Media & Society, 20(3), 973– 989. https://doi.org/10.1177/1461444816676645
Antignac, T., Sands, D., & Schneider, G. (2016). Data Minimisation: A Language-Based Approach (Long Version). ArXiv:1611.05642 [Cs]. Retrieved from http://arxiv.org/abs/1611.05642
Arnold, T., & Scheutz, M. (2018). The “big red button” is too late: An alternative model for the ethical evaluation of AI systems. Ethics and Information Technology, 20(1), 59–69. https://doi.org/10.1007/s10676-018-9447-7
Arvan, M. (2014). A Better, Dual Theory of Human Rights: A Better, Dual Theory of Human Rights. The Philosophical Forum, 45(1), 17–47. https://doi.org/10.1111/phil.12025
Arvan, M. (2018). Mental time-travel, semantic flexibility, and A.I. ethics. AI & SOCIETY. https://doi.org/10.1007/s00146-018-0848-2
Beijing AI Principles. (n.d.). Retrieved from Beijing Academy of Aritifical Intelligence website: https://www.baai.ac.cn/blog/beijing-ai-principles
Bibal, A., & Frénay, B. (2016). Interpretability of Machine Learning Models and Representations: An Introduction.
Binns, R. (2018a). Algorithmic Accountability and Public Reason. Philosophy & Technology, 31(4), 543–556. https://doi.org/10.1007/s13347-017-0263-5
Binns, R. (2018b). What Can Political Philosophy Teach Us about Algorithmic Fairness? IEEE Security & Privacy, 16(3), 73–80. https://doi.org/10.1109/MSP.2018.2701147
Binns, R., Van Kleek, M., Veale, M., Lyngs, U., Zhao, J., & Shadbolt, N. (2018). ‘It’s Reducing a Human Being to a Percentage’: Perceptions of Justice in Algorithmic Decisions. Proceedings of the 2018 CHI Conference on Human Factors in Computing Systems - CHI ’18, 1– 14. https://doi.org/10.1145/3173574.3173951
Buhmann, A., Paßmann, J., & Fieseler, C. (2019). Managing Algorithmic Accountability: Balancing Reputational Concerns, Engagement Strategies, and the Potential of Rational Discourse. Journal of Business Ethics. https://doi.org/10.1007/s10551-019-04226-4
Burrell, J. (2016). How the machine ‘thinks’: Understanding opacity in machine learning algorithms. Big Data & Society, 3(1), 205395171562251. https://doi.org/10.1177/2053951715622512
Cath, C. (2018). Governing artificial intelligence: Ethical, legal and technical opportunities and challenges. Philosophical Transactions of the Royal Society A: Mathematical,
https://doi.org/10.1098/rsta.2018.0080
Cath, C., Wachter, S., Mittelstadt, B., Taddeo, M., & Floridi, L. (2017). Artificial Intelligence and the ‘Good Society’: The US, EU, and UK approach. Science and Engineering Ethics. https://doi.org/10.1007/s11948-017-9901-7
Cath, C., Zimmer, M., Lomborg, S., & Zevenbergen, B. (2018). Association of Internet Researchers (AoIR) Roundtable Summary: Artificial Intelligence and the Good Society Workshop Proceedings. Philosophy & Technology, 31(1), 155–162. https://doi.org/10.1007/s13347-018-0304-8
Cavoukian, A., Taylor, S., & Abrams, M. E. (2010). Privacy by Design: Essential for organizational accountability and strong business practices. Identity in the Information Society, 3(2), 405–413. https://doi.org/10.1007/s12394-010-0053-z
Clarke, R. (2019). Principles and business processes for responsible AI. Computer Law and Security Review. https://doi.org/10.1016/j.clsr.2019.04.007
Coeckelbergh, M. (2012). Moral Responsibility, Technology, and Experiences of the Tragic: From Kierkegaard to Offshore Engineering. Science and Engineering Ethics, 18(1), 35–48. https://doi.org/10.1007/s11948-010-9233-3
Cookson, C. (2018, September 6). Artificial intelligence faces public backlash, warns scientist. Financial Times. Retrieved from https://www.ft.com/content/0b301152-b0f8-11e8-99ca-68cf89602132
Cowls, J., King, T., Taddeo, M., & Floridi, L. (2019). Cowls, Josh and King, Thomas and Taddeo, Mariarosaria and Floridi, Luciano, Designing AI for Social Good: Seven Essential Factors (May 15, 2019). Available at SSRN: https://ssrn.com/abstract=.
