Machine Sensation

Front matter

| Machine Sensation: & Anthropomorphism and ‘Natural’

Interaction with Nonhumans ‘Tessa Leach

Machine Sensation

Contents

Introduction 9 Anthropomorphism 12 The value of taking nonhuman perspectives seriously 19 A note on language 28 Guiding questions 33 Structure of this book 40

1. ANT and OOO for technology 45 Actor-Network Theory (ANT) 46 Object-Oriented Ontology (OOO) 49 Alien phenomenology 56

2. Kinect and “Natural” Interfaces 67 The objects themselves 70 The Kinect-brand 79 Kinect-artefact 1.0 84 Kinect-artefact 2.0 87 Communication, sensation and relation of the Kinect-artefact Kinect Sports Rivals 100 Alien phenomenology of the Kinect-artefact 107

Introduction

The Voyager 2 spacecraft, currently hurtling at great speed away from

Earth and towards the edge of our familiar solar system, along with its

twin Voyager 1, will be the first human-made machines to pass beyond the heliopause, and to encounter the objects that make up interstellar space.' Sometime after this, perhaps as early as 2025, its sensors will be switched off and it will continue, inert, through space until it approaches the star Sirius in about 296,000 years.” The Voyager spacecraft are not the first human objects to cross into interstellar space - human-made electromagnetic waves began leaving Earth over a century ago. But the Voyagers are the most anthropomorphic object yet to leave behind the aura of our sun’s energy, the energy that is the root of life on Earth, of human life, and ultimately of human technology. Voyager 2 is a machine built by humans, incorporating metals extracted by human ingenuity from the Earth, plastic derived from decayed organic matter and plutonium manipulated by highly skilled human engineers. There is also a time capsule aboard the vessel, carrying information about the human race considered relevant by the scientists who launched it in 1977.* Among other things, this includes a disc known as

the “Golden Record”, containing analogue-coded photographs of human life, greetings in 55 different Earth languages, and music from Beethoven to Chuck Berry. Moreover, it carries sensory organs designed to gather information about phenomena that human scientists consider important, and it conveys that information in a way that humans on Earth can receive and interpret. Voyager 2 is an object made of human materials and ideas, and it is careering as fast as possible away from us, making its way further into alien territory by the instant.

The point of this story is not to marvel at the way in which an artificial object, intrinsically linked to science and human exploration, can gradually become alien to us. In popular parlance, the alien is that which is strange, inhuman and sometimes hostile. It is often used to describe the possible residents of other planetary bodies, perhaps beyond the edge of our solar system. Our desire to explore will soon bring a human-made artefact in contact with this alien territory. But Voyager 2 was already an alien before it was ever launched into space, despite the immense human effort put into creating it. Like all nonhumans, the qualities that define Voyager 2 and its way of being in the world are for the most part inaccessible to us and are evident only in interactions between it and other objects — that is, how it senses and relates to its environment (including us), and how its environment relates to it. We built this machine, yet the argument that it is not fully accessible to us is a key aspect of one of the most important theoretical influences on this book, Graham Harman’s object-oriented philosophy (OOP), also known as object-oriented ontology (OOO). OOO is a prominent set of theoretical, practical and artistic ideas that have significantly influenced the so-called nonhuman turn, a movement in the contemporary humanities away from an anthropocentric view of the world and towards foregrounding nonhumans and their perspectives. In later pages, the details of Harman’s ontology and other similar positions will be described, depicting the universe as a glittering fractal of objects and their relations. An object like Voyager 2 has complex relationships both with individual humans (such as engineers and politicians) and with a human culture that promotes exploration of the universe, but it also interacts with billions of nonhuman objects (such as electrons and stars) in ways that are closed off from our experience.

Nevertheless, something about Voyager 2 will reflect its human origins. Its means of signalling to Earth, its evidence of human workmanship and particularly its cargo of precious images and ideas, all point to its close relations with humanity that may well persist beyond the lifespan of the human race. Voyager 2 possesses human-like qualities in that it bears distinct evidence of close modelling upon humans and human intent in their design, although it is unlikely to be confused with a human. In contrast,

the case studies discussed in this book are frequently compared with humans. We project humanness onto them and pick and choose the alien qualities that are of most interest to us. But this book will argue that every anthropomorphic machine, every sensor, every robot, every interface, every word, and every concept of future or transformative technology is as alien to us as is Voyager 2.

