Professional Documents
Culture Documents
AI - and LAw
AI - and LAw
Judicature 23
ALGORITHMS,
ARTIFICIAL
INTELLIGENCE,
AND THE LAW
BY PHILIP SALES
easier in various ways by informa- ALGORITHMS AND AI credit and to insurance can be greatly
tion technology. But the water, little An algorithm is a process or set of rules expanded, enhancing human capacity
by little, gets imperceptibly hotter to be followed in problem solving. It to take action to create prosperity and
and hotter, until we find we have gone is a structured process. It proceeds protect against risk.
past a crisis point and our lives have in logical steps. This is the essence of The use of digital solutions to deliver
changed irrevocably, in ways outside processes programmed into comput- public welfare assistance offers the
our control and for the worse, without ers. They perform functions in logical prospect of greatly reduced cost of
us even noticing. The water becomes sequence. Computers are transforma- administration as well as, in theory,
boiling and the frog is dead.1 tional in so many areas because they the possibility of diverting the savings
Often there is no one to blame. As are mechanically able to perform these into more generous benefits. It also
James Williams points out in his book functions at great speed and in relation offers the potential to tailor delivery of
Stand Out of Our Light: to huge amounts of data, well beyond assistance in a more fine-grained way
what is practicable or even possible in order to deliver resources to those
At “fault” are more often the for human beings. They give rise to a who need them most. The emergence
emergent dynamics of complex form of power that raises new chal- of online courts through use of infor-
multiagent systems rather than lenges for the law, in its traditional mation technology offers the potential
the internal decision-making roles of defining and regulating rights to improve access to justice and greatly
dynamics of a single individual. As and of finding controls for illegitimate reduce the time and cost taken to
W. Edwards Deming said, “A bad or inappropriate exercise of power. At achieve resolution of disputes.
system will beat a good person the same time, alongside its duty to More widely, people increasingly live
every time.”2 control abuse of power and abuse of their lives in fundamentally important
rights, law has a function to provide a ways online, via digital platforms. They
This aspect of the digital world and framework in which this new power find it convenient, and then increas-
its effects poses particular problems can be deployed and used effectively ingly necessary, to shop online, access
for legal analysis. for socially valuable purposes. In that vital services online, and to express
For purposes of this discussion, I sense, law should “go with the flow” themselves and connect with other
draw upon the distinction between and channel this power, rather than humans online.
algorithmic analysis, on the one hand, merely resist it. Artificial intelligence is part of
and artificial intelligence on the other. The potential efficiency gains are this brave new world. It is some-
The tech world defines an algorithm huge, across private commercial activ- thing at the stage beyond mere
as simply an automated instruction, ity and governmental, legislative, and algorithmic analytical processes. I
or a “coded recipe that gets executed judicial activity. Information technol- use the term as a shorthand for self-
when it encounters a trigger.” It can ogy provides platforms for increased directed and self-adaptive computer
be as simple as a mere “if, then” state- connectivity and speed of transacting. activity. It arises, for instance, where
ment. AI, by contrast, is a configuration So-called smart contracts are devised computer systems perform more com-
of algorithms that can self-modify and to allow self-regulation by algorithms, plex tasks that previously required
create new algorithms “in response in order to reduce the costs of con- human intelligence and the applica-
to learned inputs and data as opposed tracting and of policing the agreement. tion of on-the-spot judgment, such
to relying solely on the inputs it was Distributed ledger technology, such as as driving a car. In some cases, AI
designed to recognize as triggers.” That blockchain, can create secure prop- involves machine learning, whereby
capacity to “change, adapt and grow erty and contractual rights with much an algorithm optimizes its responses
based on new data” is AI.3 The main reduced transaction costs and reduced through experience as embodied in
substance of my article is directed to need for reliance on state enforce- large amounts of data, with limited or
algorithmic analysis. But many of my ment.4 Fintech is being devised to no human interference.6 AI can involve
comments apply also to artificial intel- allow machines to assess credit risks machines that are capable of analyzing
ligence more broadly. and insurance risks at a fraction of the situations to learn for themselves and
cost of performing such exercises by then generating answers that may not
human agents.5 In this way, access to even be foreseen or controlled by their
Published by the Bolch Judicial Institute at Duke Law. Reprinted with permission.
