International Human Rights Law and Artificial Intelligence: Applicability of International Legal Framework on Freedom of Expression to Speech Generated by Artificial Intelligence

(1)

International Human Rights Law

and

Artificial Intelligence:

Applicability of International Legal Framework on Freedom of

Expression to Speech Generated by Artificial Intelligence

Veronika Rokavec, 12141097 veronika.rokavec@student.uva.nl

Master’s in European and International Law Track: Public International Law

Supervised by prof. dr. Janne Elisabeth Nijman

Words: 12.994 including annotations, excluding, abstract and bibliography Submitted: 24 July 2020

(2)

Abstract

This research focused on speech generated by artificial intelligence (AI) and applicability of existing international legal framework for freedom of expression (FoE). In order for an entity to be a bearer of rights and duties, it must have (international) legal personality. According to non-traditional ILP theories, it is potentially possible to grant ILP to AI. Challenges arise when discussing free speech protection, since FoE is defined as a human right. However, nowadays human-robot interaction is increasingly common and communication is no longer limited to humans. AI is designed to “speak” the language we understand and to interact in a human-like manner. Throughout this work it is sustained that FoE does not merely protect the speaker but also the listener. In order to ensure that people have access to AI-speech they find worthwhile without undue State interference, some protection must be granted to AI. Heretofore, no other international standard on free speech has been developed, thus the only potential protection remains FoE. It is true that granting human rights to AI remains undesirable, yet it is analyzed that free speech protection is often focused primarily on the listeners rather than on the speakers. Therefore, this thesis proposes that if listener-focused approach is advanced, the existing FoE framework offers very little obstacles to AI-speech protection. With such an approach, dignity of the receivers remains at the center of FoE protection and sets limitations to AI-speech protection, namely only speech valuable to humans could potentially be protected. Hence, it is important to distinguish AI outputs without any informational value or of insufficient quality from those that do have them. This paper follows a line of literature assertign that society shall rather focus on AI-human relationships. In other words, to rather keep the spotlight on whether the robot appears to possess human-like characteristics while interacting with humans than on the question whether it actually possesses all necessary and/or sufficient qualities. According to this view, protection of fundamental rights of the end-users would be strengthened. All those who find AI-speech worthwhile and their interaction with AI important would likely find robot protection necessary for their own well-being. It is unpredictable if AI will be ever recognized as a bearer of rights under Article 19 ICCPR (or a comparable freedom in an alternative treaty) but not impossible: nothing in FoE inherently requires a human speaker. It rather seems that the speaker’s contribution is decisive for protection.

(3)

Acknowledgements

I wish to thank everybody whose assistance was a milestone in competition of this Master thesis.

First and foremost, I would like to express my gratitude to my thesis supervisor prof. dr. Janne Elisabeth Nijman for patient guidance, encouragement and genuine interest in my ideas. Collaboration with prof. Nijman was a really rewarding and inspiring experience combining all the best aspects of the academic year. During the last semester, prof. Nijman contributed to my work with her extensive expertise as well as with much needed support and advice.

None of this would be possible without my family, primarily without my father. Words cannot describe how thankful I am. Allow me to give my grandmother the special place she deserves. She has been nothing but supportive, from home-baked sweets to encouraging postcards, even a Skype call every now and then, she has always given me comfort and solace.

This year would have been very different without my best friend Rubén, who has spent countless days by my side while I was busy with this LLM. He was always there to share the joy, carry the load and follow the road.

Last but not least, I would like to thank all my friends and colleagues who offered a much-needed escape from studying, including the lovely staff from Smeks Lounge Bar for providing me with enough coffee to finish this thesis.

(4)

Should robots be given rights? Yes. Humanity has obligations toward our ecosystem and social system. Robots will be part of both systems. We are morally obliged to protect them, design them to protect themselves against misuse, and to be morally harmonized with humanity.

– Hussein A. Abbas, professor at the School of Engineering & IT

(5)

3.4.2CREATIVITY ... -29- 3.4.3HUMAN SPEAKER ... -30- 3.5INTERIM CONCLUSION ... -31- CHAPTER 4: AI-SPEECH ... - 33 - 4.1INTRODUCTION ... -33- 4.2.ROBOT SPEECH ... -33- 4.2.1INFORMATIONAL VALUE ... -34- 4.2.2SPEECH QUALITY ... -35- 4.2.3LIMITATIONS ... -36- 4.3INTERIM CONCLUSION ... -36-

CHAPTER 5: RECOMMENDATIONS AND CONCLUSION ... - 38 -

BIBLIOGRAPHY ... - 40 -

1BOOKS AND BOOK SECTIONS ... -40-

2JOURNAL ARTICLES ... -42-

3REPORTS ... -45-

4NEWSPAPER AND MAGAZINE ARTICLES ... -45-

5BLOG ENTRIES ... -46-

(7)

Abbreviations

AI – Artificial intelligence AO – Advisory opinion CoE – Council of Europe

CRD – Convention on the Rights of the Child ECHR – European Convention on Human Rights ECtHR – European Court of Human Rights FoE – Freedom of expression

FoI – Freedom of Information GC34 – General Comment No. 34 GPS – Global Positioning System HR – Human Rights

HRC – Human Rights Committee

ICCPR – International Covenant on Civil and Political Rights ICJ – International Court of Justice

IHRL – International Human Rights Law IL – International Law

ILP – International legal personality IO – International organization LA – Los Angeles

LP – legal person

UDHR – Universal Declaration of Human Rights

UK – United Kingdom of Great Britain and Northern Ireland UN – United Nations

UNC – United Nations Charter US – United States of America WWII – The Second World War

(8)

Table of International Documents

‘Concluding Observations of the HRC: Japan’ (HRC 2005) CCPR/C/JPN/CO/5 ‘Spotlight on: Artificial Intelligence & Freedom of Expression’ (OSCE 2020) <https://www.osce.org/representative-on-freedom-of-media/447829>

Charter of the International Military Tribunal (adopted and entered into force 8 August 1945) 82 UNTS 279 (Nuremberg Charter)

Charter of the United Nations (adopted 26 June 1945, entered into force 24 October 1945) 892 UNTS 119 (UNC)

European Convention for the Protection of Human Rights and Fundamental Freedoms (adopted 4 November 1950, entry into force 3 September 1953) ETS 5 (ECHR)

First Optional Protocol to ICCPR (adopted 16 December 1966, entry into force 23 March 1976) 99 UNTS 171

Helberger N, ‘Artificial Intelligence - Intelligent Politics: Challenges and Opportunities for Media and Democracy’ (Council of Europe 2020) Ministerial Conference

<https://rm.coe.int/cyprus-2020-ai-and-freedom-of-expression/168097fa82> HRC, ‘General Comment No. 34: Article 19’ (2011) CCPR/C/GC/34 (GC 34)

International Covenant on Civil and Political Rights (adopted 16 December 1966, entered into force 23 March 1976) 999 UNTS 171 (ICCPR),

Jakubowicz K, ‘A New Notion of Media?’ (Council of Europe 2009) <https://rm.coe.int/168048622d>

Kaye D, ‘Promotion and Protection of the Right to Freedom of Opinion and Expression’ (UNGA 2018) A/73/348

Rome Statute of the International Criminal Court (adopted 17 July 1998, entry into force 1 July 2002) 2187 UNTS 3 (Rome Statute)

Universal Declaration of Human Rights (adopted 10 December 1948) UNGA Res 217A(III) (UDHR)

(9)

Table of Cases

Abrams v United States (1919) 250 U.S 616 630 (Supreme Court) Barthold v Germany [1985] ECtHR Application No. 8734/79 Benhadj v Algeria [2007] HRC Communication No. 1173/2003

