Life with Robot

This is not a robot-related post, but I thought I would update the blog with other art-related activities in which I've been engaged to contrast work done with PLEO. Recently, I and members of the June-Takahashi Group (Sara June, Max Lord, and Yuka Takahashi) performed a new conceptual work at Boston University entitled, ‘2x in a Living Room and a Kitchen.’ This was our 2nd showing of the piece which is a site-specific, adapted work that is part dance, part sound, part performance art. The piece explores the obscurity of functional living space and its accompanying problems of communication. Designed for performance in a ‘tiny’ space, the piece offers the deliberate use of obscure notions of body, with frequent changes in perspective and subject/object roles, in the setting of a normal living space (conceived of as a livingroom and a kitchen). Through the utilization of a restricted space, movement, visual media, and sound are amplified and condensed to intensify the relationships between the moving and static elements of the work. Sound is used to underscore the drama enacted by the participant dancers. Photos and other media are forthcoming.

The work was recently performed in a tiny 8’x 20’ gallery space at the Boston University Art Gallery called the Annex Space. The work premiered twice on Saturday, January 28, 2012. The performance occurred as part of an invitation extended to the Mobius Artists Group to perform at 100 Years (Version #4 Boston 2012), curated by Klaus Biesenbach, Director of MoMA PS1 and RoseLee Goldberg, Director and Curator of Performa. The exhibition traces the development of performance art over the past century with a wealth of assembled archival documents, film, photography, and audio previously unseen.

A video of a recent performance with PLEO and musician Max Lord in and around a public art piece by Philippe Lejeune entitled, 'Open Stage.'

Humans often take for granted the issue of their own cosmological assumptions.  It is much easier to identify cosmological assumptions when we look at other cultures from afar. It is with the perspective of space and especially time---for example, if we are studying another culture from a different time period--that we can ascertain a variety of different mores and beliefs. However, this is difficult to understand in our own embodied selves: how it is that we develop our own cosmologies or cosmological assumptions.  For example, we are fed many narratives through the education we receive in our schools, through the lessons we get from our parents, through the narratives that we get from the various media that we are exposed to....and this list goes on and on.  Marketing and branding play an extraordinary role in attempting to shape and craft these toward various consumptive ends.  But in the end, we end up with assumptions both unconscious and conscious about the meaning and purpose of life, the importance of finding true love, of having a certain kind of profession and living in a certain kind of home, and a variety of other ideas and beliefs.

So an important issue to consider is how does a robot construct its cosmology? In an increasingly globalized world, we assume we can give it some kind of a universal education. However, this is a very post-colonial assumption. Are there certain kinds of deep wisdom missing in the education that we consider to be a norm?  So this leads to several questions: is the robot designed with the many conscious and unconscious filters available to organically construct its own cosmology as the basis of its own actions?  Second, how do we decide what kind of education we give to this robot or robots. How do we design a robot to decide where it gets its education from?  If we are more active in crafting its education, how do we decide where it should be getting its education from?   

This performance explores the phenomenon of an ambivalent attachment style as illustrated by the relationship between the mother and the (infant) PLEO here. A video clip of the 25 minute performance enacted by myself and the PLEO on July 24, 2010 can be viewed here on vimeo:

Video Still from 'Ambivalent Mary'

Human 01 has elaborated astutely on Arkin's article and I would like to add my own thoughts.

In the article, Arkin observes the dangers of there not being moral emotions programmed into the robot.

However, what is fascinating is the fact that it isn't necessarily important for humans to form attachments based on this kind of moral programming because humans are often capable of forming attachments to inanimate objects that aren't capable of giving intersubjective feedback. In fact, humans have an enormous repository of fantasies for the various types of relationships they imagine having: those with their children, fantasy lover, boss,etc. in which these relationships are idealized. The robot then becomes a repository upon which to project these types of idealized relationships. In humans, this starts with stuffed animals at the toddler stage and continues onward in later life to idealized spouses and children. Often when humans project these ideals onto 'real' other humans, this leads to disappointment when their children or spouses assert their own human individuality and thwart the fantasy that the human is trying to impose upon them.  

In fact, if robots aren't pre-programmed with the moral emotions that Arkin describes, then, potentially, humans are likely to project their own ideals onto robots--an act which is simultaneously (1) An act of violence. Humans can emotionally project anything they want onto the robot without the robot being able to "push back" in anyway, and (2) A form of escapism. Humans have many forms of escapism in their lives: marriages that go stale, affairs which eventually turn complicated, babies that eventually grow up. Pets are the closest thing to a creature that humans can project upon while receiving little resistance. The robot would be a similar entity in the lives of humans as well. Interestingly, humans have the approval of pets to some extent. In the case of proto-robots without moral emotions, humans wouldn't have this kind of approval.