© 2021 Duke University School of Law. All rights reserved. JUDICATURE.DUKE.EDU
Judicature 25
Judicature 27
this shift as it changes the traditional, ingly dominated by AI, we humans ex ante design stage. We also need to
familiar ways of aligning power with become vulnerable to power outside find effective mechanisms to allow for
human interests through democratic our knowledge and control; therefore, systematic ex post review of how digi-
control by citizens, regulation by gov- he says, we should program into the tal systems are working and — without
ernment, and competition in markets. machines a capacity for mercy.18 destroying the efficiency gains they
At the same time, looking from the The republican response to the dan- offer — for ex post challenges to individ-
other end of the telescope, from the ger of power and domination, namely ual concrete decisions to correct legal
point of view of the individual receiv- of arming citizens with individual errors and ensure equity and mercy.
ing or seeking access to services, one rights, will still be valuable. But it will Precisely because algorithmic sys-
might have a sense of being subjected not be enough if the asymmetries of tems are so important in the delivery
to power that is fixed and remorseless16 knowledge and power are so great of commercial and public services,
— an infernal machine over which one that citizens are in practice unable to they must incorporate human val-
has no control, and which is immune deploy their rights effectively. So what ues and protections for fundamental
to any challenge or appeal to consider we need to look for are ways of trying human interests.19 For example, sys-
extenuating circumstances, or to any to close the gap between democratic, tems need to be checked for biases
plea for mercy. For access to digital public control and technical exper- based on gender, sexuality, class, age,
platforms and digital services in the tise to meet the problem identified and ability. As Jamie Susskind observes
private sphere, the business model is by David Runciman; ways of trying to in Future Politics,20 progress is being
usually take it or leave it: Accept access build into our digital systems a capac- made toward developing principles of
to digital platforms on their terms ity for mercy, responsiveness to human algorithmic audit. On Feb. 12, 2019, the
requiring access to your data and on need, and equity in the application of European Parliament adopted a res-
their very extensive contract terms rules to meet the problem identified olution declaring that “algorithms in
excluding their legal responsibility, by Malcolm Bull; and ways of fashion- decision-making systems should not
or be barred from participating in an ing rights that are both effective and be deployed without a prior algorith-
increasingly pervasive aspect of the suitable to protect the human interests mic impact assessment . . . .”21
human world. This may be experienced that are under threat in this new world. The question then arises, how should
as no real choice at all. The movement We are not at a stage to meet Malcolm we provide for ex ante review of code
begins to look like a reversal of Sir Bull’s challenge, and rights regimes in the public interest? One idea is to
Henry Maine’s famous progression will not be adequate. People are not follow the European Parliament’s res-
from status to contract. We seem to be being protected by the machines and olution that a government department
going back to status again. often are not capable of taking effec- intending to deploy an algorithmic
Meanwhile, access to public services tive action to protect themselves. program should conduct an impact
is being depersonalized. The individual Therefore, we need to create laws that assessment, much as it does now in
seems powerless in the face of machine require those who design and operate relation to the environmental impacts
systems and loses all dignity in being algorithmic and AI systems to consider and equality impacts in relation to
subjected to their control. The move- and protect the interests of people the introduction of policy. But gov-
ment here threatens to be from citizen who are subject to those systems. ernment may not have the technical
to consumer and then on to serf. capability to do this well, particularly
Malcolm Bull argues in On Mercy17 EVALUATING when one bears in mind that it may
that it is mercy rather than justice that TECHNOLOGICAL have contracted out the coding and
is foundational for politics. Mercy, as a SYSTEMS design of the system on the grounds
concession by the powerful to the vul- Because digital processes are more that the relevant expertise lies in the
nerable, makes rule by the powerful fixed in their operation than the human private sector. Moreover, those in
more acceptable to those on the receiv- algorithms of law and operate with the legislature who are supposed to
ing end and hence more stable. In a few immense speed at the point of appli- be scrutinizing what the government
suggestive pages at the end of the book, cation of rules, we need to focus on does are unlikely to have the necessary
under the heading “Robotic Politics,” ways of scrutinizing and questioning technical expertise either. Further, it
Bull argues that as the world is increas- the content of digital systems at the might also be said that provision needs u
Published by the Bolch Judicial Institute at Duke Law. Reprinted with permission.