Citizens United v Federal Election Commission (2010) 558 U.S. 310 (US Supreme Court) Coleman v Australia [2006] HRC Communication No. 1157/2003

Fedotova v Russian Federation [2012] HRC Communication No. 1932/2010 Fernando v Sri Lanka [2005] HRC Communication No. 1189/2003

First Nat’l Bank of Boston v Bellotti [1978] US Supreme Court No. 76-1172 Gauthier v Canada [1999] HRC CCPR/C/65/D633/1995

Gillberg v Sweden [2012] ECtHR Application No 41723/06 Handyside v the United Kingdom [1976] ECtHR 5493/72

Khadija Ismayilova v Azerbaijan [2019] ECtHR Applications nos. 65286/13 and 57270/14 LaGrand (Germany v United States of America), Judgement [2001] ICJ Rep 2001 466 (ICJ) Legal presentation in criminal trial (Ceferin) [2008] Constitutional Court of Republic of

Slovenia Up-309/05-25

Mavlonov and Sa’di v Uzbekistan [2009] HRC Communication No. 1334/2004 Mika Miha v Equatorial Guinea [1994] HRC Communication No. 414/1990

Reparation for Injuries in nthe Service of the UN [1949] ICJ AO, 1949 ICJ Rep 174 Ross v Canada [2000] HRC Communication No. 736/1997

Shin v Republic of Korea [2004] HRC Communication No. 926/2006 Toktakunov v Kyrgyzstan [2011] HRC CCPR/C/101/D/1470/2006 Velichkin v Belarus [2005] HRC Communication No. 1022/2001

(10)

Chapter 1: Introduction

1.1 Problematique

In the last decades, humanity witnessed unparalleled technological development in artificial intelligence (AI). It can now perform numerous complex tasks long believed to be exclusively within human domain, most surprisingly including artistic expression.1_{The focus}

of this thesis will be limited to AI able to create artistic pieces, such as stories, paintings and music, with “limited to no human interaction beyond the initial programming choices.”2_This

new technology has blurred differences between humans and robots,3_{therefore it is suitable to}

start the debate about its (international) legal position. AI cannot yet act completely without human interference, however this may be likely in the not too distant future. As we increasingly interact with AI,4_{it is time to engage in the discussion regarding protection (and/or restriction)}

of these algorithmic outputs.5

This research proceeds from a preliminary claim that for AI benefiting humanity,6

regulation must be rights-based to ensure human rights (HR) protection of individuals.7

International human rights law (IHRL) defines principles for the treatment of individuals and

1_{Douglas Heaven (ed), Machines That Think: Everything You Need to Know about the Coming Age of Artificial}

Intelligence (Nicholas Brealey Publishing 2017), 1936.

2_{Matt Carlson, ‘The Robotic Reporter: Automated Journalism and the Redefinition of Labour, Compositional}

Forms, and Journalistic Authority’ (2015) 3 Digital Journalism, 416.

3_{Robot and AI are not synonyms: former focuses on robots that aim to simplify lifestyle and increase work}

productivity, the latter deals with development of software that can do the tasks for which certain level of intelligence is needed. AI is all about humanizing the technological experience. Nevertheless, the terms can be connected: when the term “robot” is used in this thesis, it refers to artificially intelligent robots, namely robots which have been endowed with AI. (Ashok Sharma, ‘Artificial Intelligence vs Robotics: All the Facts You Need to Know’ (MediumDataDrivenInvenstor, 10 September 2019).

4_{Examples range from Amazon’s Alexa, Apple’s Siri, Microsoft’s Cortana to news stories and reports produced}

by “robo-journalists”, such as Homeowners bot and Quakebot.

5_{John Frank Weaver, ‘Why Robots Deserve Free Speech Rights’ [2018] Slate.}

6_{Future of Life Institute, ‘An Open Letter: Research Priorities for Robust and Beneficial AI’}

<https://futureoflife.org/ai-open-letter/?cn-reloaded=1>.

7_{‘Participants at Global UNESCO Conference on Artificial Intelligence Urge Rights-Based Governance of AI’}

(11)

applies to every individual within jurisdiction of a state party to a HR treaty.8_Undeniably,

applicability of a HR treaty to a non-human entity is dubious, nevertheless, the only hitherto developed international standard on free speech is freedom of expression (FoE) defined in International Covenant on Civil and Political Rights (ICCPR). Therefore, until an alternative instrument on “robot (speech) rights” is proposed, FoE remains the only legal framework protecting speech generated by AI (AI-speech) and individuals’ access to it. Up to date, this debate proposed four different models to govern AI-speech.

First, AI-speech is not protected under FoE standards.9_{This gives States leeway to}

prohibit it, although this raises problem of enforceability.10_{Taking social media as an example,}

they have billions of accounts11_{and it would be logistically impossible to effectively locate}

accounts operated by AI, moreover, it is far from settled what could an effective (preventive) penalty be. What is more, FoE is not just about the speakers, but also about the listeners; 12

which means that such a ban on AI speech would hurt individuals’ rights who find AI-speech worthwhile.

Second, AI-speech was proposed to be another form of human speech, namely that of the programmer,13_{but this does not accurately capture the reality. AI has already generated}

problematic speech, the most obvious case being Tay, an AI system created by Microsoft, which was shut down within 24 hours of its launch for displaying racist tendencies14_{that the}

developer never implemented.

Third approach is that AI-speech is only protected when it represents the speech of its human programmer.15_{The problem is that this question is often impossible to answer, most}

notably as it was illustrated in 2016 US presidential campaign, due to anonymity of most of the (Twitter)bots.16_{Additionally, rarely there is a single programmer behind the development}

8_{International Covenant on Civil and Political Rights (adopted 16 December 1966, entered into force 23 March}

1976) 999 UNTS 171 [ICCPR], art. 2(1); European Convention for the Protection of Human Rights and Fundamental Freedoms (adopted 4 November 1950, entry into force 3 September 1953) ETS 5 [ECHR], art. 2(1).

9_{Generally Tim Wu, ‘Machine Speech’ (2013) 161 University of Pennsylvania Law Review.} 10_{Weaver (n5).}

11_{https://arxiv.org/pdf/2007.03604.pdf}

12_{Tim Wu, ‘Is the First Amendment Obsolete?’ (Knight First Amendment Institute at Columbia University, 1}

September 2017).

13_{Lata Nott, ‘Does the First Amendment Protect Speech Made by Artificial Intelligence?’ (Freedom Forum}

Institute, 19 September 2019).

14_{Sarah Perez, ‘Microsoft Silences Its New A.I. Bot Tay, after Twitter Users Teach It Racism’ (TechCrunch, 24}

March 2016).

15_{Weaver (n5).}

(12)

of such an advanced technology therefore, it would be impossible to track whose ideas such AI reflects.