Crawford, K., & Calo, R. (2016). There is a blind spot in AI research. Nature, 538(7625), 311– 313. https://doi.org/10.1038/538311a
D’Agostino, M., & Durante, M. (2018). Introduction: The Governance of Algorithms. Philosophy & Technology, 31(4), 499–505. https://doi.org/10.1007/s13347-018-0337-z
Dennis, L. A., Fisher, M., Lincoln, N. K., Lisitsa, A., & Veres, S. M. (2016). Practical verification of decision-making in agent-based autonomous systems. Automated Software Engineering, 23(3), 305–359. https://doi.org/10.1007/s10515-014-0168-9
Diakopoulos, N. (2015). Algorithmic Accountability: Journalistic investigation of computational power structures. Digital Journalism, 3(3), 398–415. https://doi.org/10.1080/21670811.2014.976411
Doshi-Velez, F., & Kim, B. (2017). Towards A Rigorous Science of Interpretable Machine Learning. ArXiv:1702.08608 [Cs, Stat]. Retrieved from http://arxiv.org/abs/1702.08608
DotEveryone. (n.d.). The DotEveryone Consequence Scanning Agile Event. Retrieved from https://doteveryone.org.uk/project/consequence-scanning/
Dressel, J., & Farid, H. (2018). The accuracy, fairness, and limits of predicting recidivism. Science Advances, 4(1), eaao5580. https://doi.org/10.1126/sciadv.aao5580
Durante, M. (2010). What Is the Model of Trust for Multi-agent Systems? Whether or Not ETrust Applies to Autonomous Agents. Knowledge, Technology & Policy, 23(3–4), 347–366. https://doi.org/10.1007/s12130-010-9118-4
Edwards, L., & Veale, M. (2018). Enslaving the Algorithm: From a “Right to an Explanation” to a “Right to Better Decisions”? IEEE Security & Privacy, 16(3), 46–54. https://doi.org/10.1109/MSP.2018.2701152
European Commission. (2019). Ethics Guidelines for Trustworthy AI. Retrieved from https://ec.europa.eu/futurium/en/ai-alliance-consultation
Floridi, L, & Clement-Jones, T. (2019, March 20). The five principles key to any ethical framework for AI. Tech New Statesman. Retrieved from https://tech.newstatesman.com/policy/ai-ethics-framework
Floridi, L., & Strait, A. (Forthcoming). Ethical foresight analysis: What it is and why it is needed.
Floridi, L. (2016a). Faultless responsibility: On the nature and allocation of moral responsibility for distributed moral actions. Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences, 374(2083), 20160112. https://doi.org/10.1098/rsta.2016.0112
Floridi, L. (2016b). Tolerant Paternalism: Pro-ethical Design as a Resolution of the Dilemma of Toleration. Science and Engineering Ethics, 22(6), 1669–1688. https://doi.org/10.1007/s11948-015-9733-2
Floridi, L. (2017). The Logic of Design as a Conceptual Logic of Information. Minds and Machines, 27(3), 495–519. https://doi.org/10.1007/s11023-017-9438-1
Floridi, L. (2018). Soft ethics, the governance of the digital and the General Data Protection Regulation. Philosophical Transactions of the Royal Society A: Mathematical,
https://doi.org/10.1098/rsta.2018.0081
Floridi, L. (2019a). Establishing the rules for building trustworthy AI. Nature Machine Intelligence. https://doi.org/10.1038/s42256-019-0055-y
Floridi, L. (2019b). The logic of information: A theory of philosophy as conceptual design (1st edition). New York, NY: Oxford University Press.