© 2021 Duke University School of Law. All rights reserved. JUDICATURE.DUKE.EDU
to be made for impact assessment of that society must be prepared to More difficult is to find a way to inte-
major programs introduced in the pri- devote the resources and expertise grate ways of challenging individual
vate sector, where, again, government to perform this scrutiny to a proper decisions taken by government pro-
is unlikely to have the requisite expert standard. It will not be cheap. But the grams as they occur while preserving
capability. Because of lack of informa- impact of algorithms on our lives is so the speed and efficiency that such pro-
tion and expertise, the public cannot great that I suggest the likely cost will grams offer. It will not be possible to
be expected to perform their usual be proportionate to the risks which have judicial review in every case. I
general policing function in relation to this will protect us against. make two suggestions. First, it may be
service providers. There should also be scope for legal possible to design systems whereby if
There seems to be a strong argu- challenges to be brought regarding a service user is dissatisfied, they can
ment for creating a new agency to the adoption of algorithmic programs, refer the decision to a more detailed
scrutinize AI programs from the per- including at the ex ante stage. In fact, assessment level — a sort of “advanced
spective of the public interest, which this seems to be happening already.22 search option,” which would take a lot
would constitute a public resource for This is really no more than an extension more time for the applicant to com-
government, legislators, the courts, of the well-established jurisprudence plete but might allow for more fine-
and the public generally. It would be an on challenges to adoption of poli- grained scrutiny. Secondly, the courts
expert commission staffed by coding cies which are unlawful23 and is in line and litigants, perhaps in conjunction
technicians, with lawyers and ethicists with recent decisions on unfairness with an algorithm commission, could
to assist them. The commission could challenges to entire administrative sys- become more proactive in identifying
be given access to commercially sen- tems.24 However, the extension will cases that raise systemic issues and
sitive code on strict condition that its have procedural consequences. The marshalling them together in a com-
confidentiality is protected. However, claimant will need to secure disclosure posite procedure, by using pilot cases
it would invite representations from of the coding in issue. If it is commer- or group litigation techniques.
interested persons and groups in civil cially sensitive, the court might have to The creation of an algorithm com-
society and, to the fullest extent pos- impose confidentiality rings, as happens mission would be part of a strategy
sible, it would publish reports from its in intellectual property and competi- for meeting the first and second
reviews, to provide transparency in tion cases. And the court will have to be challenges I mentioned — (i) lack of
relation to the digital processes. educated by means of expert evidence, technical knowledge in society and (ii)
Perhaps current forms of pre- which on current adversarial mod- preservation of commercial secrecy
legislative scrutiny of Acts of els means experts on each side with in relation to code. The commission
Parliament here in the UK offer the live evidence tested by cross examina- would have the technical expertise
beginnings of an appropriate model. tion. This will be expensive and time and all the knowledge necessary to
For example, the Joint Committee on consuming, in ways that feel alien in be able to interrogate specific cod-
Human Rights scrutinizes draft legis- a judicial review context. I see no easy ing designed for specific functions. I
lation for its compatibility with human way round this, unless we create some suggest it could provide a vital social
rights and reports back to Parliament system whereby the court can refer the resource to restore agency for public
on any problems. But those introduc- code for neutral expert evaluation by institutions — to government, legisla-
ing algorithmic systems are widely an algorithm commission or an inde- tors, the courts, and civil society — by
dispersed in society and across the pendently appointed expert. supplying the expert understanding
globe, so one would need some form The ex ante measures should operate required for effective lawmaking,
of trawling mechanism to ensure that in conjunction with ex post measures. guidance, and control in relation to
important algorithms are available How well a program is working and digital systems. It would also be a way
for scrutiny by the commission. That the practical effects it is having may of addressing the third challenge — (iii)
is by no means straightforward. The only emerge after a period of oper- rigidity in the interface between law
emphasis may have to be more on ex ation. There should be scope for a and code — because the commission
post testing and audit checking of pri- systematic review of results as a check would include experts who understand
vate systems after deployment. Also, after a set time, to see if the program and can constantly remind govern-
it cannot be emphasized too strongly needs adjustment. ment, legislators, and the courts about
Published by the Bolch Judicial Institute at Duke Law. Reprinted with permission.
© 2021 Duke University School of Law. All rights reserved. JUDICATURE.DUKE.EDU
Judicature 29
islatures. Coding is structuring our encryption solutions to create virtual of fiduciary obligation in the conduct
lives more and more. No longer are the currencies free from state control. of relationships35 and an expansion of
material conditions of nature, molded However, as Karen Yeung observes, doctrines of abuse of rights, which in
by industrial society, the main ground- points of contact between these the UK currently exist only in small
ing of our existence. Law has been able currency regimes and national juris- pockets of the common law36 and stat-
to operate effectively as a management dictions will continue to exist. The ute;37 to control of unfair terms and
tool for that world. But now coding is state will not simply retreat from legal rebalancing of rights at the micro level
becoming as important as nature in control. There will still need to be ele- of individual contracts.
forming the material grounds of our ments of state regulation in relation to Third, intellectual property has
existence.29 It is devised and manipu- the risks they represent. She maps out grown in importance, and will con-
lated by humans and will reflect their three potential forms of engagement, tinue to, as economic value shifts
own prejudices and interests. Its direc- which she characterizes as (a) hostile ever-more to services and intangibles.