If none of the aforementioned governance models resolves challenges posed by AI-speech, one is left with the fourth theory: AI-speech is protected by FoE.17

The claim that AI should have any human right or fundamental freedom raises considerable controversy. Opposing arguments originate from unhuman nature of AI18_and

machines being only a tool/instrument,19_property,20_{or even slaves}21_{to humans. Regarding}

AI-speech, Wu argued that free speech protection is to be afforded exclusively to human expression.22_{Similarly Bracha and Pasquale claim that AI-speech is performative, which}

means its dominant function is not expression but “doing things.”23_{But it is also undeniable}

that throughout legal history rights of any new entity has been theretofore unimaginable.24_One

cannot overlook a vast amount of information being produced, synthesized, and disseminated through increasingly sophisticated AI.25_{These systems provide helpful information about}

stock-trading, shopping, navigation or news, to name a few. Therefore, there are compelling arguments in favor of recognizing AI as a subject of international law (IL). First and foremost, entities that operate (to some degree) autonomously in (international) legal system should possess legal status of subjects rather than objects.26_{This approach would reflect the reality}

that communication is no longer only confined to discourse between humans and a considerable amount of information is generated by AI. What is more, lack of human control

17_{Generally Toni M Massaro and Helen L Norton, ‘Siri-Ously? Free Speech Rights and Artificial Intelligence’}

(2015) 110 Northwestern University Law Review; Toni M Massaro, Helen L Norton and Margot E Kaminski, ‘SIRI-OUSLY 2.0: What Artificial Intelligence Reveals About the First Amendment’ [2017] Arizona Legal Studies Discussion Paper No. 17-01; Ronald KL Collins and David M Skover, Robotica: Speech Rights and Artificial Intelligence (1st edn, Cambridge University Press 2018); Stuart Minor Benjamin, ‘Algorithms and Speech’ [2013] University of Pennsylvania Law Review.

18_{Abeba Birhane and Jelle van Dijk, ‘Robot Rights? Let’s Talk about Human Welfare Instead’, 207.}

19_{Interview with Hussein Abbas, Benjamin Kuipers and Birgit Schippers, ‘Do Robots Deserve Human Rights?’}

(5 December 2017).

20_{George Dvorsky, ‘Experts Sign Open Letter Slamming Europe’s Proposal to Recognize Robots as Legal}

Persons’ (Gizmodo, 13 March 2018).

21_{Joanna J Bryson, ‘Robots Should Be Slaves’ in Yorick Wilks (ed), Natural Language Processing, vol 8 (John}

Benjamins Publishing Company 2010).

22_{Tim Wu, ‘Free Speech for Computers?’ The New York Times (19 June 2012).}

23_{Frank Pasquale and Oren Bracha, ‘Federal Search Commission? Access, Fairness and Accountability in the}

Law of Search’ [2008] Cornell Law Review, 1148-49.

24_{Christopher D Stone, ‘Should Trees Have Standing? - Toward Legal Rights for Natural Objects’ (1972) 45}

Southern California Law Review, 453.

25_{Massaro and Norton (n17), 1170.}

26_{Janne Elisabeth Nijman, ‘Non-State Actors and the International Rule of Law: Revisiting the “realist Theory”}

of International Legal Personality’ in M Noortman and C Ryngaert (eds), Non-state actor dynamics in international law: from law-takers to law-makers (Farnham: Ashgate 2010), 4.

(13)

over machines employing deep learning,27_{the process in which the machine is “trained,” in}

comparison to traditional programming where an engineer writes step-by-step instructions for computers to follow, creates gaps in identification of the responsible agent. The most obvious problem is the “black box problem” when it is impossible to explain how the machine transforms an input into an output. Furthermore, Petrasek argues that recognizing AI-rights and responsibilities might make the programmers more responsible developing and deploying.28

People increasingly forming relationships29_{with AI might also lead to find robot protection}

necessary. Including AI-speech under FoE could potentially strengthen full realization of FoE of individuals due to protection from state censorship.30_{Advocates favoring FoE protection}

stress the nexus that algorithmic output has with daily human-robot interaction31_{and the}

importance of substantive messages generated by AI.32

It is true that currently expressive AI is relatively unsophisticated, yet dramatic advances are pursued to create the true Second Order Robotics.33_{Therefore, this thesis attempts}

to encourage the discussion whether the existing FoE framework can protect AI-speech. When information is generated by AI, FoI could be one possible source of protection. Traditionally, the focus of FoE is the speaker while FoI centers on the recipient,34_{thus the connection seems}

obvious. Both freedoms have been linked in international HR instruments,35_{yet FoI, in}

comparison to FoE, can be limited, to e.g. political context. FoI legislation has not yet been universally recognized36_{and an intrinsic right to information has not so far been included in a}

HR treaty. Relying exclusively on FoI of individuals in the context of AI-speech could amount to unfair disadvantage for those finding such speech useful. Moreover, FoE has proved to be central to the future prosperity of humanity, namely for democracy, accountability of States

27_{Massaro, Norton and Kaminski (n17), 2486.}

28_{David Petrasek, ‘Human and Non-Human Rights – Convergence or Conflict?’ (OpenGlobalRights, 10}

December 2018).

29_{Herman Tavani, ‘Can Social Robots Qualify for Moral Consideration? Reframing the Question about Robot}

Rights’ (2018) 9 Information 73, 2-3.

31_{Generally Josh Blackman, ‘What Happens If Data Is Speech?’ [2013] SSRN Electronic Journal; Benjamin}

(n17).

32_{Benjamin (n17), 1461.}

33_{Truly intelligent AI (more in 1.2) as distinguished from First Order Robotics which only appears to be}

intelligent.

34_{M McDonagh, ‘The Right to Information in International Human Rights Law’ (2013) 13 Human Rights Law}

Review 25, 29.

35_{Generally Anthony Mason, ‘The Relationship between Freedom of Expression and Freedom of Information’,}

Freedom of Expression and Freedom of Information: Essays in Honour of Sir David Williams (OUP 2000).

36_{Helen Darbishire, ‘Ten Challenges for Right to Information’ in Tarlach McGonagle and Yvonne Donders}

(eds), The United Nations and freedom of Expression and Information (Cambridge University Press 2015), 280-1.

(14)

and for individuals’ development.37_{FoE enables individuals to get as humanly possible to the}

truth,38_{meaning that enabling free speech is a precondition for emergence of quality}

information. This illustrates that FoE is not only focused on speaker’s humanness but rather on ensuring receiver’s full realization of FoE.

1.2 Defining AI

Numerous problems regarding AI originate from lack of disagreement on definition.39

Generally speaking, there are two types of AI, namely First Order Robotics40_{or weak AI}41_and

Second Order Robotics42_{or strong AI.}43_{The former encompasses AI which is driven by the}

human dictates therefore its algorithmic outputs are determined by developers’ parameters.44

Such AI only appears to be intelligent. On the other hand, strong AI is perceived as actually intelligent in the conventional human sense and generates content independently of human input.45

For the purpose of this research, AI refers to deep learning machines, a technique that uses “neural networks” inspired by the human brain.46_{Deep learning model is designed to}

analyse data with human-like logic: the software attempts to mimic the activity of the human brain.47_{Such systems require a lot of “training,” but once it is completed, it has proven to}

outperform human brain in numerous tasks. In this thesis, “robot” refers to physically embodied artificially intelligent system48_{which employs deep learning. In order for AI to raise}

37_{Sylvie Coudray, ‘UNESCO: Freedom of Expression, Information and the Media’ in Tarlach McGonagle and}

Yvonne Donders (eds), The United Nations and Freedom of Expression and Information (Cambridge University Press 2015), 208.

38_{Timothy Garton Ash, Free Speech: Ten Principles for a Connected World (Atlantic Books 2016), 119.} 39_{David J Gunkel, Robot Rights (MIT Press 2018), 597.}

40_{Collins and Skover (n17), 27.}

41_{Stuart J Russell, Peter Norvig and Ernest Davis, Artificial Intelligence: A Modern Approach (3rd ed, Prentice}

Hall 2010), 1020.