Floridi, L. (2019c). Translating Principles into Practices of Digital Ethics: Five Risks of Being Unethical. Philosophy & Technology, s13347-019-00354–x. https://doi.org/10.1007/s13347-019-00354-x
Floridi, L, & Cowls, J. (2019). A Unified Framework of Five Principles for AI in Society. Harvard Data Science Review. https://doi.org/10.1162/99608f92.8cd550d1
Floridi, L, Cowls, J., Beltrametti, M., Chatila, R., Chazerand, P., Dignum, V., … Vayena, E. (2018). AI4People—An Ethical Framework for a Good AI Society: Opportunities, Risks, Principles, and Recommendations. Minds and Machines, 28(4), 689–707. https://doi.org/10.1007/s11023-018-9482-5
Floridi, L, & Taddeo, M. (2016). What is data ethics? Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences, 374(2083), 20160360. https://doi.org/10.1098/rsta.2016.0360
Friedler, S. A., Scheidegger, C., & Venkatasubramanian, S. (2016). On the (im)possibility of fairness. ArXiv:1609.07236 [Cs, Stat]. Retrieved from http://arxiv.org/abs/1609.07236
Goodman, B., & Flaxman, S. (2017). European Union regulations on algorithmic decision-making and a ‘right to explanation’. AI Magazine, 38(3), 50. https://doi.org/10.1609/aimag.v38i3.2741
Green, B. P. (2018). Ethical Reflections on Artificial Intelligence. Scientia et Fides, 6(2), 9. https://doi.org/10.12775/SetF.2018.015
Guidotti, R., Monreale, A., Ruggieri, S., Turini, F., Giannotti, F., & Pedreschi, D. (2018). A Survey of Methods for Explaining Black Box Models. ACM Computing Surveys, 51(5), 1– 42. https://doi.org/10.1145/3236009
Hagendorff, T. (2019). The Ethics of AI Ethics—An Evaluation of Guidelines. ArXiv:1903.03425 [Cs, Stat]. Retrieved from http://arxiv.org/abs/1903.03425
Heath, J. (2014). Rebooting discourse ethics. Philosophy and Social Criticism, 40(9), 829–866. https://doi.org/10.1177/0191453714545340
Hevelke, A., & Nida-Rümelin, J. (2015). Responsibility for Crashes of Autonomous Vehicles: An Ethical Analysis. Science and Engineering Ethics, 21(3), 619–630. https://doi.org/10.1007/s11948-014-9565-5
Holland, S., Hosny, A., Newman, S., Joseph, J., & Chmielinski, K. (2018). The Dataset Nutrition Label: A Framework To Drive Higher Data Quality Standards. ArXiv:1805.03677 [Cs]. Retrieved from http://arxiv.org/abs/1805.03677
Holm, E. A. (2019). In defense of the black box. Science, 364(6435), 26–27. https://doi.org/10.1126/science.aax0162
Holzinger, A. (2018). From Machine Learning to Explainable AI. 2018 World Symposium on Digital Intelligence for Systems and Machines (DISA), 55–66. https://doi.org/10.1109/DISA.2018.8490530
idea.org. (n.d.). The Field Guide to Human-Centered Design. Retrieved from http://www.designkit.org/resources/1
Involve, & DeepMind. (n.d.). How to stimulate effective public engagement on the ethics of Artificial Intelligence. Retrieved from https://www.involve.org.uk/sites/default/files/field/attachemnt/How%20to%20stimul ate%20effective%20public%20debate%20on%20the%20ethics%20of%20artificial%20in telligence%20.pdf
Jacobs, N., & Huldtgren, A. (2018). Why value sensitive design needs ethical commitments. Ethics and Information Technology. https://doi.org/10.1007/s10676-018-9467-3
Jobin, A., Ienca, M., & Vayena, E. (2019). Artificial Intelligence: The global landscape of ethics guidelines. ArXiv:1906.11668 [Cs]. Retrieved from http://arxiv.org/abs/1906.11668
Johansson, F. D., Shalit, U., & Sontag, D. (2016). Learning Representations for Counterfactual Inference. ArXiv:1605.03661 [Cs, Stat]. Retrieved from http://arxiv.org/abs/1605.03661
Kemper, J., & Kolkman, D. (2018). Transparent to whom? No algorithmic accountability without a critical audience. Information, Communication & Society, 1–16. https://doi.org/10.1080/1369118X.2018.1477967
