Title: Re-Centralization of AI focusing on Social Justice Author: Adnan Hadzi, Denis Roio

RE - CENTRALIZATION OF AI FOCUSING ON SOCIAL JUSTICE

In order to lay the foundations for a discussion around the argument that the adoption of artificial intelligence (AI) technologies benefits the powerful few, 1 focussing on their own existential concerns, 2 we decided to narrow down our analysis of the argument to social justic (i.e. restorative justice). This paper signifies an edited version of Adnan Hadzi’s text on Social Justice and Artificial Intelligence, 3 exploring the notion of humanised artificial intelligence 4 in order to discuss potential challenges society might face in the future. The paper does not discuss current forms and applications of artificial intelligence, as, so far, there is no AI technology, which is self-conscious and self- aware, being able to deal with emotional and social intelligence. 5 It is a discussion around AI as a speculative hypothetical entity. One could then ask, if such a speculative self-conscious hardware/software system were created, at what point could one talk of personhood? And what criteria could there be in order to say an AI system was capable of committing AI crimes? Concerning what constitutes AI crimes the paper uses the criteria given in Thomas King et al.’s paper Artificial Intelligence Crime: An Interdisciplinary Analysis of Foreseeable Threats and Solutions, 6 where King et al. coin the term “AI crime”. We discuss the construction of the legal system through the lens of political involvement of what one may want to consider to be ‘powerful elites’ 7 . In doing so we will be demonstrating that it is difficult to prove that the adoption of AI technologies is undertaken in a way, which mainly serves a powerful class in society. Nevertheless, analysing the culture around AI technologies with regard to the nature of law with a philosophical and sociological focus enables us to demonstrate a utilitarian and authoritarian trend in the adoption of AI technologies. Mason argues that “virtue ethics is the only ethics fit for the task of imposing collective human control on thinking machines” 8 and AI. We will apply virtue ethics to our discourse around artificial intelligence and ethics. As expert in AI safety Steve Omonhundro believes that AI is “likely to behave in antisocial and harmful ways unless they are very carefully designed.” 9 It is through virtue ethics that this paper will propose for such a design to be centred around restorative justice in order to take control over AI and thinking machines, following Mason’s radical defence of the human and his critique of current thoughts within trans- and post- humanism as a submission to machine logic.

The paper will conclude by proposing an alternative practically unattainable, approach to the current legal system by looking into restorative justice for AI crimes, 10 and how the ethics of care could be applied to AI technologies. In conclusion the paper will discuss affect 11 and humanised artificial intelligence with regards to the emotion of shame, when dealing with AI crimes. In order to discuss AI in relation to personhood this paper follows the descriptive psychology method 12 of the paradigm case formulation 13 developed by Peter Ossorio. 14 Similar to how some animal rights activists call for certain animals to be recognised as non-human persons, 15 this paper speculates on the notion of AI as a non-human person being able to reflect on ethical concerns. 16 Here Wynn Schwartz argues that “it is reasonable to include non-humans as persons and to have legitimate grounds for disagreeing where the line is properly drawn. In good faith, competent judges using this formulation can clearly point to where and why they agree or disagree on what is to be included in the category of persons.” 17 In the case of AI technologies we ask whether the current vision for the adoption of AI technologies, a vision which is mainly supporting the military-industrial complex through vast investments in army AI, 18 is a vision that benefits mainly powerful elites. In order to discuss these questions, one has to analyse the history of AI technologies leading to the kind of ‘humanised’ AI system this paper posits. The old-fashioned approach, 19 some may still say contemporary approach, was to primarily research into ‘mind-only’ 20 AI technologies/systems. Through high level reasoning, researchers were optimistic that AI technology would quickly become a reality. Those early AI technologies were a disembodied approach using high level logical and abstract symbols. By the end of the 80s researchers found that the disembodied approach was not even achieving low level tasks humans could easily perform. 21 During that period many researchers stopped working on AI technologies and systems, and the period is often referred to as the “AI winter”. 22 Rodney Brooks then came forward with the proposition of “Nouvelle AI”, 23 arguing that the old-fashioned approach did