tion and content are inevitably political evasion (or cat and mouse), (b) efficient On one hand, a major project is likely
issues.30 We need to find effective ways alignment (or the joys of [patriarchal] to be the conceptual development of
to manage this dimension of our lives marriage), and (c) supporting novel the idea that one’s personal data is his
collectively, in the interests of all. forms of peer-to-peer coordination to or her property, which ought to be por-
reduce transactional friction associ- table for one’s own benefit and over
ADAPTING LAW ated with the legal process (or uneasy which one has rights to control its
Legal doctrine may have to adapt in coexistence).32 commercial exploitation. On the other
the increasingly digital age. Such are Second, there is the loss of individ- hand, the veto rights created by intel-
the demands of bringing expertise and uals’ control over contracting and the lectual property are likely to become
technical knowledge to bear that it is related issue of accessibility to dig- qualified, so as not to impede the inter-
not realistic to expect the common law, ital platforms. Online contracting connected and global nature of the
with its limited capacity to change and has taken old concerns about boil- digital world. Intellectual property
the slow pace at which it does so, to play erplate clauses to new extremes. To rights may be subjected to regimes
a major role.31 It may assist with adap- access digital tools, one has to click to that allow them to be overridden or
tation in the margins. But the speed of accept terms that are extremely long bought out in return for a fair pay-
change is so great and the expertise that and rarely read. Margaret Radin has ment to the property owner. They may
needs to be engaged is of such a techni- written about the deformation of con- become points creating rights of fair
cal nature that the main response must tract in the information society33 and return to encourage innovation as eco-
come in legislative form. What is more, describes what she calls “massively nomic life flows through and round
the permeability of national borders distributed boilerplate” removing ordi- them, as has happened with patent
to the flow of digital technologies is so nary remedial rights. She argues for a rights under so-called FRAND regimes.
great that there will have to be interna- new way of looking at the problem, In these regimes, as the price of being
tional cooperation to provide common involving a shift from contract to tort, part of global operating standards,
legal standards and effective cross-bor- via a law of misleading or deceptive patent holders give irrevocable uni-
der regulation. disclosure. A service provider would be lateral undertakings for the producers
liable for departures from reasonable and consumers of tech products to use
The Challenges of an expectations that are insufficiently sig- their patents on payment of a fee that
Algorithmic World nalled to the consumer. is fair, reasonable, and nondiscrimina-
In the space available I offer some The information and power asym- tory.38 It is possible that these sorts of
thoughts at a very high level of gen- metries in the digital world are so solutions may come to be imposed by
erality in relation to three areas: (1) great that we need a coherent strate- law by states operating pursuant to
commercial activity; (2) delivery of gic response along a spectrum: from international agreements.
public services; and (3) the political competition law at the macro level, The fourth topic is the use of digital
sphere. to protect against abuse of dominant techniques to reduce transaction costs
(1) Commercial Activity. First, positions;34 to rights of fair access to in policing of contracts, through smart
there is the attempt to use digital and digital platforms; to extended notions contracts that are self-executing
Published by the Bolch Judicial Institute at Duke Law. Reprinted with permission.
© 2021 Duke University School of Law. All rights reserved. JUDICATURE.DUKE.EDU
Judicature 31
LEGISLATIVE FORM. WHAT IS MORE, will still play a significant role, but
may crowd back in. Ideas of consent
Judicature 33
vitude we need to exploit these new sis of self-regulation, obesity, through a European Parliament Resolution and
avenues to keep the power of the supportive public regulation. Similarly, Report in January 2017.60 On the issue
supercharged state in check.51 In rela- in relation to the digital world, as of liability for the acts of robots and
tion to the tech companies, he argues Williams points out, it is not realistic other AIs, the resolution proposes61
for regulation to ensure transparency to expect people to “bear the burdens establishing a compulsory insurance
and structural regulation to break of impossible self-regulation, to sud- scheme, a compensation fund, and, in
up massive concentrations of power. denly become superhuman and take the case of sophisticated AIs, “a specific
Structural regulation would be aimed on the armies of industrialized persua- legal status for robots in the long run.”