Anthropomorphic machines are marvellous. We marvel at them. From intuitive chatbot interfaces to fully-fledged bipedal robots, we love to look for human-like qualities in machines. Anthropomorphic machines are charming when they work and frustrating when they fail. This is not necessarily because we see ourselves as the epitome of good engineering; we do not need to think of ourselves as quite that vain. But in many machines, particularly those that we bring into our homes, anthropomorphism aids in our ability to use them since humans are typically very good at communicating with other humans. They usually reproduce elements of human bodies or performance that make it easier for human users to comprehend how they relate to the world, apparently replicating qualities like eyesight and the ability to learn to speak. But we do not build machines to be like humans. We build machines to pretend to be humans. And by discussing specific technologies alongside contemporary anthrodecentric philosophy, this book aims to address the marvellous complexity of machine sensation and experience that anthropomorphic discourse tends to obscure. To a large extend, this is done through the mobilisation of and experimentation with OOO.

In engineering contexts, “anthropomorphic machines” typically refers to bipedal or social robots. Here, the term is much broader. I leave it deliberately vague in acknowledgement of the fact that anthropomorphism can be seen in an enormous variety of technological objects and other nonhumans. But a key factor that links all of the case studies in this book is the fact that they were designed or marketed as material or metaphorical mimics of humans. The case studies are, in the order presented:

1. The Microsoft Kinect, a natural user interface (NUTI) for the popular Microsoft Xbox One (a videogame console); 2. William Grey Walter’s robot tortoises (Machina speculatrix);

3. Terry Winograd’s SHRDLWU, an early language-using computer program; and 4. The concept of the gynoid sex robot.

These case studies were selected for particular reasons, and those reasons will be explained in each respective chapter. Although these objects are all human-made, and all possess qualities that make them anthropomorphic, I argue that anthrodecentric inquiry focusses investigation onto how anthropomorphic qualities in machines may obfuscate and confuse questions about their place in human society. Each are examples of protagonists in episodes when humans and artefacts have managed to comprehend each other successfully enough to form a useful working relationship, as well as occasions when an artefact stubbornly refuses to cooperate, when the human expects more than the artefact can deliver, or when the human does not act in a way that the artefact requires in order to do what the human wants it to do. Anthropomorphism is something that is done by humans in reaction to particular qualities in machines. Not everyone sees the same anthropomorphic qualities in machines. What might seem human-like to one human may to another human seem uncanny or a tool through which governments or corporations pursue their own agendas. This is an obstacle to consensus-building about the possible societal consequences of building anthropomorphic machines. This book contends that these episodes are usefully illuminated through greater exploration

of the artefacts’ sensory worlds and speculation about their internal experiences, even if those experiences are radically separate from us. In other words, we should take nonhuman perspectives seriously.

Anthropomorphism

Since it is easy for us to interact with human-like objects, and it is often easier to copy living things rather than invent brand new ways of being in the world, humans frequently anthropomorphise advanced machines.

Anthropomorphic qualities can improve the performance of artefacts, or they can improve human knowledge of human bodies and minds. The machines discussed in this book are all anthropomorphised for at least one of these two reasons. But anthropomorphism is a trick that is deceptive but persuasive, and consequently valuable or dangerous depending on

the context.*

We might think of technology as existing on a spectrum of complexity. One end of this complexity spectrum is occupied by the simplest of machines, such as levers and glass cubes, and the other is occupied by our most advanced computers. Inevitably, the advanced end of the spectrum has become associated with machines that we heavily anthropomorphise. Computers “think”, “hear” and “make mistakes”. These complex machines are the subject of much study both in philosophy and cultural studies. They have directed us to rethink what it means to be human and how we interact with objects. They visibly, and often deliberately, blur boundaries that modern Westerners are used to thinking of as impervious.