43_{Russell, Norvig and Davis (n41), 1026.} 44_{ibid, 1020-26.}

46_{Lindsey Andersen, ‘Human Rights in the Age of Artificial Intelligence’ (AccessNow 2018), 9.} 47_{Jason Brownlee, ‘What Is Deep Learning?’ (Machine Learning Mastery, 20 December 2019).} 48_{Raya A Jones, Personhood and Social Robotics: A Psychological Consideration (2018), 5.}

(15)

concerns in the realm of speech, it must be able to generate speech without human interference. To sum up, this paper focuses on “strong AI” with the ability to act as an expressing agent.

Boden argued twenty years ago that creativity poses a fundamental challenge to AI.49

Similarly, Nake and Cohen are sceptical of the possibility for the AI to be ever autonomously creative,50_{nevertheless, quite remarkable progress has been made delivering promising results.}

McCormack51_{designed a computer demonstrating artistic behaviour not explicitly}

programmed into it. Two painting masters were created: AARON, a machine who “mixes its own paint, creates striking artwork, and even washes its own brushes,”52_{and Painting Fool.}53

This year, a team of programmers proved that AI is capable of producing a “catchy” song.54

Llano is creating some of the most human-like technology:55_{what-if machines.}56_Another

example is Scheherezade, a system learning from questions.57_{When it lacks knowledge, it asks}

the internet58_{and can consequently write fiction about anything that can be found online. These}

examples illustrate how AI has managed to penetrate into creative art world.

1.3 Research Question

This research aims to answer whether the existing FoE framework can be applicable to AI with the ability to generate information. Therefore, two interrelated questions are considered. First, if AI can be a bearer of FoE and second, if it follows that AI-speech qualifies for protection under Article 19 ICCPR.

49_{Margaret A Boden, ‘Creativity and Artificial Intelligence’ (1998) 103 Artificial Intelligence, 347.}

50_{Frieder Nake, ‘Construction and Intuition: Creativity in Early Computer Art’ in Jon McCormack and Mark}

d’Inverno (eds), Computers and Creativity (Springer Verlag 2012), 61-94.

51_{Jon McCormack, ‘Creative Ecosystem’ in Jon McCormack and Mark d’Inverno (eds), Computers and}

Creativity (Springer Verlag 2012), 39-60.

52_{‘Harold Cohen - AARON’ Genetics and Culture}

<http://geneticsandculture.com/genetics_culture/pages_genetics_culture/gc_w05/cohen_h.htm>.

53_{Simon Colton, ‘The Painting Fool: Stories from Building an Automated Painter’ in Jon McCormack and}

Mark d’Inverno (eds), Computers and Creativity (Springer Verlag 2012), 3-38.

54_{‘Australia Wins Artificial Intelligence Song Contest’}

<https://www.vprobroadcast.com/titles/ai-songcontest/articles/australia-wins-ai-song-contest.htmlAustralia wins Artificial Intelligence Song Contest>.

55_{Heaven (n1), 1941.}

56_{‘WHIM - The What-If Machine’ (WHIM project)}

<http://ccg.doc.gold.ac.uk/research/whim/resources/overview_sheet.pdf>.

57_{Heaven (n1), 2005.} 58_{ibid, 2017.}

(16)

1.4 Methodology and Argument Structure

In order to understand the tensions arising from AI-speech, Chapter 2 establishes the legal framework of this research – Article 19 ICCPR – and the relationship between FoE and FoI.

This analysis uses FoE as analytical framework and considers two criteria that must be met for its protection. First, the entity seeking protection must qualify as a bearer of FoE rights and second, AI-speech must qualify as protected. The former discussion (Chapter 3) will be considered in the light of international legal personality (ILP) theories, IHRL and justifications for FoE. ILP enables entities to be bearers of rights and duties under IL. Several theories seem to offer leeway in accommodating non-traditional subjects, including individuals. Yet, special character of FoE is implied since it falls under human rights. Universal Declaration of Human Rights (UDHR) stipulates that reason and consciousness59_{are necessary and sufficient}

pre-conditions for HR protection. However, FoE is not only focused on speaker’s humanness but also on receiver’s full realization of FoE. Accordingly, FoE will be applied to AI and a conclusion drawn on its applicability to non-human entities. Putting it shortly, nexus with human-robot interaction and listener-focused approach will be in the centre of Chapter 3 arguments. In Chapter 4, informational value and quality of AI-speech will be proposed as one of the possible limitations for protection. To summarize, throughout this thesis, it will be sustained that the existing FoE framework can be applied to AI-speech.

(17)

Chapter 2: Freedom of Expression

2.1 Introduction

Technological development has far reaching consequences on how information is generated. There was never a time in human history for ideas to be accessed so easily, to illustrate, nowadays 56%60_{of humankind has access to the internet. An additional challenge}

for the society represents the development of AI,61_{which decisively shapes the information}

people receive.62_{These new developments in the “information ecosystem”}63_{raise new}

regulatory dilemmas.64_{In the light of the technological revolution, FoE became an essential}

freedom,65_{enabling realization of several other fundamental rights.}66_{While there are still no}

international norms that explicitly deal with FoE and AI, international standards on FoE were developed which are potentially applicable to AI-speech. Already legally non-binding UDHR contained FoE which was further defined in the subsequent treaty, ICCPR, defining obligations of the State parties towards individuals under the party’s jurisdiction. This Chapter describes hitherto developed FoE standards and attempts to illustrate that innovative interpretation of the existing legal framework would enhance protection of HR in general.

60_{Ying Lin, ‘10 Internet Statistics Every Marketer Should Know in 2020’ (Oberlo, 8 November 2019).} 61_{Corinne Cath and others, ‘Artificial Intelligence and the “Good Society”: The US, EU, and UK Approach’}

[2017] Science and Engineering Ethics, 10-17.

62_{‘Privacy and Freedom of Expression in the Age of Artificial Intelligence’ (2018), 2.}

63_{Tarlach McGonagle, ‘The Development of Freedom of Expression and Information within the UN: Leaps and}

Bounds or Fits and Starts?’ in Tarlach McGonagle and Yvonne Donders (eds), The United Nations and Freedom of Expression and Information (Cambridge University Press 2015), 5.

64_{Generally Karol Jakubowicz, ‘A New Notion of Media?’ (CoE 2009).} 65_{Benhadj v Algeria [2007] HRC Communication No. 1173/2003, par.8.10.}

(18)

2.2 FoE

FoE is defined in universal and regional HR treaties, e.g. UDHR, ICCPR, Convention on the Rights of the Child (CRD), European Convention on HR (ECHR), etc. Nevertheless, international bodies differ on the levels of protection afforded to different content. Whereas European Court of Human Rights (ECtHR) recongized high protection of political speech,67

the Human Rights Committee (HRC) refuses to acknowledge different levels of protection based on the speech content and, to date, follows John Stuart Mill’s approach of search for truth.68

2.2.1 UDHR and ICCPR

While Article 19 UDHR has been drafted over seventy years ago, it still expresses FoE relevant in today’s digital era:

“Everyone has the right to freedom of opinion and expression; this right includes freedom to hold opinions without interference and to seek, receive

and impart information and ideas through any media regardless of frontiers.”69

FoE is set out in more detail in Article 19(2) ICCPR which articulates it as a compound right,70_{including right to express and freedom to seek, receive and impart information.}71_The

formulation acknowledges that individuals communicate with others and therefore Article 19 includes freedom to hear others and to exchange information with them.72

67_{Handyside v the UK [1976] ECtHR 5493/72, par.49.}

68_{Nicola Wenzel, ‘Opinion and Expression, Freedom of, International Protection’, Max Planck Encyclopaedia}

of Public International Law (2014), para.6.