Kleinberg, J., Lakkaraju, H., Leskovec, J., Ludwig, J., & Mullainathan, S. (2017). Human Decisions and Machine Predictions*. The Quarterly Journal of Economics. https://doi.org/10.1093/qje/qjx032
Kleinberg, J., Mullainathan, S., & Raghavan, M. (2016). Inherent Trade-Offs in the Fair Determination of Risk Scores. ArXiv:1609.05807 [Cs, Stat]. Retrieved from http://arxiv.org/abs/1609.05807
Knight, W. (2019). Why does Beijing suddenly care about AI ethics? MIT Technology Review. Retrieved from https://www.technologyreview.com/s/613610/why-does-china-suddenly-care-about-ai-ethics-and-privacy/
Knoppers, B. M., & Thorogood, A. M. (2017). Ethics and big data in health. Current Opinion in Systems Biology, 4, 53–57. https://doi.org/10.1016/j.coisb.2017.07.001
Kolter, Z., & Madry, A. (n.d.). Materials for tutorial Adversarial Robustness: Theory and Practice. Retrieved from https://adversarial-ml-tutorial.org/
Kroll, J. A. (2018). The fallacy of inscrutability. Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences, 376(2133), 20180084. https://doi.org/10.1098/rsta.2018.0084
La Fors, K., Custers, B., & Keymolen, E. (2019). Reassessing values for emerging big data technologies: Integrating design-based and application-based approaches. Ethics and Information Technology. https://doi.org/10.1007/s10676-019-09503-4
Lakkaraju, H., Kleinberg, J., Leskovec, J., Ludwig, J., & Mullainathan, S. (2017). The Selective Labels Problem: Evaluating Algorithmic Predictions in the Presence of Unobservables. Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining - KDD ’17, 275–284. https://doi.org/10.1145/3097983.3098066
Lepri, B., Oliver, N., Letouzé, E., Pentland, A., & Vinck, P. (2018). Fair, Transparent, and Accountable Algorithmic Decision-making Processes: The Premise, the Proposed Solutions, and the Open Challenges. Philosophy & Technology, 31(4), 611–627. https://doi.org/10.1007/s13347-017-0279-x
Lessig, L., & Lessig, L. (2006). Code (Version 2.0). New York: Basic Books.
Lighthill, J. (1973). ‘Artificial Intelligence: A General Survey’ in Artificial Intelligence: A paper symposium. Retrieved from UK Science Research Council website: http://www.chiltoncomputing.org.uk/inf/literature/reports/lighthill_report/p001.htm
Lipton, Z. C. (2016). The Mythos of Model Interpretability. ArXiv:1606.03490 [Cs, Stat]. Retrieved from http://arxiv.org/abs/1606.03490
Lundberg, S. M., & Lee, S.-I. (2017). A Unified Approach to Interpreting Model Predictions. In I. Guyon, U. V. Luxburg, S. Bengio, H. Wallach, R. Fergus, S. Vishwanathan, & R. Garnett (Eds.), Advances in Neural Information Processing Systems 30 (pp. 4765–4774). Retrieved from http://papers.nips.cc/paper/7062-a-unified-approach-to-interpreting-model-predictions.pdf
Makri, E.-L., & Lambrinoudakis, C. (2015). Privacy Principles: Towards a Common Privacy Audit Methodology. In S. Fischer-Hübner, C. Lambrinoudakis, & J. López (Eds.), Trust, Privacy and Security in Digital Business (Vol. 9264, pp. 219–234). https://doi.org/10.1007/978-3-319-22906-5_17
Matzner, T. (2014). Why privacy is not enough privacy in the context of “ubiquitous computing” and “big data”. Journal of Information, Communication and Ethics in Society, 12(2), 93–106. https://doi.org/10.1108/JICES-08-2013-0030
Mikhailov, D. (2019). A new method for ethical data science. Retrieved from Medium website: https://medium.com/wellcome-data-labs/a-new-method-for-ethical-data-science-edb59e400ae9