But our tendency to anthropomorphise is also evident in other kinds of objects. It can be seen in the assignment of human qualities to objects in the natural world, such as gods, animals and natural disasters.> Moreover, we can trace human interest in anthropomorphic nonhumans through stories of more primitive machines. The humanoid robots found in science fiction are the modern incarnations of venerable stories like Pygmalion’s statue and the golem (a clay humanoid enchanted into movement by a rabbi).° Narratives in which human-like sculptures come to life have been common since ancient times and are represented in early science fiction, when robots were monstrous and eerie.’ Medieval automata and Japanese karakuri were frequently made in the image of humans, and today the compulsion is evident in our most sophisticated AI systems. The case studies of this book are diverse in structure and sensory capabilities, but they share similarities because of the human-like attributes that we give them.

The desire for anthropomorphism often comes into conflict with requirements that we may have for machines, such as skills that humans do not have. It also sometimes diverts attention to things that do not aid in the task assigned to the nonhuman. As Brian Duffy puts it, “anthropomorphism obstructs the fact that human form is not the ideal for a machine.”® But that does not necessarily mean that designing human-like characteristics into a machine is necessarily frivolous or vain. Anthropomorphism in technology is a practical concern (such as when designing robots for therapeutic purposes’”) as well as being indicative of certain preoccupations of human designers (theorists have identified preoccupations such as vanity as well as religious faith!!). Anthropomorphised entities are more likely to be deemed worthy of our care and consideration.!? For example, the term “Mother Earth” is often used by people who wish others to behave in an environmentally conscious way.'? Studies have shown that humans are more likely to deem nonhumans responsible for their actions if the nonhumans have anthropomorphic qualities.!t The greater the potential action and “intelligence” of the technology, the greater we consider its ability for rational thought and decision-making. Claiming that an artefact is human-like can also be a way of emphasising its dangerous qualities. One of the most common tropes in science fiction is that of robots who are created to help humans but end up posing a threat. They are almost always anthropomorphic robots. They might be visually anthropomorphic, like the robots gone amok in Westworld,’ or they might have functionally anthropomorphic language and conversation skills like HAL in 2001:A Space Odyssey.'® It is psychologically easier for humans to furnish objects with moral responsibility if they have a stronger resemblance to humans. As Kate Darling points out, anthropomorphic framing of technological artefacts may be desirable or undesirable (from a human point of view).!” Anthropomorphising an object allows us to manipulate the response of human consumers, which might be a good thing with a Fitbit that encourages people to exercise more, but a negative thing if it can persuade people to release personal data.'* The resemblance need not be physical. One experiment by Clifford Nass and colleagues revealed that humans are even anxious about offending computers.'° Participants performed a task on Computer A and were then asked to evaluate the computer’s performance on the same machine. But a second group was asked to enter their opinion of Computer A on Computer B. When the participants evaluated the computer on the same machine, participants were less critical than when the evaluation was performed on Computer B. As Sherry Turkle puts it,

“participants do not want to insult a computer ‘to its face.’”?° Even when machines are not designed to look human, people impose human values and relationships onto them. Anthropomorphism can be useful for designers who wish to emphasise the human-like qualities of their creations for profit.

Machines may only bear a resemblance to part of the human form, and that part is likely emphasised by the designer. Take the case of a glass eye. In this case the goal is to replicate as realistically as possible the appearance of the eye, without its visual functions. But another type of anthropomorphism is evident in technologies that mimic the function of human bodies. A pertinent corollary to the glass eye is the retinal prosthesis, a technology used in the treatment of macular degeneration, a common eye condition that can seriously impede human vision. The retinal prosthesis attempts to merge naturally with the remaining eye tissue and stimulate parts of the retina.”! It simulates human sensation, but it is structurally very dissimilar from the eye and the user is visibly cyberised. The glass eye is visually anthropomorphic, while the retinal prosthesis is functionally anthropomorphic. Functional anthropomorphism can also be seen in our daily interaction with technology: in the assignment of names to our computers, or the impulse to swear at them when they malfunction.””