69_{UDHR (n59), art.19.} 70_{McGonagle (n63), 3.} 71_{ICCPR (n8), art.19(2).}

72_{Dominic McGoldrick, ‘Thought, Expression, Association, and Assembly’ in Daniel Moeckli, Sangeeta Shah}

(19)

Special Rapporteurs have further developed FoE scope. General Comment No.34 (GC34) renders “any form of idea or opinion capable of transmission to others”73_{as protected,}

e.g. political discourse,74_{commentary on one’s own}75_{and on public affairs,}76_canvassing,77

discussion of HR,78_journalism,79_{cultural and artistic expression,}80_{teaching, religious}

discourse,81_{and commercial advertising.}82_{19(2) ICCPR protects even expression that may be}

offensive,83_{nevertheless, such expression can be limited.}84_{The freedom to impart encompasses}

any expression of any kind of information through any media of choice,85_{“either orally, in}

writing or in print [or] in form of art.”86

2.2.2 FoI in 19 ICCPR

19(2) ICCPR includes FoI, namely right to seek, receive and impart information87

which was endorsed in GC34. The relationship between the two freedoms is characterized as mutual dependency: expression and information contribute to the forming of opinions. Information is the basis for expression seeking to communicate with others. Therefore, information can be understood as preceding expression. Yet, expression produces information, which leads to a more symbiotic relationship.88_{Accordingly, a question emerges whether}

AI-speech and individuals’ access to it could be sufficiently protected under FoI.

There are several differences between both freedoms. While FoE is centered around the agent generating information, FoI focuses on the receiver. HR bodies have been slow in interpreting FoE as encompassing FoI, although recent developments show willingness to

73_{GC34 (n66), par.11.}

74_{Mika Miha v Equatorial Guinea [1994] HRC Communication No. 414/1990, par.6.8.} 75_{Fernando v Sri Lanka [2005] HRC Communication No. 1189/2003, par.9.2.}

76_{Coleman v Australia [2006] HRC Communication No. 1157/2003, par.7.3.}

77_{‘Concluding Observations of the HRC: Japan’ (HRC 2005) CCPR/C/JPN/CO/5, par.26.} 78_{Velichkin v Belarus [2005] HRC Communication No. 1022/2001, par.7.2-7.3.}

79_{Mavlonov and Sa’di v Uzbekistan [2009] HRC Communication No. 1334/2004, par.8.4.} 80_{Shin v Republic of Korea [2004] HRC Communication No. 926/2006, par.7.2.}

81_{Ross v Canada [2000] HRC Communication No. 736/1997, par.11.6.} 82_{GC34 (n66), par.11.}

83_{Ross v. Canada (n81), par.11.6.} 84_{ICCPR (n8), art.19(3) and 20.} 85_{GC34 (n66), par.12.}

86_{ICCPR (n8), art.19(2).} 87_Ibid.

(20)

recognize this right, at least in certain contexts.89_{Nevertheless, while there is little}

disagreement among States on existence of FoE, several states still refuse to recognize FoI. Until 2013, 95 States worldwide have adopted FoI legislation which is less than half of UN Members.90_{Even some HR-observant States, such as Germany, have been reluctant to adopt}

information legislation.91_{Thus, it is clear that FoI is not yet universally recognized and in many}

States underdeveloped. Moreover, while FoE is not limited to neither certain situations nor content, FoI has been often protected only in limited contexts, such as information about elected State bodies,92_{statistics on death penalty}93_{or information held by public bodies.}94_Even

ECtHR, with its generally evolutive interpretation, did not unequivocally recognize FoI95_and

held that it can arise under FoE,96_{implying that it could be limited to certain situations. Scope}

of FoI as developed under FoE appears to be limited to certain information,97_{namely it shall}

meet “public interest qualitative test.”98_{The treaty bodies have so far emphasized the}

“watchdog”99_{role of those requesting access to information.}100_{Similarly, the purpose for}

which the information is sought seems to be determinative, the focus being exclusively political reasons.101

It follows that treating AI-speech as an “idea capable of transmission,”102_{awarding it}

FoE protection, is considerably different from protection arising from FoI as defined in 19 ICCPR. Undoubtedly, the idea that FoI is necessary for participation in public affairs offers a much broader basis for its protection,103_{however, information relevant for individual’s}

personal development could potentially still be excluded. Many might not have the means104

and/or the knowledge to rely on their FoI, they would have to be aware of the existence of such speech (and of AI being able to generate information). Such a regime could provide protection only to certain speech, e.g. politically motivated. If individuals had to invoke their FoI, granting

89_{McDonagh (n34), 31.}

90_{https://www.right2info.org/access-to-information-laws/access-to-information-lawsc#_ftnref7} 91_{Darbishire (n36), 280-1.}

92_{Gauthier v Canada [1999] HRC CCPR/C/65/D633/1995, para.13.4.} 93_{Toktakunov v Kyrgyzstan [2011] HRC CCPR/C/101/D/1470/2006.} 94_{GC34 (n66) par.19.}

95_{Toby Mendel, ‘Freedom of Information as an Internationally Protected Human Right’ (Article19), 2.} 96_{Gillberg v Sweden [2012] ECtHR Application No 41723/06.}

97_{Toby Mendel, Freedom of Information (UNESCO 2008), 16-20.} 98_{McDonagh (n34), 47-8.}

99_{Barthold v Germany [1985] ECtHR Application No. 8734/79, par.59.} 100_{Toktakunov v Kyrgyzstan (n97); McDonagh (n34), 46-7.}

101_{McDonagh (n34), 49.} 102_{GC34 (n66), para.11.} 103_{McDonagh (n34), 46.}

(21)

access would essentially depend on the authorities, consequently creating a burden on public administration105_{and leave considerable discretion on AI-speech protection}106_{to the national}

authorities which could be interested in censoring107_{certain content. Furthermore, AI-speech}

relevant to minorities could face censorship108_{due to the lack of access of some groups to State}

administration.109_{It remains disputed to what extent the scope of FoI extends to private actors}

and information held by them.110

Some scholars argue that there is an intrinsic right to information rather than just instrumental. Florini writes that FoI is a fundamental HR.111_{Yet, as analyzed, FoI legislation}

has not yet been universally accepted.112_{Realization of such a stand-alone right depends on the}

political will to include it in HR treaties. In the light of its absence from ICCPR, it would be hard to recognize its unequivocal existence and the humanity is likely left with FoI as developed in the scope of FoE with several limitations.

2.2.2.1 Justifications

Justifications for FoE protection offer further arguments supporting the need for inclusion of AI-speech in the scope of FoE. In the first place, they illustrate reasons for speech protection, clarify why the freedom is so fundamentally important and further advance the distinctive characteristics of FoI and FoE. There are three main diverging but complementing justifications of FoE.113

First can be traced to J.S. Mill who defined it as a tool in the search for truth: enabling everybody to express their opinion is a precondition for the emergence of truth.114_Justice

Holmes expressed the same idea in Abrams v. US: the only remedy for bad speech is more speech and, thus, the best way to test an idea is “to get itself accepted in the competition of the

106_{Steven G Gey, ‘The First Amendment and the Dissemination of Socially Worthless Untruths’ (2009) 36}

Florida State University Law Review, 21.

107_{Natali Helberger and others, ‘Implications of AI-Driven Tools in the Media for Freedom of Expression’}

(IViR 2019), 16.

108_{Collins and Skover (n17), 54.} 109_{Helberger and others (n107), 12.} 110_{Darbishire (n36), 288-9.}

111_{Generally Ann Florini (ed), The Right to Know: Transparency for an Open World (Columbia University}

Press 2007).