at ensuring individual liberty and that sion.”57 But at the moment, it is unclear In one approach,62 sophisticated AIs
the power of the tech companies does how public regulation would work and with physical manifestations, such
not go unchecked.52 whether there would be the political as self-driving cars, could be given
Third, James Williams, in Stand Out will to impose it. legal personhood, like a company.63
of Our Light,53 identifies a further sub- Fourth, the law has an important role However, types of AI differ consider-
tle threat to democracy arising from to play in protecting the private sphere ably, and a one-size-fits-all approach is
the pervasiveness of information tech- in which individuals live their lives and unlikely to be appropriate.64 It may be
nology and the incessant claims that it in regulating surveillance. For exam- necessary to distinguish between ordi-
makes on our attention. According to ple, the case law of the European Court nary software used in appliances, for
him, the digital economy is based on of Human Rights58 and of the UK’s which a straightforward product lia-
the commercial effort to capture our Investigatory Powers Tribunal59 sets bility approach is appropriate, and that
attention. In what he calls the “Age of conditions for the exercise of surveil- used in complex AI products.65
Attention,” information abundance lance powers by intelligence agencies A contrary approach is to maintain the
produces attention scarcity. At risk is and provides an effective way of moni- traditional paradigm of treating even
not just our attention, but our capac- toring such exercise. sophisticated AIs as mere products for
ity to think deeply and dispassionately liability purposes.66 A middle way has
about issues and hence even to form The Challenges of Artificial also been proposed, in which some but
what can be regarded as a coherent Intelligence not all AIs might be given separate legal
will in relation to action. He points Some of the challenges to legal doctrine personality, depending on their degree
out that the will is the source of the in relation to AI will be extrapolations of autonomous functionability and
authority of democracy. He observes from those in relation to algorith- social need,67 but may be denied “[i]f the
that as the digital attention economy mic programming. But some will be practical and legal responsibility asso-
compromises human will, it strikes “at different in kind. At the root of these ciated with actions can be traced back
the very foundations of democracy” is the interposition of the agency of to a legal person.”68 There are concerns
and may “directly threaten not only machines between human agents and about allowing creators or operators of
individual freedom and autonomy, but events that have legal consequences. AIs to enjoy a cap on liability for the acts
also our collective ability to pursue any An much-discussed example is that of of such machines, which Jacob Turner
politics worth having.”54 a driverless car that has an accident. calls the “Robots as Liability Shields”
He argues that we must reject “the Existing legal doctrine suggests pos- objection.69 However, legal personal-
present regime of attentional serfdom” sible analogies on which a coherent ity for AIs could be used in conjunction
and instead “re-engineer our world legal regime might be based. The mer- with other legal techniques, such as
so that we can give attention to what its and demerits of each have to be ideas of vicarious liability and require-
matters.”55 That is a big and difficult compared and evaluated before final ments for compulsory insurance.70
project. As Williams says, the issue is decisions are made. We should be try- These are familiar ways of distributing
one of self-regulation, at both individ- ing to think this through now. There risk in society.
ual and collective levels.56 It seems that is already a burgeoning academic lit-
law will need to support this effort in erature in this area, engaging with CONCLUSION
some way, perhaps through some form fundamental legal ideas. Legislation Algorithms and AI present huge
of public regulation. We have made the at the EU level is beginning to come opportunities to improve the human
first steps to try to fight another cri- under consideration, stemming from condition. They also pose grave u
Published by the Bolch Judicial Institute at Duke Law. Reprinted with permission.
© 2021 Duke University School of Law. All rights reserved. JUDICATURE.DUKE.EDU
threats. These exist in relation to both order to reaffirm human agency at the
PHILIP SALES,
of the diverging futures the digital individual level and at the collective,
LORD SALES
world seems to offer: technical effi- democratic level. It needs to find points
is a justice of the
ciency and private market power for in the stream where it can intervene
Supreme Court of the
Silicon Valley, on the one hand, and and ways in which the general flow
United Kingdom. This
more authoritarian national control, as can be controlled, even if not in min-
article is based on the
exemplified by China, on the other. ute detail. Law is a vehicle to safeguard
text of the Sir Henry
The digitization of life is overwhelm- human values. The law has to provide
Brooke lecture for the British and Irish Legal
ing the boundaries of nation states and structures so that algorithms and AI
Information Institute, delivered on Nov. 12,
conventional legal categories, through are used to enhance human capaci-
2019. Lord Sales is grateful to Philippe Kuhn
the volume of information that is gath- ties, agency, and dignity, not to remove
for his research assistance and to Richard
ered and deployed and the speed and them. It has to impose its order on the
Susskind for his comments on a draft of the
impersonality of decision-making that digital world and must resist being
lecture. The views expressed and any errors
it fosters. The sense is of a flood in reduced to an irrelevance.
are the sole responsibility of Lord Sales. A
which the flow of water moves around Analyzing situations with care and
revised and updated version of the lecture
obstacles and renders them meaning- precision with respect to legal rela-
was published in the British journal Judicial
less. Information comes in streams tionships, rights, and obligations is
Review in 2020.
that cannot be digested by humans, what lawyers are trained to do. They
and decisions flow by at a rate that the have a specific form of technical exper-
court process cannot easily break up tise and a fund of knowledge about
for individual legal analysis. Law needs potential legal solutions and analogies should be engaging with the debates
to find suitable concepts and prac- which, with imagination, can be drawn about the digital world now, and as a
tical ways to structure this world in upon in this major task. Lawyers matter of urgency.