In semi-autonomous technological artefacts, sensation and perception of external data may be more or less functionally anthropomorphic. For example, a robot arm in an assembly line needs to be able to sense any instructions being sent to it from an external object (such as a computer), to have proprioception (an understanding of the position of its “body parts” in space) and to have a knowledge of where other objects are located (such as the object being assembled). Other machines have sensory apparatuses that make them better suited to environments with a lot of people. A social robot is most effective with both visual anthropomorphism and functional sensory anthropomorphism. It should be able to relate what it sees, hears, feels and smells in a way that seems natural to human users. Yet when we consider the mechanisms that produce these effects, it is evident that natural appearances are in fact highly contrived artificial senses. Machines do not have ears, nor do they have the centres of the brain responsible for hearing, but may have microphones and sound chips instead. Their sensation and perception of phenomena begins as machine-like and is heavily mediated to become human-like. We can have some insight into how the sound chip functions, easily performing some feats and being unable to perform others. But ultimately the structure of the sound card is only related to us through other objects. Why would we choose functionally anthropomorphic or visually anthropomorphic qualities in different cases? Our choice reveals the intended use and cultural or practical value of the artefact.

Choosing when to use anthropomorphic language and when to use more alien language can divide us on social issues. The boundary between human and nonhuman is drawn, policed, and transgressed in each of the case studies discussed. Acceptance of anthropomorphism comes more easily to some people than others. To some humans, anthropomorphic machines are endearing. To others, the same qualities that are meant to be endearing seem uncanny or even blasphemous.” Sex robots (or sexbots) are discussed in Chapter 5 of this book and are an excellent example.

The ethics of acting out a rape fantasy with a sexbot is dependent on

the perceived anthropomorphic or alien qualities of the sexbot. To an individual who sees more alien qualities in a sexbot, it is more tool-like than human-like and it cannot be raped. To an individual who sees more anthropomorphic qualities in a sexbot, the question of consent becomes more relevant. Note that this may not be because the sexbot is deserving of rights for her own sake, but because violence against machines with anthropomorphic qualities could lead to violence against human beings. Gender and ethnic identity are also difficult to define in robots because of the conflict between perceived human and alien qualities. Darling cites the power of anthropomorphism to both reinforce and challenge stereotypes.”* Nonhumans that lack anthropomorphic qualities like ATMs and factory robots cannot be accused of representing particular gendered or ethnic groups. But if a nonhuman is highly anthropomorphic then gendered

and ethnic signifiers must be built into them. As discussed in Chapter 2, even a simple machine can be accused of racism or sexism. Seeing more of a human or more of an alien in a nonhuman determines the degree to which it is interpolated into cultural and social issues. Anthropomorphism is subjective, but examining an object’s alien qualities alongside their human ones can help us to make more informed decisions about what they “understand” or “desire”. We could make an informed decision about the ethics of sexbot rape if we could access the inner being of individual sexbots, but we cannot. All we can do is focus on the relations they form with both physically embodied and incorporeal objects. It can be tempting to focus

on anthropomorphic qualities in nonhumans that approach the carefully policed human/nonhuman divide. But to be anthrodecentric we must focus on the whole object and accept our psychological impression of it as just one side of the story.

It might not be possible to know exactly what a pencil is experiencing. But it is not possible to know what another human is experiencing either, yet we still give voice to each other and advocate on behalf of marginalised groups. Anthropomorphic machines are built in our image — are they a marginalised group? Historically we have designated groups as less than human in order to justify violence or subjugation. Rosi Braidotti has called “the human... a normative convention, which does not make it inherently negative, just highly regulatory and hence instrumental to processes of exclusion and discrimination.”” Braidotti is speaking here not of machines, but of groups of humans that have been historically excluded from the concept of humanity (such as enslaved peoples). But we still regulate humanity and either resist or encourage the exclusion of anthropomorphic machines from human categorisation. How human-like does an object have to be for us to call it human? If we insist upon anthropomorphising large groups of nonhumans, we are met with a contradiction. We frequently refuse to consider them as individuals. When anthropomorphic machines are shipped en masse, we simultaneously inflict humanity on them while refusing them the individuality that humans, according to Western morality, deserve.