not take into consideration motor skills and neural networks. Only by the end of the 90s did researchers develop statistical AI systems without the need for any high-level logical reasoning; 24 instead AI systems were ‘guessing’ through algorithms and machine learning. This signalled a first step towards humanistic artificial intelligence, as this resembles how humans make intuitive decisions; 25 here researchers suggest that embodiment improves cognition. 26 With embodiment theory Brooks argued that AI systems would operate best when computing only the data that was absolutely necessary. 27 Further in Developing Embodied Multisensory Dialogue Agents Michal Paradowski argues that without considering embodiment, e.g. the physics of the brain, it is not possible to create AI technologies/systems capable of comprehension. Foucault’s theories are especially helpful in discussing how the “rule of truth” has disciplined civilisation, allowing for an adoption of AI technologies which seem to benefit mainly the upper-class. But then should we think of a notion of ‘deep-truth’ as the unwieldy product of deep learning AI algorithms? Discussions around truth, Foucault states, form legislation into something that “decides, transmits and itself extends upon the effects of power” 28 . Foucault’s theories help to explain how legislation, as an institution, is rolled out throughout society with very little resistance, or “proletarian counter-justice” 29 .

Foucault explains that this has made the justice system and legislation a for-profit system. With this understanding of legislation, and social justice, one does need to reflect further on Foucault’s notion of how disciplinary power seeks to express its distributed nature in the modern state. Namely one has to analyse the distributed nature of those AI technologies, especially through networks and protocols, so that the link can now be made to AI technologies becoming ‘legally’ more profitable, in the hands of the upper-class. In Protocol, Alexander Galloway describes how these protocols changed the notion of power and how “control exists after decentralization” 30 . Galloway argues that protocol has a close connection to both Deleuze’s concept of control and Foucault’s concept of biopolitics 31 by claiming that the key to perceiving protocol as power is to acknowledge that “protocol is an affective, aesthetic force that has control over life itself.” 32 Galloway suggests that it is important to discuss more than the technologies, and to look into the structures of control within technological systems, which also include underlying codes and protocols, in order to distinguish between methods that can support collective production, e.g. sharing of AI technologies within society, and those that put the AI technologies in the hands of the powerful few. 33 Galloway’s argument in the chapter Hacking is that the existence of protocols “not only installs control into a terrain that on its surface appears actively to resist it” 34 , but goes on to create the highly controlled network environment. For Galloway hacking is “an index of protocological transformations taking place in the broader world of techno-culture.” 35 Having said this, the prospect could be raised that restorative justice might offer “a solution that could deliver more meaningful justice” 36 . With respect to AI technologies, and the potential inherent in them for AI crimes, instead of following a retributive legislative approach, an ethical

discourse, 37 with a deeper consideration for the sufferers of AI crimes should be adopted. 38 We ask: could restorative justice offer an alternative way of dealing with the occurrence of AI crimes? 39 Dale Millar and Neil Vidmar described two psychological perceptions of justice. 40 One is behavioural control, following the legal code as strictly as possible, punishing any wrongdoer, 41 and second the restorative justice system, which focuses on restoration where harm was done. Thus an alternative approach for the ethical implementation of AI technologies, with respect to legislation, might be to follow restorative justice principles. Restorative justice would allow for AI technologies to learn how to care about ethics. 42 Julia Fionda describes restorative justice as a conciliation between victim and offender, during which the offence is deliberated upon. 43 Both parties try to come to an agreement on how to achieve restoration for the damage done, to the situation before the crime (here an AI crime) happened. Restorative justice advocates compassion for the victim and offender, and a consciousness on the part of the offenders as to the repercussion of their crimes. The victims of AI crimes would