112_{Darbishire (n36), 280-1.}

113_{L.A. Powe, The Fourth Estate and the Constitution: Freedom of the Press in America (University of}

California Press 1992), 240.

(22)

free market.”115_{In other words, FoE enables us to get as close as humanly possible to the truth}

on a certain subject.116

The second approach to FoE analysis focuses on its link with democracy117_and

identifies it as a prerequisite for a functioning democratic system.118_{If citizens are enabled to}

access all kinds of information and ideas, they can form their own opinion on public matters,119

while expression allows their participation in public debate. It facilitates the development of a pluralistic society with access to different philosophies and ideas.120

Lastly, FoE may be explained as fundamental for enhancing individual’s personality.121

It enables individuals to develop their own identity122_{through the exposure to all kinds of}

opinions,123_{as well as improves their ability to gather relevant information for autonomous}

decision-making.124

In conclusion, FoE facilitates individuals to express their ideas and to hear ideas from others whose expression is equally protected from State censorship. If certain speech, including AI-speech, which proves to be worthwhile (to some) is not accessible, these particular individuals could only rely on their FoI as developed in the scope of FoE to access the information in question. Since FoI was limited by HRC, ECtHR and several states to certain contexts and situations and an intrinsic right to information has not yet been recognized in ICCPR, a big part of information would fall short of any protection.

2.2.3 The Media and AI

FoE analysis is incomplete without including the media, which plays a crucial role in its realization. It has been advanced in IHRL that the media possess considerable power over

115_{Abrams v United States [1919] Supreme Court 316, dissenting opinion, 250.} 116_{Garton Ash (n38), 119.}

117_{Generally Alexander Meiklejohn, Free Speech and Its Relation to Self-Government (Lawbook Exchange}

2000).

118_{Wenzel (n68), par.4.} 119_Ibid.

120_{Sarah Joseph and Melissa Castan, ‘Freedom of Expression - Articles 19 and 20’, The International Covenant}

on Civil and Political Rights: Cases, Materials and Commentary (3rd edn, OUP 2013), 590.

121_{Garton Ash (38), 73.} 122_{GC34 (n66), par.2.} 123_{Wenzel (n68), par.5.}

(23)

individuals.125_{AI is widely used in the media, e.g. chatbots, content creation and}

distribution,126_{therefore, the role that media have in society is shaped by technology.}127_{It has}

been emphasized by Special Rapporteur Kay that algorithms are not capable of evaluating cultural context and are more likely to engage in content blocking and undermine individuals’ FoI.128_{It can be concluded that challenges posed by AI have been acknowledged to a certain}

point, yet the specific problem of AI-speech has been up to date considered only by the Council of Europe (CoE).129

On 17 March 2014 at 6:28am, LA Times published an article about an earthquake130

taking place only three minutes earlier. The piece was plain but informative, authored by Schwenke, however, the credit ought to go to his computer, Quakebot, writing without any human interference.131_{For years now, machines have been deployed writing minor league}

sports game summaries,132_{news about stock market and financial reports for inter alia}

Groupon, Forbes and USAA.133_{In MittMeddia, use of bots is popular on real-estate news and}

almost 40,000 articles were created up until only February 2019.134_{So far, the AI is not able to}

write long pieces completely independently and its writing can at times present to be awkward and rough. However, the software is improving: in 2016, Matsubara developed a novella in collaboration with an AI-system and it passed the first judging round for the Hinoshi Shinichi Literary Award, it was evaluated as well-structured, although lacking in some aspects.135

It is reasonable to expect that the public will encounter more and more information generated by AI. According to report by Reuters Institute, in 2018 72% of publishers were

125_{McGonagle (n63), 9.}

126_{‘Spotlight on: Artificial Intelligence & Freedom of Expression’ (OSCE 2020), 2.}

127_{Jack M Balkin, ‘Free Speech in the Algorithmic Society: Big Data, Private Governance, and New School}

Speech Regulation’ [2017], 1151.

128_{David Kaye, ‘Promotion and Protection of the Right to Freedom of Opinion and Expression’ (UNGA 2018)}

A/73/348, par.29.

129_{Natali Helberger and others, ‘Artificial Intelligence - Intelligent Politics: Challenges and Opportunities for}

Media and Democracy’ (CoE 2020) Ministerial Conference.

130_{The news story:}

https://www.latimes.com/local/lanow/earthquake-47-quake-strikes-near-westwood-california-yxdnr8-story.html#axzz2wGVTRtfC

131_{Heaven (n1), 1967.}

132_{Jonathan Follet, ‘A Story Told by a Machine’ (towards data science, 8 November 2019)}

<https://towardsdatascience.com/a-story-told-by-a-machine-328002e3487>.

133_{Burk Krohe, ‘What Is Content AI?’ (Powerpost, 28 February 2018).}

134_{Nils Jacobsen, ‘Wie Ein Schwedischer Verlag Mit Robo-Journalismus Dauerhaft Digitalabonnenten}

Gewinnt’ Meedia (12 February 2019).

(24)

actively experimenting with AI.136_{In 2019, Press Association produced 30,000 news a month}

using AI and the content has expanded beyond a mere formulaic writing.137

2.3 Interim Conclusion

Regardless the problems that technology still face, AI plays an increasingly important role: from fully automated stories to art work.138_{The latest report by Reuters Institute finds}

that more than half of the publishers believe that robo-journalism will be at least somewhat important in the current year.139_{Therefore, the debate on AI-speech regulation is increasingly}

relevant. Keeping in mind that an intrinsic right to information is absent from HR treaties, recognizing AI that generates information as a bearer of FoE seems to be, at least for the time being, a plausible solution for protecting AI-speech from State censorship and to enable individuals’ access to it.

136_{Nic Newman, ‘Journalism, Media, and Technology Trends and Predictions 2018’ (Reuters Institute for the}

study of journalism 2018), 5.

137_{Bernard Marr, ‘Artificial Intelligence Can Now Write Amazing Content -- What Does That Mean for}

Humans?’ Forbes (29 March 2019) <https://www.forbes.com/sites/bernardmarr/2019/03/29/artificial-intelligence-can-now-write-amazing-content-what-does-that-mean-for-humans/#5a48e0c350ab>.

138_{Helberger (n129), 1.} 139_{Newman (n136), 22.}

(25)

Chapter 3: FoE for AI

3.1 Introduction

Hitherto was sustained that existing FoI framework is insufficient to offer adequate protection to AI-speech, namely due to the limitation introduced by HR bodies. Several authors have already argued in the realm of the First Amendment that AI-speech shall be constitutionally protected. It is true that the United States (US) are exceptional in broad free-speech protection, nevertheless use of AI with the capabilities to generate information is spread throughout the world. Accordingly, it is necessary to address whether the current IHRL framework can accommodate speech by a non-human entity. It is questionable to claim that any HR could be applicable to an artificial agent, yet as illustrated above, human individuals could face denial to access information generated by AI which could potentially undermine one’s search for the truth, democratic participation and personal development. This Chapter will examine whether AI with the ability to generate information without human interference (Second Order Robotics) could qualify as a bearer of FoE. In order for an entity to be a bearer of rights and duties, it must possess (international) legal personality.140_{This suggests that it is}

necessary to first explore if AI can have ILP and, then, move towards the analysis of FoE for AI.