1 5 15
James Williams, Stand Out of Our Light: Freedom See Lord Hodge, Justice of the Supreme Court David Runciman, How Democracy Ends (2018).
and Resistance in the Attention Economy 93–94 of the United Kingdom, The Potential and Perils 16
This sense exists in some contexts, while in oth-
(2018). of Financial Technology: Can the Law Adapt to ers the emerging digital systems may be hugely
2 Cope?, The 1st Edinburgh FinTech Lecture (Mar.
Id. at 102. empowering, enabling far more effective access
3
Kaya Ismail, AI vs. Algorithms: What’s the 14, 2019). to a range of goods, such as, education, medical
6
Difference?, CMS Wire (Oct. 26, 2018) https:// Financial Stability Board, “Artificial Intelligence guidance and assistance, and help in under-
www.cmswire.com/information-management/ and machine learning in financial services” (Nov. standing legal entitlements. See Richard Susskind
ai-vs-algorithms-whats-the-difference/. See 1, 2017), https://www.fsb.org/2017/11/artifi- and David Susskind, The Future of the Professions:
also Stephen F. Deangelis, Artificial Intelligence: cial-intelligence-and-machine-learning-in-fi- How Technology Will Transform the Work of
How Algorithms Make Systems Smart, Wired (last nancial-service/. Human Experts (2015). Of course, what is needed
7 are legal structures which facilitate this process
updated May 25, 2018) https://www.wired.com/ Jamie Susskind refers to this effect as “force”:
insights/2014/09/artificial-intelligence-algo- algorithms which control our activity force of enhancing individuals’ agency while avoiding
rithms-2/ (“Rather than follow only explicitly certain actions upon us, and we can do no other. the possible negative side-effects which under-
programmed instructions, some computer Jamie Susskind, Future Politics: Living Together in a mine it.
17
algorithms are designed to allow computers World Transformed by Tech ch. 6 (2018). Malcolm Bull, On Mercy (2019).
to learn on their own (i.e., facilitate machine 8 18
Id. at ch. 8. Id. at 159–61.
learning).”); Artificial Intelligence Algorithms: All 9
Ben Golder, Foucault and the Politics of Rights 19
See Williams, supra note 1, at 106. The goal is “to
you need to know, Edureka (Nov. 25, 2020) https:// (2015). bring the technologies of our attention onto our
www.edureka.co/blog/artificial-intelligence- 10 side. This means aligning their goals and values
See, e.g., Philip Sales, “A Comparison of the
algorithms (“Generally, an algorithm takes some with our own. It means creating an environ-
Principle of Legality and Section 3 of the Human
input and uses mathematics and logic to pro- ment of incentives for design that leads to the
Rights Act 1998” (2009) 125 LQR 598. These
duce the output. In stark contrast, an Artificial creation of technologies that are aligned with
are but two specific examples of a much wider
Intelligence Algorithm takes a combination of our interests from the outset.”
phenomenon.
both — inputs and outputs simultaneously in 11 20
Aristotle, The Ethics of Aristotle: the Nicomachean Susskind, supra note 7, at 355.
order to ‘learn’ the data and produce outputs 21
when given new inputs.”). Ethics (J.A.K. Thomson trans., 1953) at 198–200. Resolution on a comprehensive European indus-
4 12 trial policy on artificial intelligence and robotics,
Solvej Karla Krause, Harish Natarajan, & Helen For a discussion of the possibilities, see Richard
Susskind, Online Courts and the Future of Justice Eur. Parl. Doc. A8-0019/2019, 2018/2088(INI),
Luskin Gradstein, “Distributed Ledger Technol-
Part IV (2019). (2019) https://www.europarl.europa.eu/doceo/
ogy (DLT) and Blockchain” (World Bank Group,
13
See supra note 7, at 107–10, on the ability of AI to document/TA-8-2019-0081_EN.html#ref_1_1 .
Report No. 122140, 2017), http://documents1. 22
worldbank.org/curated/en/177911513714062215/ apply standards as well as rules. See Henry McDonald, AI system for granting UK
pdf/122140-WP-PUBLIC-Distributed-Ledger- 14
Human Rights Council Res. 35/19, U.N. Doc. visas is biased, rights groups claim, The Guardian
Technology-and-Blockchain-Fintech-Notes.pdf. A/74/48037, (Oct. 18, 2019). (Oct. 29, 2019) https://www.theguardian.com/
Published by the Bolch Judicial Institute at Duke Law. Reprinted with permission.