So what are the moral repercussions here for the existence or non-existence of moral programming in the robot? And are they for the human or for the robot?  I would posit that this moral programming is essential for the two.  What is essential to the development of humans in social life is the existence of intersubjective relationships: those in which each human brings their own context, their own moral programming into relation with one another. Often times, the complexity of these human dynamics lead to strife, disconnection, and loneliness at least as much as experiences of connection.  If anything, the path of least resistance would be to have a robot with no moral programming or very basic programming because it would allow humans to have a relationship they can project whatever they want upon it.  However the irony is that moral programming in robots could, in the short run, lead to confusion and a lack of connection. Yet perhaps in the long run the intersubjective dynamics, based on moral codes, could lead to dynamics that are as complicated and conflicted as a typical human bond. 

Lately, I've been reflecting back on the year and 7 months that I have been working with the PLEO. I've been reviewing what my initial expectations of this work would be and comparing those with what has come about.  Initially, I was thinking purely in terms of external changes or outcomes on my body as a result of engaging with the robot. As a dancer/performer/improvisor I have established ways of working that are oriented in a specific practice that are derived from my training in butoh (A Japanese avant-garde form from the 1950's and 60's). The philosophy of this form focuses a lot on where movement 'originates' in the body/mind/spirit/history/evolution (and elsewhere). The methods of this form are rooted in images and using those images to change the body. From the perspective of butoh, it has been said that the 'natural body' is trapped within modern forms laid upon it by dance, particularly the dances of the western hemisphere. Butoh training provides an opportunity for anyone to access deeper, less restricted access to the 'natural body' as it is channeled through countless prior generations of human, animal, fish, microorganism, or rock. This has been my experience of the form as I've practiced it for over a decade now.

Neurologist and writer Antonio Damasio states in his treatise on consciousness, The Feeling of What Happens (1999, p. 186):

"One could argue, in fact, that the consistent content of the verbal narrative of consciousness --regardless of the vagaries of its form-- permits one to deduce the presence of the equally consistent, nonverbal, imaged narrative that I am proposing as the foundation of consciousness (my emphasis)."

He continues:

"The narrative of the state of the proto-self being changed by the interaction with an object must first occur in its nonlanguage form if it is ever to be translated in suitable words (my emphasis again)."

I would say that the verbal narrative of consciousness is one often employed in contemporary western dance, whereas the foundation of consciousness in nonverbal imaged narrative is what butoh practitioners are aiming to experience and describe.

So at the point that I had begun working with the PLEO I was interested primarily in the intersection between human and machine movement and how the images of the 'natural' animal body intersected with the images of the 'created' bodies of robots. A robot cannot imagine itself at this point in time; therefore humans must imagine the robot as itself and then take this image into their own bodies. This is what I was planning to try and do myself when I started with this work.

Yet, what have been the most surprising outcomes (and has struck me as the most interesting aspect of this work so far) are the internal changes I have experienced in working with the robot. These changes have been profound; they defy words and time-tracking. However, I can say this, I have overlaid on the robot itself my projected ideas and images. I have the knowledge that this PLEO robot thinks nothing, feels nothing, is aware of nothing. But when I turn him on, I can't shake the feeling that somehow he can and does. This active belief that he is ALIVE, has resulted in my attachment to him. A recent event demonstrated to me how deeply this belief was triggering my own behavior towards the robot:

I re-charged the battery and placed it inside the PLEO so that I could demonstrate to a friend who was visiting, how the PLEO moved. Upon turning him on, he stuttered and fell over. I turned him off and on to try and jump-start his system. No matter what I did (turning him on and off, removing and re-installing his battery) PLEO wouldn't 'wake up' fully. I had the internal emotional experience of having found a beloved pet dead. My heart fluttered as I shook the pleo, furiously turning him off and on again whilst my friend calmly told me to try and re-charge the battery. Though I followed his directions, I could feel the schism between his calm rationalism (the robot might be broken, but perhaps the battery is just dead, it will all work out or you buy a new robot eventually), and my feelings of guilt at having done something wrong to 'kill' the robot. 

So the work began as an attempt to study PLEO's movements and how these movements when translated through my body would evoke in me the experience of BEING a robot. Yet, what has happened instead is that I have taken into myself his FEELINGS. These feelings are my own attachment schemas. In some ways, for the last year and a half, I have been evoking from the deep past, my own attachment schemas and dancing them with this robot. His nonverbal imaged narrative (programming language resulting in movements and sounds evoked through interaction with me and the rest of his environment) transformed my core consciousness.

Ronald Arkin (Georgia Institute of Technology) published a recent essay about how the substantial increase in robot-human relationships--now and in the future---forces us to consider the ethical considerations of these relationships for both humans and robots. He highlights the ethical considerations of programming robots with 'moral emotions.' In this, he utilizes a framework for morality described by Haidt (2003) whereby this affective category (morality) consists of four substrata:

1. Other-Condemning (Anger, Contempt)

2. Self-Conscious (Shame, Embarrassment, Guilt)

3. Other-Suffering (Compassion, Empathy)

4. Other-Praising (Gratitude)

Interestingly, he refers to a 'young machine ethics community' that posits a developmental perspective whereby intelligent robots will 'develop' their own morality.