Miller, C., & Coldicott, R. (2019). People, Power and Technology: The Tech Workers’ View. Retrieved from Doteveryone website: https://doteveryone.org.uk/report/workersview/
Mingers, J. (2011). Ethics and OR: Operationalising discourse ethics. European Journal of Operational Research, 210(1), 114–124. https://doi.org/10.1016/j.ejor.2010.11.003
Mingers, J., & Walsham, G. (2010). Toward ethical information systems: The contribution of discourse ethics. MIS Quarterly: Management Information Systems, 34(4), 855–870. Retrieved from https://www.scopus.com/inward/record.uri?eid=2-s2.0-79953280231&partnerID=40&md5=7a63836e6f0ff04a1baa17cb9ebf76d8
Mitchell, M., Wu, S., Zaldivar, A., Barnes, P., Vasserman, L., Hutchinson, B., … Gebru, T. (2019). Model Cards for Model Reporting. Proceedings of the Conference on Fairness, Accountability, and Transparency - FAT* ’19, 220–229. https://doi.org/10.1145/3287560.3287596
Mittelstadt, B. D., Allo, P., Taddeo, M., Wachter, S., & Floridi, L. (2016). The ethics of algorithms: Mapping the debate. Big Data & Society, 3(2), 205395171667967. https://doi.org/10.1177/2053951716679679
Nissenbaum, H. (2004). Privacy as Contextual Integrity. WASHINGTON LAW REVIEW.
OECD. (2019). Forty-two countries adopt new OECD Principles on Artificial Intelligence. Retrieved from https://www.oecd.org/science/forty-two-countries-adopt-new-oecd-principles-on-artificial-intelligence.htm
OECD. (n.d.). Recommendation of the Council on Artificial Intelligence. Retrieved from https://legalinstruments.oecd.org/en/instruments/OECD-LEGAL-0449
Oetzel, M. C., & Spiekermann, S. (2014). A systematic methodology for privacy impact assessments: A design science approach. European Journal of Information Systems, 23(2), 126– 150. https://doi.org/10.1057/ejis.2013.18
Overdorf, R., Kulynych, B., Balsa, E., Troncoso, C., & Gürses, S. (2018). Questioning the assumptions behind fairness solutions. ArXiv:1811.11293 [Cs]. Retrieved from http://arxiv.org/abs/1811.11293
Oxborough, C., Cameron, E., Rao, A., Birchall, A., Townsend, A., & Westermann, C. (n.d.). Explainable AI: Driving Business Value through Greater Understanding. Retrieved from PWC website: https://www.pwc.co.uk/audit-assurance/assets/explainable-ai.pdf
Peters, D., & Calvo, R. A. (2019, May 2). Beyond principles: A process for responsible tech. Retrieved from Medium website: https://medium.com/ethics-of-digital-experience/beyond-principles-a-process-for-responsible-tech-aefc921f7317
Polykalas, S. E., & Prezerakos, G. N. (2019). When the mobile app is free, the product is your personal data. Digital Policy, Regulation and Governance, 21(2), 89–101. https://doi.org/10.1108/DPRG-11-2018-0068
Poursabzi-Sangdeh, F., Goldstein, D. G., Hofman, J. M., Vaughan, J. W., & Wallach, H. (2018). Manipulating and Measuring Model Interpretability. ArXiv:1802.07810 [Cs]. Retrieved from http://arxiv.org/abs/1802.07810
PWC. (n.d.). The PwC Responsible AI Framework. Retrieved from https://www.pwc.co.uk/services/audit-assurance/risk-assurance/services/technologyrisk/technology-risk-insights/accelerating-innovation-through-responsible-ai.html
Reisman, D., Schultz, J., Crawford, K., & Whittaker, M. (2018). Algorithmic Impact Assessments: A Practical Framework for Public Agency Accountability. Retrieved from AINow website: https://ainowinstitute.org/aiareport2018.pdf
Ribeiro, Singh, S., & Guestrin, C. (2016, August 12). Local Interpretable Model-Agnostic Explanations (LIME): An Introduction A technique to explain the predictions of any machine learning classifier. Retrieved from https://www.oreilly.com/learning/introduction-to-local-interpretable-model-agnostic-explanations-lime
