Artificial Intelligence and Theory of Mind in Affective Artefacts: Anthropological Issues in Legal and Ethical Philosophy

By Denis Larrivee and Luis Echarte

Introduction

Affective, often autonomous, computational and robotic artifacts constitute a rapidly growing sector of artificial intelligence application [1]. Deployed for use in a diversity of socially interactive domains as companions, health assistants, and elder and child caretakers such systems are able to tap into the affective system of humans; accordingly, they have both the potential to support, but also to benefit from, interactions that are structured [2]. Such artifacts raise specific anthropological and ethical issues related to human flourishing, dignity, and autonomy and, by extension, to the social structures flowing from human attachments. Many such systems now avail neuroscientific knowledge that is specifically designed to appeal to underlying neural correlates that evoke interactive responsivity. Among neural correlates yet to be considered are processes that are capable of inferring human intentions, a human ability whose neural correlates are designated theory of mind. Theory of mind is a generalized class of neural structures - e.g., mirror neurons - domains, and operations that draw intentional inferences from the actions of others [3], capacities that have facilitated evolutionary and cultural progress in human social performance. Combined with affective abilities, the duplication of these capacities in autonomous machines that are purposed to enhancing social interaction with humans, has the pragmatic outcome of broader anthropological assimilation. Such assimilation represents a widening incursion of ontological equivalence that relativizes anthropological meta-reality through technology. Accordingly, the evolution of AI/robotic artefacts through a simulated appropriation of human social and computational capacities constitutes a multi-dimensional influence, affecting not only social, legal and ethical praxis but also how these influences are articulated through evolving conceptions of technology's relation to anthropology. This poster explores the role of the growing intersection of neuroscience and technology in affective/intentional AI/robotic artefacts as a source for articulating an evolving anthropology-technology relation through legal and ethical praxis.

1. Is Technology Meant to Execute or to Share our Goals?

The use of technology has always had a close association with the human being. Impressed with a telos that is externally imposed technology is necessarily bound to the human being by its capacity to extend human ability. Technology’s proximity to human beings, however, was the subject of Heidegger’s critique where humans were themselves made subject to the manipulation of the technology they had created [4]. Heidegger’s concern is echoed in the recently released document Ethically Aligned Design [1] on the prospects, but also the dangers, posed by artificial intelligence technology. Accordingly, the document extends the notion of progress to include not merely the enhancement of computational or related abilities but also to align with human moral values to allow an elevated level of trust to permeate their interaction.

To fully benefit from the potential of AI and Autonomous Systems, we need to… go beyond… more computational power… make sure they are aligned to human ethical principles… to elevate trust between Humans and technology….’

Hence, trust is a crucial ethical variable and flows only from alignment with human value and well being. However, the development of ‘trust’ in technology that will benefit human flourishing is likely to be challenged by autonomous systems designed to elicit affective interaction.

2. Social capacities: A Dual Outcome Challenge

The remarkable social capabilities of humans have evolved to enhance group and cultural advance, features that require extended intervals for neural development [5]. Increased knowledge of social neuroscience has propelled advances in psychiatric care [6] , but has also yielded an abundance of knowledge on neural features that could be appropriated for social interaction through responsive AI device systems. Such devices, in fact, either have been implemented or are in advanced stages of development [2]. Their development raises at least two ethical metaconcerns. First, as Heidegger pointed out, they risk a misrepresentation of ontology, which is to say they introduce risk through the perception of being human, and the influence on ethical and juridical structure. Second, they risk ‘trust’, a risk flowing from the absence of value paradigms oriented to human well being. For example, in the absence of such moral framing, such systems are open to autonomous and intentionally deceptive actions, a circumstance likely to be exacerbated as knowledge of social neuroscience grows (Table 1).

3. The Challenge of Theory of Mind.

Affective computational and robotic technologies capable of sensing modeling, or exhibiting affective behavior by means of emotions, moods, personality are especially likely to elicit trust on the part of human partners. Such capacities can be combined with abilities for inferring human intentions, like that now studied in social neuroscience termed Theory of Mind [3]. These capacities are instantiated by a unique group of cells, e.g. mirror neurons, circuits, eg., fronto parietal, and subdomains, e.g. temporo-parietal junction, that are instrumental in their ability to facilitate group and social interaction. Their likely consideration or even appropriation in responsive AI systems design can be expected to further erode distinctions with human abilities and exacerbate the sort of dual challenge risks already introduced by intentionally affective devices.

4. Embedding AI Artefacts in Social Structure: Themes.

The embedding of AI artefacts in social order is intentional and pervasive, engaging a confluence of social and engineering disciplines. Though marked by multiple motivations, a consensus posits that humanoid simulation – appearing in multiple guises (Table 2) - will enable more rapid and adaptive appropriation. Campa, for example, identifies two key concepts that inform design efforts directed to the mimicking human features, scenario, narratives to achieve goals, and persona, concrete actors in the narratives [7]. Campa’s identification of simulated properties, however, does not exhaust the themes employed for assimilation. Beyond simulation, efforts aimed at broadening AI repertoires seek to achieve cognitive and computational properties that more closely resemble those of humans to facilitate conceptual exchange with artefacts in real world, joint tasking [8].