not only be placed in front of a court, but also be offered engagement in the process of seeking justice and restoration. 44 Restorative justice might support victims of AI crimes better than the punitive legal system, as it allows for the sufferers of AI crimes to be heard in a personalised way, which could be adopted to the needs of the victims (and offenders). As victims and offenders represent themselves in restorative conferencing sessions, these become much more affordable, 45 meaning that the barrier to seeking justice due to the financial costs would be partly eliminated, allowing for poor parties to be able to contribute to the process of justice. This would benefit wider society and AI technologies would not only be defined by a powerful elite. Restorative justice could hold the potential not only to discuss the AI crimes themselves, but also to get to the root of the problem and discuss the cause of an AI crime. For John Braithwaite restorative justice makes re-offending harder. 46 In such a scenario, a future AI system capable of committing AI crimes would need to have knowledge of ethics around the particular discourse of restorative justice. The implementation of AI technologies will lead to a discourse around who is responsible for actions taken by AI technologies. Even when considering clearly defined ethical guidelines, these might be difficult to implement, 47 due to the pressure of competition AI systems find themselves in. That said, this speculation is restricted to humanised artificial intelligence systems. The main hindrance for AI technologies to be part of a restorative justice system might be that of the very human emotion of shame. Without a clear understanding of shame it will be impossible to resolve AI crimes in a restorative manner. 48

Thus one might want to think about a humanised symbiosis between humans and technology, 49 along the lines of Garry Kasparov’s advanced chess, 50 as in advanced jurisprudence. 51 A legal system where human and machine work together on restoring justice, for social justice. Furthering this perspective, we suggest that reflections brought by new materialism should also be taken into account: not only as a critical perspective on the engendering and anthropomorphic representation of AI, but also to broaden the spectrum of what we consider to be justice in relation to all the living world. Without this new perspective the sort of idealized AI image of a non-living intelligence that deals with enormous amounts of information risks to serve the abstraction of anthropocentric views into a computationalist acceleration, with deafening results. Rather than such an implosive perspective, the application of law and jurisprudence may take advantage of AI’s computational and sensorial enhanced capabilities by including all information gathered from the environment, also that produced by plants, animals and soil.

44 Cp. Nils Christie, “Conflicts as Property”, The British Journal of Criminology, 17 (1), 1977, pp. 1–15. 45 Cp. J. Braithwaite, “Restorative Justice and a Better Future”, in E. McLaughlin and G. Hughes (eds.), Restorative Justice: Critical Issues, London, SAGE, 2003, pp. 54–67. 46 Cp. J. Braithwaite, Crime, Shame and Reintegration, Cambridge, Cambridge University Press, 1989. 47 Cp. A. Conn, “Podcast: Law and Ethics of Artificial Intelligence”, Future of Life, March 31, 2017. Available at: https://futureoflife.org/2017/03/31/podcast-law-ethics-artificial-intelligence/ [accessed September, 22 2018]. 48 Cp. A. Rawnsley, “Madeleine Albright: ‘The Things that are Happening are Genuinely, Seriously Bad’”, The Guardian, July 8, 2018. Available at: https://web.archive.org/web/20190106193657/https://www.theguardian.com9

/books/2018/jul/08/madeleine-albright-fascism-is-not-an-ideology-its-a- method-interview-fascism-a-warning [accessed October 25, 2019]. 49 Cp. D. Haraway, “A Cyborg Manifesto”, Socialist Review, 15 (2), 1985. Available at: http://www.stanford.edu/dept/HPS/Haraway/CyborgManifesto.html [accessed October 25, 2019]; C. Thompson, “The Cyborg Advantage”, Wired, March 22, 2010. Available at: https://www.wired.com/2010/03/st-thompson- cyborgs/ [accessed October 25, 2019]. 50 Cp. J. Hipp et al., “Computer Aided Diagnostic Tools Aim to Empower Rather than Replace Pathologists: Lessons Learned from Computational Chess”, Journal of Pathology Informatics, 2, 2011. Available at: https://doi.org/10.4103/2153-3539.82050 [accessed October 25, 2019]. 51 Cp. J. Baggini, “Memo to Those Seeking to Live for Ever: Eternal Life Would be Deathly Dull”, The Guardian, July 8, 2018. Available at: https://web.archive.org/web/20181225111455/https://www.theguardian.com /commentisfree/2018/jul/08/live-for-ever-eternal-life-deathly-dull-immortality [accessed October 25, 2019].