Royakkers, L., Timmer, J., Kool, L., & van Est, R. (2018). Societal and ethical issues of digitization. Ethics and Information Technology, 20(2), 127–142. https://doi.org/10.1007/s10676-018-9452-x
Russell, C., Kusner, M. J., Loftus, J., & Silva, R. (2017). When Worlds Collide: Integrating Different Counterfactual Assumptions in Fairness. In I. Guyon, U. V. Luxburg, S. Bengio, H. Wallach, R. Fergus, S. Vishwanathan, & R. Garnett (Eds.), Advances in Neural Information Processing Systems 30 (pp. 6414–6423). Retrieved from http://papers.nips.cc/paper/7220-when-worlds-collide-integrating-different-counterfactual-assumptions-in-fairness.pdf
Saltz, J. S., & Dewar, N. (2019). Data science ethical considerations: A systematic literature review and proposed project framework. Ethics and Information Technology. https://doi.org/10.1007/s10676-019-09502-5
Samuel, A. L. (1960). Some Moral and Technical Consequences of Automation—A Refutation. Science, 132(3429), 741–742. https://doi.org/10.1126/science.132.3429.741
Selbst, A. D. (2017). Disparate Impact in Big Data Policing. Georgia Law Review, 52(1), 109–196. Retrieved from https://heinonline.org/HOL/P?h=hein.journals/geolr52&i=121.
Spielkamp, M., Matzat, L., Penner, K., Thummler, M., Thiel, V., Gießler, S., & Eisenhauer, A. (2019). Algorithm Watch 2019: The AI Ethics Guidelines Global Inventory. Retrieved from https://algorithmwatch.org/en/project/ai-ethics-guidelines-global-inventory/.
Stahl, B. C., & Wright, D. (2018). Ethics and Privacy in AI and Big Data: Implementing Responsible Research and Innovation. IEEE Security & Privacy, 16(3), 26–33. https://doi.org/10.1109/MSP.2018.2701164
Taddeo, M., & Floridi, L. (2018). How AI can be a force for good. Science, 361(6404), 751–752. https://doi.org/10.1126/science.aat5991
Turilli, M. (2007). Ethical protocols design. Ethics and Information Technology, 9(1), 49–62. https://doi.org/10.1007/s10676-006-9128-9
Turilli, M. (2008). Ethics and the practice of software design. In A. Briggle, P. Brey, & K. Waelbers (Eds.), Current issues in computing and philosophy. Amsterdam: IOS Press.
Turilli, M., & Floridi, L. (2009). The ethics of information transparency. Ethics and Information Technology, 11(2), 105–112. https://doi.org/10.1007/s10676-009-9187-9
Vakkuri, V., Kemell, K.-K., Kultanen, J., Siponen, M., & Abrahamsson, P. (2019). Ethically Aligned Design of Autonomous Systems: Industry viewpoint and an empirical study. ArXiv:1906.07946 [Cs]. Retrieved from http://arxiv.org/abs/1906.07946
Vaughan, J., & Wallach, H. (2016). The inescapability of Uncertainty: AI, Uncertainty, and Why You Should Vote No Matter What Predictions Say. Retrieved 4 July 2019, from Points. Data Society website: https://points.datasociety.net/uncertainty-edd5caf8981b
Wachter, S., Mittelstadt, B., & Floridi, L. (2017). Why a Right to Explanation of Automated Decision-Making Does Not Exist in the General Data Protection Regulation. International Data Privacy Law, 7(2), 76–99. https://doi.org/10.1093/idpl/ipx005
Wiener, N. (1961). Cybernetics: Or control and communication in the animal and the machine (2d ed.). New York: M.I.T. Press.
Winfield, A. (2019, April 18). An Updated Round Up of Ethical Principles of Robotics and AI. Retrieved from http://alanwinfield.blogspot.com/2019/04/an-updated-round-up-of-ethical.html
Yetim, F. (2019). Supporting and Understanding Reflection on Persuasive Technology Through a Reflection Schema. In H. Oinas-Kukkonen, K. T. Win, E. Karapanos, P. Karppinen, & E. Kyza (Eds.), Persuasive Technology: Development of Persuasive and Behavior Change Support Systems (pp. 43–51). Cham: Springer International Publishing.