5. Ethical Meta-Principles for Enlightened AI Implementation: Relating Ontology to Praxis

Heidegger’s critique of technology specifically addressed the lack of transparency concerned with technology’s impact on the human being [4,9]. Current attempts to frame the ethical issue (Ethically Aligned Design Document) situate the lack of transparency in the context of manipulative self interest and conflicts of interest. While serious issues, Heidegger’s reading also concerns the subtle issue wherein the order of being fails to conciliate with the reality of the technology [4]. This ontological reconfiguration elicits a human affective investment not coincident with the nature of the artefact.

6. Current Efforts to Expand Value Paradigms in RoboEthics.

The sophistication and complexity of AI/Robotic devices, and their potential for acquiring more advanced functionalities, led to the early recognition not only of the benefit of extending human capability, but also of inflicting harm. Accordingly, initial ethical statements were framed in terms of meta principles that clearly asserted human control and safety, e.g, the frequently cited Asimov Laws (Table 3). Social assimilation and joint tasking roles, however, have been the stimulus for consensus based approaches that capitalize on stakeholder contribution [10]. Rather than presuppose a human normative standard, value is defined in terms of utility and a median position among competing interests. Improved autonomous capabilities, moreover, have stimulated value derivation paradigms premised on functionalist notions of property parity rather than ontological distinction. For example, notions of whether robot labor constitutes servitude emerge from value models that equate cognitive performance with ontological parity [11].

7. Human Rights? Or New Legal Philosophies.

The development of devices with capabilities for human interaction at the scale of affectivity and intention suggests that fundamentally new relations between technology and humans will be structured, with new susceptibilities to ontological misrepresentation and affective exploitation. Which legal philosophies best analogize these new circumstances and how can they best be used to enhance human flourishing? What interactions ought to be governed by statutory provision? In recognition of their enhanced autonomous capability The European Union’s RoboLaw Project identified the notion of accountability gaps emerging in liability definitions [12] (Table 4). For affective artefacts liability concerns mark a special sphere of interactive scenarios relating to disclosure, as well as emotive association and the notion of harm type. However, as an extension of ethical value, legal philosophy and praxis can also be expected to adjust to shifting value paradigms, accommodating a redistribution of value investment. Consensus ethics and the ontological functionalizing of anthropology have been taken up in new metaphorical notions that have shaped legal praxis [13]. Moreover, human rights can be expected to undergo similar redistribution, invested in artefacts in manners That parallel egalitarian actor network philosophies that have emerged from eco-ethics models [14].

References

[1] Ethically Aligned Design (2016) A Vision for Prioritizing Human Wellbeing with Artificial Intelligence and Autonomous Systems. IEEE Global Initiative http://standards.ieee.org/develop/indconn/ec/ead_v1.pdf.

[2] Sabonovic S (2014) Inventing Japan’s robotics culture: the repeated assembly of science, technology, and culture in social robotics. Social Studies Science 44(3):342-367.

[3] RizzolattiG (2010) Mirror neurons: from discovery to autism. Exp Brain Res 200(3):223-237.

[4] Onishi B (2010) Information, bodies, and Heidegger: tracing visions of the posthuman. Sophia 50:101-112.

[5] Decety J, Cowell JM (2014) Friends or foes: is empathy necessary for moral behavior. Perspectives Psychol 9(5):525-537.

[6] Cacioppo JT, Cacioppo S, Dulawa S, Palmer AA (2014) Social neuroscience and its potential contribution to psychiatry World Psych 13:131-139.

[7] Campa R (2016) The rise of social robots: a review of the recent literature. J Evolution Tech 26(1):106-113.

[8] Lin et al (2011) Robot ethics: Mapping the issues for a mechanized world. Artificial Intel 175:942-949.

[9] Rae G (2014) Being and technology: Heidegger on the overcoming of metaphysics. J British Soc Phenom 43(3):305-325.

[10] Rae G (2014) Heidegger’s influence on posthumanism: the destruction of metaphysics, technology, and the overcoming of anthropocentrism. His Human Sci 27(1):51-69.

[11] Stahl BC, Coekelbergh (2016) Ethics of healthcare robotics: towards responsible research and innovation. Robotics Auto Sys 86:152-161.

[12] Petersen S (2007) The ethics of robot servitude. J Exper Theor Artificial Intel 19(1):43-54.

[13] The RoboLaw Project. Regulating emerging robotic technologies in Europe: robotics facing law and ethics. (2014). www.robolaw.eu

[14] Calo R (2016) Robots as legal metaphors. Harvard J Law Tech 30(1):209-237.

[15] Chandler D (2013) The world of attachment: the post-humanist challenge to freedom and necessity Millennium J Inter Studies 41(3):516-534.