(26)

3.2 ILP

There is no numerus clausus of entities with ILP nor any clear criteria, therefore the dictum of International Court of Justice (ICJ) is still relevant: ILP is a concept “giving rise to controversy.”141_{Yet, a number of theories deals with who or what can be an ILP.}

Historically, only states,142_{and potentially international organizations (IO),}143_{had ILP.}

However, state-centric system has been challenged by various non-state actors acting in the international arena,144_{including individuals. HR are traditionally granted to individuals; hence}

the present part will discuss non-traditional theories, ILP of individuals and significance of human features. In the last part, eligibility of AI to be protected under IHRL, in particular FoE, will be addressed.

In line with ICJ case law145_{and literature,}146_{the terms “subject of IL” and “ILP” are}

used interchangeably. Moreover, it shall be noted that this thesis does not claim that AI could and will act in all ways identically to a human being, but rather that strong AI may be granted

certain rights/duties.

3.2.1 ILP for AI

Keeping in mind that AI was created to serve humans rather than to enjoy independent existence, experts are reluctant towards granting it ILP.147_{Though, Shaw wrote: “personality}

is a relative phenomenon varying with the circumstances,”148_{and implicitly acknowledged that}

entities could be recognized as ILP depending on the needs of the international community. Hence why this part analyzes whether AI could technically become a subject of IL.

141_{Reparation for Injuries in the Service of the UN [1949] ICJ AO, 1949 ICJ Rep 174, 178.}

142_{Andrea Bianchi, ‘The Fight for Inclusion: Non-State Actors and International Law’ in Ulrich Fastenrath and}

others (eds), From Bilateralism to Community Interest (OUP 2011), 40.

143_{Reparation for Injuries (n141), 318.}

144_{Janne Elizabeth Nijman, The Concept of International Legal Personality: An Inquiry into the History and}

Theory of International Law (TMC Asser Press 2004), 354.

145_{Reparation for Injuries (n141), 179.}

146_{Kate Parlett, The Individual in the International Legal System: Continuity and Change in International Law}

(Cambridge University Press 2011), 29-30.

147_{Open Leter (n6).} 148_{Shaw (n140), 156.}

(27)

3.2.1.1 Non-traditional ILP theories

Contrary to the traditional view, the non-traditional theories include non-state actors and, thus, seem more promising for the present discussion. According to Gierke, legal person (LP) can149_{emerge from factual existence, this means that LP arises from social circumstance}

and it has capacity to bear rights and duties, act and decide freely.150_{It has been shown that}

robots, when acting independently of human interference, can possess their own autonomy, very similar to individuals, therefore, it is not impossible to conclude that it could appear that they have factual existence. Although, as Nijman writes, such perspective of “something real” is not common for IL, on the contrary, IL itself determines which entity has ILP,151_{rather than}

“reality.”

The “actor conception” uses term “actor/participant” and, hence, considers all entities exercising powers in IL.152_{This theory is promising since it rather focuses on what the entity}

does and much less on what it is. Up till now, AI does not yet possess any considerable power

in IL, yet in the future, more advanced machines could be perceived as participants.

Clapham focused on addressees of legal rules rather than on their status under IL.153

Therefore, any entity, including individuals, which is a bearer of rights and duties under IL, is a LP.154_{Accordingly, if AI were to become an addressee of IL, it would automatically have}

ILP.

3.2.1.2 Individuals and ILP

Due to HR focus, it is necessary to rely on scholars who considered individual-centered ILP conception. Kelsen presented it as follows: entity has ILP if it is an addressee of international norm(s).155_{The theory builds on the assumption that every legal norm regulates}

human behavior.156_{It does not follow, that individuals are the exclusive ILP, when law}

149_{The other two being explicit and implicit attribution. In Nijman (n26), 32.} 150_{Nijman (n144), 33.}

151_{ibid, 37.}

152_{Rosalyn Higgins, Problems and Process: International Law and How We Use It (Repr. Clarendon Press}

2001), 50.

153_{Bianchi (n142), 43.} 154_{Discussed in 4.2.2.}

155_{Astrid Kjeldgaard-Pedersen, The International Legal Personality of the Individual, vol 1 (Oxford University}

Press 2018), 26.

(28)

addresses collective entity, this entity is endowed with ILP. Kelsen was not the only scholar opposing the state-only ILP view, among others, Scelle157_{and Politis}158_supported

individual-focused IL system.159_{Similarly, Lauterpacht wrote: “There is no rule of international law}

which precludes individuals […] from acquiring direct rights under […] IL […].”160_Turning

the attention towards a more recent practice of ICJ, existence of individual rights under international treaties was clearly upheld in LaGrand,161_{confirming that their protection}

expands beyond HR.162

3.2.1.3 Human characteristics

In this section, “human characteristics” refer equally to biological and psychological qualities. Some authors seek to assimilate LP to the philosophical notion of a person, yet others do not require such a connection.163_{The former is supported by Austin, who suggests that a}

natural person is a human being with rights and duties, an equivalent to a biological being;164

diversely Kelsen supports the latter view arguing that it is wrong to identify a natural person with the biological status.165_{Accordingly, it will be considered whether human characteristics}

of individuals are a decisive factor in recognizing their ILP.

Perhaps the terms “person” and “personality” are misleading. In everyday conversations, the term “person” is often used as a synonym for a human being: “[…] quality which makes a being human.”166_{Contrary, the legal definition refers to the capacity of}

possessing rights and duties and, as such, may exclude some humans and include some non-human entities.167

157_{Only individuals are the subject of law in international law. In Georges Scelle, Précis de Droit Des Gens,}

Principes et Systématique, vol 1 (Librairie du Recueil Sirey 1932), 42.

158_{“It is no longer exclusively the law of States, but not yet completely that of human beings.” In Nicolas}

Politis, Les Nouvelles Tendances Du Droit International (Librairie Hachette 1927), 91.

159_{Kjeldgaard-Pedersen (n153), 30.}

160_{Hersch Lauterpacht, ‘The Subjects of the Law of Nations’ in Elihu Lauterpacht (ed), International Law:}

Being the Collected Papers of Hersch Lauterpacht, vol 2 (Cambridge University Press 2009), 526.

161_{LaGrand (Germany v United States of America), Judgement [2001] ICJ Rep 2001 466, par.77.} 162_{Kjeldgaard-Pedersen (n155), 57.}

163_{In general David J Calverley, ‘Imagining a Non-Biological Machine as a Legal Person’ (2008) 22 AI &}

SOCIETY 523.

164_{John Austin, Lectures on Jurisprudence: Or the Philosophy of Positive Law (1874) (Kessinger Publishing,}

LLC 2007), 350.

165_{Kelsen (n156), 94.}

166_{John Simpson, ‘Personality, n. and Adj.’ <https://www.oed.com>.}

167_{Samir Chopra and Laurence F White, A Legal Theory for Autonomous Artificial Agents (University of}

(29)

The human-qualities vs. AI debate has been raised by Solum, who considered a scenario in which a robot demands certain rights. A relevant objection for Solum is the absence of “some critical component of personhood,”168_{implying those could be certain human characteristics.}

Taking into account the most obvious example, corporations and IO, it can be deduced that human character of the entity is not a necessary condition for personhood.169_{Correspondingly,}

some national jurisdictions recognized entities, which are traditionally objects, as LP. For instance, Whanganui River, New Zealand, has been recognized as a LP170_{and orangutan}

Sandra was granted certain rights in Argentina.171_{While it is true that national legal personality}

does not automatically grant ILP, recognizing certain objects’ rights could serve as the first step towards ILP. The obvious difference between these entities and AI is that machines are human creations, while animals and rivers originate from nature and, to a certain point, evade human control. Additionally, the law neither recognized being human or possessing human capacities as a sufficient condition,172_{the obvious examples being slaves and nasciturus. Given}

the above, it appears irrelevant, or at least not conclusive, whether the entity possesses human characteristics when considering legal personality.