© 2021 Duke University School of Law. All rights reserved. JUDICATURE.DUKE.EDU
Judicature 35
uk-news/2019/oct/29/ai-system-for-granting- decisions in relation to the management of the according to which “no firm is allowed a monop-
uk-visas-is-biased-rights-groups-claim. property or affairs of another). oly over each of the means of force, scrutiny,
23 36 and perception-control” and “no firm is allowed
Gillick v. West Norfolk and Wisbech Area Health See, e.g., John Murphy, Malice as an Ingredient of
Auth. [1985] Eng. Rep. 402, [1986] 1 AC 112; R Tort Liability, 79 Cambridge L.J. 355 (2019). significant control over more than one of the
(Suppiah) v. SOS for the Home Dep’t [2011] EWHC 37
See, e.g., U.K. Pub. Gen. Acts, Companies Act means of force, scrutiny, and perception control
2 (Admin), [137]; R (S and KF) v. SOS for Just. [2012] together.” Id. at 354–59.
2006 S. 994 (giving members of a company 53
EWHC 1810 (Admin), [37]. the right to complain of abuse of rights by the Williams, supra note 1.
24 54
See, e.g., R (Det. Action) v. First-tier Tribunal (Im- majority where this constitutes unfair prejudice Id. at 47.
migr. & Asylum Chamber) [2015] EWCA Civ 840; 55
to the interests of the minority). Id. at 127.
[2015] 1 WLR 5341; R (Howard League for Penal 38
Huawei Tech. Co. Ltd v. Unwired Planet Int’l Ltd 56
Id. at 20.
Reform) v. LC [2017] EWCA Civ 244; LC v. Det. Ac- [2018] EWCA Civ 2344 (under appeal to the 57
Id. at 101.
tion [2017] 4 WLR 92; Fredrick Powell, Structural Supreme Court). See also the discussion about 58
Procedural Review: An Emerging Trend in Public E.g., Liberty v. UK app. 58243/00, Eur. Ct. H.R
FRAND regimes in the communication from (2008).
Law, 22 Judicial Rev. 1 83 (2017). the Commission, the Council, and the European
25 59
For instance, in the field of digital healthcare E.g., Privacy Int’l v. SOS for Foreign and Common-
Economic and Social Committee dated 2017
systems, the International Digital Health and wealth Aff. [2018] UKIP Trib IPT 15_110_CH 2 and
(COM (2017) 712 final), referred to at para. 60 in
AI Research Collaborative was established in related judgments.
the Court of Appeal judgment. 60
October 2019 to bring together health experts, 39 Report with Recommendations to the Comm.
[2019] SGHC (I) 03 (Sing.); on appeal at Quoine Pte
tech experts and ethicists to establish common on Civil Law Rules of Robotics, Eur. Parl. Doc.
Ltd v B2C2 Ltd [2020] SGCA(I) 02) (Sing.).
standards for delivery of digital health services. 40 A8-0005/2017 (Jan. 27, 2017); Civil Law Rules on
B2C2 Ltd [2019] SGHC (I) 03 at para. 210 (Sing.). Robotics, Eur. Parl Doc. P8_TA(2017)0051 (Jan. 27,
It will have the capacity to review and critique 41
systems adopted by governments or big tech [2020] SGCA(I) 02 at paras. 98–99 (Sing.); but see 2017).
companies. [2020] SGCA(I) 02 at paras. 192–98 (Sing.) (Mance 61
Civil Law Rules on Robotics, supra note 62, at
26 IJ, J., dissenting) (finding the law has to be adapt- para. 59.
5Rights Foundation, https://5rightsfoundation.
ed to the new world of AI in a way which gave 62
com (last visited Feb. 6, 2021). See, e.g., Jiahong Chen & Paul Burgess, The
27 rise to results which reason and justice would
E.g., G20 A.I. Principles (2019), Tsubuka, boundaries of legal personhood: how spontaneous
lead one to expect; relief should be available if it
https://www.g20-insights.org/wp-content/ intelligence can problematise differences be-
would at once have been perceived by an honest
uploads/2019/07/G20-Japan-AI-Principles.pdf; tween humans, artificial intelligence, companies
and reasonable trader that some fundamental
OECD Council Recommendation on A.I., OECD/ and animals, 27 A.I. & L. 73, 73–92 (2019); see also
error had occurred).
Legal/0449 (2019) (calling for shared values of 42 Gabriel Hallevy, When Robots Kill: Artificial Intel-
See P.S. Atiyah, The Rise and Fall of Freedom of ligence under Criminal Law (2013); Shawn Bayern,
human-centredness, transparency, explainabili-
Contract, 43 Modern L. Rev. 467 (1979). See also The Implications of Modern Business-Entity Law
ty, robustness, security, safety, and accountabil-
Frederick Wilmot-Smith, Term Limits: What is a for the Regulation of Autonomous Systems, 7
ity); U.N. Secretary-General, High-Level Panel on
Term?, 39 Oxford J. of Legal Studies 705 (making European J.of Risk Reg. 297, 297–309 (2016); Shawn
Digital Cooperation report, The Age of Interde-
the point that freedom of contract is a public Bayern et al., Company Law and Autonomous
pendence (June 2019), https://www.un.org/en/
policy which can be open to challenge on policy Systems: A Blueprint for Lawyers, Entrepreneurs,
pdfs/DigitalCooperation-report-for%20web.pdf
grounds; there is no one thing that contracts and Regulators, 9 Hastings Sci. & Tech. L.J. 135,
(emphasizing multi-stakeholder coordination
necessarily are); Wilmot-Smith, Term Limits, at 135–62 (2017).