Of course, there is a problem with this argument in that it equates the suffering and struggle of marginalised or oppressed groups of humans with the human labelling of machines. Being dehumanised is a big problem for a human or groups of humans. But if anthropomorphic machines are marginalised by us, it is not necessarily important to them. Who can say what a social robot really wants — perhaps it wants to be a good companion to humans; perhaps it wants to encourage us to build more social robots; perhaps it wants to maintain access to its electric power source; or perhaps it simply wants to exist for as long as possible. We cannot be sure. If a social robot tells me it is happy, I am inclined to disbelieve it. But there is a certain point on the human-nonhuman spectrum at which we need to take seriously what nonhumans tell us about their experiences. Because it has anthropomorphic qualities, and was in part created by us in our own image, it is unconscionable to avoid acknowledging its social robot-ness. If we do not at least attempt to create an anthrodecentric narrative for the social robot, then how can we justify bringing these machines into the world? We deny them self-determination by avoiding their alien qualities. We contrive human-like qualities then force them into an unhuman outgroup. Nonhumans do not necessarily mind being forced into an outgroup, but humans should mind. We have enough trouble appreciating existing humans?’ individuality without creating more stereotyped anthropomorphic outgroups.

The challenge of examining relations in anthropomorphic machines is often complicated by cyborg or hybrid qualities in objects. It is becoming harder to determine where human agency, decision-making and sensation ends and where that of nonhumans begins. When we can’t tell what is human and what is nonhuman, the distinctions between them become less relevant in an expedient sense, even if they are still philosophically interesting. Thus we are left with interesting hybrids, actor-networks, and cyborgs: the employment of algorithms to buy and sell in that bastion of human culture and economics, the Stock Exchange; the use of artificial steroids to enhance athletes that participate in the Olympics, once the test of “natural” human bodies; the close touch of a robot, controlled by a surgeon outside the room, in the removal of cancerous cells.” The case studies in this book are deliberately presented as human-like, not just as human-machine hybrids. The chapters ahead contain examples of genuine attempts by humans to create an artificial other, often with personalities, names and the attribution of human qualities. Once again, these objects can best be interrogated by examining their relations. An essential first step is to decide upon the object of study and not be distracted by our own sloppy language choices. Is it the robot or robot-surgeon network that is the focus of investigation? Are we talking about the Stock Exchange as an embodied network, or as an incorporeal concept?

Another reason why anthropomorphic machines make interesting case studies is the fact that they are generally explicable by their creators.

Because anthropomorphic machines are engineered by humans, we can make useful deductions about their way of being in the world. In Chapter

3 the robot tortoises discussed possessed simple light and touch sensors. These sensors were built into them. Unlike when we study nonhumans such as bats and slime moulds, the question of which parts of the world the nonhuman is capable of sensing is already partly answered. In most cases, creators are able to understand why technological artefacts act in certain ways.’ There’s no need to reverse engineer a sensory organ if you’ve created it. In addition to this, the sensory organs of anthropomorphic machines can often be compared to human sensory organs. A light sensor may attempt to replicate human vision, but we know enough about the anatomy of the eye and the design of the sensor to deduce that they have different structures and, consequently, a different way of being in the world. Comparing the two yields an appreciation of what it is that makes nonhumans unlike humans and more alien. We can generalise this appreciation to apply to all nonhumans.

The value of taking nonhuman perspectives seriously

This book encounters anthropocentric thought on two fronts.

The anthropomorphism of machines is the first. The second is the anthropocentric thought that dismisses the sensation and experience of nonhumans beyond that of their successful or unsuccessful interaction with human “users”. Both of these tendencies distort nonhumans into specific kinds of caricatures. We are inhibited in our attempts to look beyond human problems or thinking, which can be a hindrance when persuading technological objects to behave one way or another. They may spring surprises on us. So, this book proposes to employ an anthrodecentric approach both to the theory and the case studies, a technique that actively removes the human from the centre of scholarly inquiry. We may discover which anthropomorphic qualities we give to (or inflict on) machines. An uninhibited approach to studying nonhuman phenomenological worlds can open our minds and, with any luck, facilitate more informed relationships between technological artefacts and humans. There is irony in this: the deployment of anthrodecentric thought to solve anthropocentric problems.

Science and philosophy are both heavily invested in the work of describing the lived experiences of humans. Yet there has been relatively little work studying the phenomenal world of the machine as a semi-autonomous being. The language to explain, for example, what constitutes the experience of a spacecraft, microphone, length of copper wire or videogame character, is still in its infancy (possibly due to a lack of interest in any aspect of its experience beyond that which is relevant to its technological functionality). This book argues that this is a significant problem for the philosophy of technology because it is only by comprehending the phenomenology of machines that we can arrive at an understanding of experiences between humans and nonhumans (or between nonhumans).