3.2.1.4 Social recognition

For untypical LP, social recognition is of bigger significance than for traditional subjects.173_{For instance, Whanganui River is indispensable for Maori people. This implies that}

personhood is a status of a (non-human) entity that humans care about. Therefore, a crucial determinative factor for AI to become LP would be the formation of genuinely meaningful relationships174_{to gain through them enough respect from humans. Even Turing test, which}

determines whether AI is capable of thinking in a human-like manner,175_{indicates that}

individuals tend to measure AI-personhood against the ability to be recognized as “human.” This suggests that humans need to see AI as an equal in social interactions. However, it is not

168_{Lawrence B Solum, ‘Legal Personhood for Artificial Intelligences’ (1992) 70 NCL Rev 1231, 1258.} 169_{Samir Chopra and Laurence White, ‘Artificial Intelligence - Personhood in Law and Philosophy’ [2004]}

ECAI’04: Proceedings of the 16th European Conference on Artificial Intelligence 635, 635.

170_{Abigail Hutchison, ‘The Whanganui River as a Legal Person’ (2014) 39 Alternative Law Journal 179.} 171_{David R Lawrence and Margaret Brazier, ‘Legally Human? “Novel Beings” and English Law’ (2018) 26}

Medical Law Review 309, 324.

172_{Chopra and White (n169), 640.}

173_{Roman Dremliuga, Pavel Kuznetcov and Alexey Mamychev, ‘Criteria for Recognition of AI as a Legal}

Person’ (2019) 12 Journal of Politics and Law 105, 109.

174_{Chopra and White (n167), 187.}

(30)

sustained that recognition is necessary or sufficient for an entity to become LP, it rather suggests that the lack of such recognition could be a crucial obstacle for untypical entities.

3.2.1.5 Dependent LP

Some scholars have pointed out the difficulties in AI-speakers claiming their rights.176

Here, this problem will be addressed only briefly keeping in mind the significant challenges remain to be addressed.

According to Chopra and White, AI could be recognized as a dependent LP, which means it could only act through the agency of another LP, similarly to children, adults without sound mind, corporations or river(s).177_{It follows that, in order for AI to claim any protections,}

it would need to be represented by another entity claiming protection in its name, e.g. AI-rights organizations, individuals interested in AI-speech or specialized lawyers.178_{Legal system}

could follow precedent currently in place for corporations. Especially in the context of false statements, it will be necessary to adjust the existing legal frameworks so AI could be held responsible.

3.3 HR for AI

HR system focuses on human beings and, as such prima facie does not correspond with AI. To say the least, claiming that machines ought to have HR is contentious. Yet, in the light of fast-paced technological development, this thesis attempts to draw attention on this thinking to possibly benefit full realization of FoE. Since HR are granted to individuals, it is necessary to first briefly look at why and how humans became bearers of such rights.

176_{Solum (n168), 1248-52.} 177_{Chopra and White (n167), 159.}

(31)

3.3.1 Development of IHRL

HR development started on domestic plane, consequently, treatment of people remained within internal sphere of respective national jurisdictions.179_{The only exception}

being treaties facilitating abolition of slavery.180

In the aftermath of the WWII, the international community needed a new approach for individual rights under IL.181_{Individuals became direct holders of enforceable}182_{rights under}

IL and bearers of responsibilities that mainly concern international crimes.183_{Adoption of UN}

Charter (UNC) finally made HR a subject of concern for the international community,184

although, due to Article 2(7)UNC, States were at most required to “promote” HR.185

Nonetheless, the UN Commission on HR was given the task to create a HR instrument applicable to all human beings and, as a result, UDHR was proclaimed in UN General Assembly.186_{It was under these circumstances, that HR of individuals were gradually}

recognized in IL.

3.3.2 Rights for AI

Hitherto was observed that untypical LP were recognized as such mainly for their own protection. Originating from fear of AI supremacy, authors tended to focus on obligations of machines rather than on their protection. Asimov, a pioneer in defining ground rules for AI,187

emphasized duties of AI towards humans. However, his Third Law188_{about the machine}

179_{Shaw (n140), 213.}

180_{Brussels Conference Act of 1890 (adopted 2 July 1890, entered into force 31 August 1891).} 181_{Ed Bates, ‘History’, International Human Rights Law (3rd edn, OUP 2018), 16.}

182_{Example: First Optional Protocol to ICCPR (adopted 16 December 1966, entry into force 23 March 1976) 99}

UNTS 171 and ECHR (n3), art.34.

183_{Charter of the International Military Tribunal (adopted and entered into force 8 August 1945) 82 UNTS 279}

(Nuremberg Charter), art. 6(b), Ibid, art. 6(c); Rome Statute of the International Criminal Court (adopted 17 July 1998, entry into force 1 July 2002) 2187 UNTS 3 (Rome Statute), arts. 5-8bis.

184_{Charter of the United Nations (adopted 26 June 1945, entered into force 24 October 1945) 892 UNTS 119}

(UNC), Preamble.

185_{Ibid, arts. 55(c), 56.} 186_{Bates (n181), 18.}

187_{Isaac Asimov, I, Robot (Bantam Spectra 2008), 183-216.}

188_{A Robot must protect its own existence as long as such protection does not conflict with the First or Second}

(32)

protecting its own existence, although of lower importance than First,189_Second190_and

Zeroth191_{Law, could be regarded as the beginning of “robot protection.”}

Some scholars advocate the obvious difference between humans and AI: machines are built by humans and, as such, inferior.192_{This understanding aligns with the main argument}

against HR for AI, the biological rationale. To question this view, it is necessary to question the limits of HR. First challenge seems to be what the term “human” encompasses. Beyond the fact that it is equated with Homo Sapiens, UDHR assigns it a further meaning: humans are “endowed with reason and conscience”193_{and have “inherent dignity […].”}194_{Individuals are}

entitled to these rights solely on the basis of being human and, thus, the idea of granting these rights to any alternative entities is, to say the least, controversial. Albeit, as mentioned, not even all human beings have been historically entitled to HR. Slaves were considered property, in some US states even until the 19th_century.195_{There are plenty of other examples where}

humans were not considered to be worthy of basic rights which they should theoretically hold, for instance women and Jews. Therefore, it seems that granting HR is largely dependent upon what the society understands as “human.” As the understanding of the term evolved through time, it progressively included individuals previously considered as inferior.196

Based on this historical development of expanding HR protection, the society could at some point in the future recognize AI as being “human-enough” to deserve HR protection. The technology is nowadays becoming increasingly “human-like,”197_{mainly interactive, but also}

affective and corporal.198_{A similar argument was made by Solum: if cognitive process of AI}

became relatively similar to those of humans, there are good reasons to treat machines as

189_{A Robot may not injure a human being or, trough inaction, allow a human being to come to harm.} 190_{A Roboto must obey the orders given to it by human beings, except where such orders would conflict with}

the First law.

191_{A Robot must not harm humanity.}

192_{Keith W Miller, ‘It’s Not Nice to Fool Humans’ (2010) 12 IT Professional 51, 51-52.} 193_{UDHR (n59), art.1.}

194_{ibid, preamble.}

195_{‘A History of Slavery in the United States’}

<https://www.nationalgeographic.org/interactive/slavery-united-states/>.

196_{Bates (n181), 11-16.}

197_{Massaro and Norton (n17), 1172.}

International Human Rights Law and Artificial Intelligence: Applicability of International Legal Framework on Freedom of Expression to Speech Generated by Artificial Intelligence