and sharing of data sets to bolster trust, policies
724 (“Rather than asking about the way courts 63
for digital inclusion and equality, review of com- The common factors being (1) physical location,
must interpret contracts, we should ask what
patibility of digital systems with human rights, (2) human creation for a purpose or function,
the law of interpretation should be. Any answer
importance of accountability and transparency; and (3) policy reasons for anchoring liability
to that question will depend upon the conditions
that report indicates that the UN’s 75th anniver- back to other natural or legal persons. Chen &
of justified legal responsibility and the values
sary in 2020 may be linked to launch of a “Global Burgess, supra note 62, at 81.
contract law should promote. Contractual schol-
Commitment for Digital Cooperation”. See 64
Id. at 74.
arship must, therefore, connect with broader
generally Anna Jobin, Marcello Ienca, & Effy Vay- 65
debates in legal and political philosophy … we Id. at 90.
ena, The global landscape of AI ethics guidelines, 1 66
are concerned with the appropriate grounds of S. M. Solaiman, Legal Personality of Robots,
Nature Machine Intelligence 9, 389–99 (2019).
28 legal obligations and, in certain cases, the use Corporations, Idols, and Chimpanzees: A Quest
OECD, Measuring the Digital Transformation — A of state power to enforce those obligations. for Legitimacy, 25 A.I. & L. 155 (2017). Solaiman
roadmap to the future (2019).
29
Conflict is inevitable.”). objects to extending the corporate model to
Simone Weil, Reflections Concerning the Causes 43
Human Rights Council Res. 35/19, supra note 14. sophisticated AIs, principally on the grounds
of Liberty and Social Oppression in Oppression 44 that this would serve the undesirable aim of ex-
Carol Harlow & Richard Rawlings, Procedural-
and Liberty (Arthur Wills and John Petrie trans.,
ism and Automation: Challenges to the Values of onerating the creators and users from liability
1958). where significant harm to humans can or has
30 Administrative Law in The Foundations and Future
See Susskind, supra note 7. of Public Law, ch. 14 (Elizabeth Fisher, Jeff King, & been caused by AIs and the inability to apply a
31
See also Lord Hodge, supra note 5. Alison Young eds., 2020). rights-duties analysis.
32 45 67
Karen Yeung, Regulation by Blockchain: the Id. at 297. Robert van den Hoven van Genderen, Legal
Emerging Battle for Supremacy between the Code 46 personhood in the age of artificially intelligent
See generally Susskind, supra note 12.
of Law and Code as Law, 82 Modern L. Rev. 207 47 robots, in Research Handbook on the Law of Artifi-
See Philip Sales, Legalism in Constitutional Law: cial Intelligence (Woodrow Barfield & Ugo Pagallo
(2019).
33 Judging in a Democracy (2018) Pub. L. 687. eds., 2018).
Margaret Radin, The Deformation of Contract 48
Runciman, supra note 15. 68
Id. at 245.
in the Information Society, 37 Oxford J. of Legal 49
Studies 505 (2017). Jürgen Habermas, Between Facts and Norms, ch. 8 69
Jacob Turner, Robot Rules: Regulating Artificial
34 (William Rehg trans., 1996); Christopher Zurn, Intelligence 191–93 (2018).
Autorité de la concurrence & Bundeskartellamt,
Deliberative Democracy and the Institutions of 70
Algorithms and Competition (Nov. 2019) (work- See Hodge, supra note 5. “The law could confer
Judicial Review 239–43 (2007); Philip Sales, The
ing paper) (on file at https://www.bundeskartel- separate legal personality on the machine by
Contribution of Legislative Drafting to the Rule of
lamt.de/SharedDocs/Publikation/EN/Berichte/ registration and require it or its owner to have
Law, 77 Cambridge L.J. 630.
Algorithms_and_Competition_Working-Paper. 50 compulsory insurance to cover its liability to
pdf?__blob=publicationFile&v=5). Susskind, supra note 7 at 346. third parties in delict (tort) or restitution.”
35 51
White v. Jones [1995] 2 AC 207, 271–72 (Browne- Id. at 347–348 and ch. 13.
52
Wilkinson, L.J., concurring) (finding fiduciary According to Susskind’s vision, the regulation
obligations are imposed on a person taking would implement a new separation of powers,