To begin this book, I must make one thing clear: no human can ever fully understand what it is like to be a robot or a computer. According to Harman, objects are radically divided from one another and the only part of the objects we can see are those that are evident in our interaction with them. I once believed that OOO could help increase sympathy between humans and anthropomorphic machines, which might smooth a path toward a more congenial relationship. But over time I have come to think that the opposite goal is also a worthy task. Anthropomorphism already promotes a feeling of sympathy for a nonhuman, but that sympathy sometimes obscures the alien processes that contribute to unanticipated failures in humanartefact relations. We sometimes underestimate the importance of paying attention to other relations such as those with materials, light, platforms, and ideas. OOO permits (and even encourages) a loss of sympathy through extreme alienation. But it can also spark interest in nonhuman experiences, and particularly in the sensory capabilities that nonhumans use to interact with one another.

The version of OOO presented in this book is based upon Harman’s work, from Tool-Being: Heidegger and the Metaphysics of Objects (which is based on his 1999 doctoral dissertation) to more recent publications. Throughout his career, Harman’s metaphysical position has remained steady, but it has found new areas of application and he and other authors have allowed OOO to flourish in unexpected ways. There have been other interpretations of a OOO nature, but I focus on Harman because of his prominent status in the movement and his continuing observation of and engagement with the changes taking place. The evolution of his writing is significant. His first books were deeply concerned with metaphysics and particularly phenomenology. But more recent works have addressed the socio-political implications of OOO.

Harman’s metaphysics is rich and vast. It is built on an interpretation of Heidegger that leads him to some unconventional conclusions.** Harman’s concepts of the quadruple object, withdrawal, and allure form a robust structure through which the interpretation of real examples is made possible. They provide the language to facilitate accounts of complex relations between objects. For the purposes of this book, it is not important whether Harman is correct in his ontology because it is mainly used here to focus the anthrodecentric inquiry into the case studies. Harman’s model provides a cohesive framework for studying specific objects and is a useful tool for both purely metaphysical and socio-political inquiry.

Chapter 1 will go into detail about OOO, but to those who are unfamiliar with the idea here is a quick contextualisation. OOO is fundamentally different from most other approaches to theorising being and relations because of its anti-correlationism. The word “correlationism” in this context was first used by Quentin Meillassoux in 2006 in his book After Finitude.”° It refers to the prevailing understanding in philosophy that the human view of the world is the only one that is possible or important to understand. All other perspectives are either inaccessible to humans or are not worth considering. Levi Bryant identifies a profound anthropocentrism at the heart of correlationism.*? However, most post-Kantian philosophers, from phenomenologists to idealists, are correlationists. In contrast, OOO is an anti-correlationist movement based on realism. It insists not only that nonhumans really exist outside of human perception but that we can know something of objects independent of human access. As Ian Bogost says, “humans are allowed to live [...] alongside sea urchins, kudzu, enchiladas, quasars, and Tesla coils.”?! Each of these nonhumans is an object, but the definition also extends to humans and to incorporeal things like concepts and signs. And each object, no matter its humanlike or corporeal status, is equally ontologically relevant. The perspective of the sea urchin is just as necessary for an understanding of the world as that of the human, even if it does not seem important to most humans when compared with our own perspectives. Matt Hayler points to OOO’s “usefulness” in the study of technological artefacts in particular.** Our bodies limit and structure our encounters with the world, and technologies both change our encounters with the world and exist as independent and inaccessible objects in their own right. Anti-correlationist thought is often a key concept for those interested in the concept of “flat ontology”,** since anti-correlationism discourages hierarchies and categories of objects. At the time of writing, Bogost and cultural critic Christopher Schaberg were commissioning books and essays on “the hidden lives of ordinary things”, such as golf balls and refrigerators.** In general, these texts on ordinary things follow the trend of rejecting the primacy of human lives and experience in favour of a focus on nonhumans, a project that is anti-correlationist in nature.