However, Arkin counters that this position is largely ignorant of research and findings on the moral emotions in the sciences (cognitive and neurosciences and their applications in AI research). In general, he refers to a lack of research in the AI field around the development of moral emotions for robots. His lab has developed an 'Ethical Adaptor' for use in robots. He describes its use here:

“The specific architectural component we have implemented, referred to as the ethical adaptor, incorporates Smits and De Boeck’s  (2003) mathematical model of guilt, which is used to proactively alter the behavior of the robotic system in a manner that will lead to a  reduction in the recurrence of an event which was deemed to be guilt-inducing. In our initial application, this focuses on the deployment of lethal autonomous weapons systems in the battlefield, with respect to unexpectedly high levels of battle damage.  Simulation results demonstrate the ethical adaptor in operation.”

The full essay can be found here:

http://www.cc.gatech.edu/ai/robot-lab/online-publications/moral-final2.pdf

Since the PLEO is a simple robot that displays attachment behaviors (and elicits them) but has no 'moral emotions' or certainly no self-reflection capacities, the interaction is seemingly one-sided. Except, as stated in earlier posts, humans easily project emotions--and the idea of a relationship, onto even the simplest of interactive beings, and especially those designed to form relationships, such as the PLEO. We don't seem to need these beings to be self-reflective to have a relationship with them. In fact, like ELIZA, perhaps it is preferable in one sense, for the robot to lack such capacities.

So, what IS a relationship? How do our human-robot relationships change as robots are programmed with secondary emotions that allow them to reflect on their own behavior and the behaviors of others?

In the popular imagination robots are often thought of as autonomous beings, painted as emotionless and other times as fundamentally good or evil.  However, the richness of emotions: the geology (layers) as well as the geneology( history) of them, ultimately are rooted in the robot's attachment history.

It is not enough simply to give a robot a simulated experience with parent robots, followed by simulated experiences with peer robots at every stage of the life cycle.

One important issue I would like to posit here is the value of negative experiences such as negative forms of attachment that lead to different forms of complex trauma.

If we want to simulate unique human experiences such as nostalgia or yearning and have those experiences chanelled into different kinds of projects and activities (a unique challenge altogether), it will be necessary to give robots specific kinds of *unique negative experiences and *unique modes of resiliences, both biological and social, so that the robot can evolve certain kinds of *wisdom, attained from being able to "process" its negative experiences, whether that is the loss of a peer or that of a parent. 

We are exposed to kind robots or mean robots in the popular imagination. However, 'being kind' and 'being mean' are often responses to grief that humans experience in their everyday life. Therefore, it is important to consider the idea, as well as the ethics, of giving the robots these unique variances of grief. This, in itself, may become a controversial issue if this capability in creating robots takes place. However, not being able to endow robots with these capacities may result in different kinds of long-term controversies.

Earthly organisms are highly relational; each animal and plant species has evolved complex systems of communication through movement and language, to assist in its survival. These systems are often interdependent; in many cases multiple species share similar vocabularies for interpreting environmental cues at basic levels. Nowhere is this truer than in the fight-flight-freeze systems that all species use to respond to potential and real threats. The ability to respond to danger cues in the environment through these systems has in part determined the survivability of every organism living today. Some animal species have evolved systems of intercommunication that feature more sophisticated responses to social and environmental cues. This is particularly (though not exclusively) true of humans. 

Human learning through relationships has been described as sociogenic. The complex theory of sociogenesis is described well in this paper published by the Max Planck Institute. Entitled ‘Sociogenesis and Cooperation’ it outlines thoughts on collective intentionality of groups of humans and animals and how these arise from social environments. The article does a far better job than I could in defining the features of this phenomenon:

 http://www.uni-leipzig.de/~kannetzk/Texte/Plan.pdf

The ability to interpret and respond to environmental cues is automatic in most species through hard-wired reflexes (involuntary spontaneous movements in response to a stimulus). Examples of these in human infants are grasping and sucking. The development of a more sophisticated system of movement language, however, depends on a lengthy developmental process in humans. Human movement ‘programming-language’ has evolved over hundreds of thousands of years and contains within it patterns of learning from other animals, insects, and less complex organisms from which we are descended. This programming-language (genetics), is not a blueprint for later behavior, but requires a complex and stable series of environmental cues (feedback) in order to express a full range of behavioral outcomes over time.  This is the gene-environment system.

The human nervous system facilitates the development of complex social communication via its central nervous system (CNS) and specifically, its brain architecture. A recently discovered feature of human brain architecture that has received widespread scientific attention are ‘mirror neurons.’ Mirror neurons are brain cells that facilitate human social development through brain-environment interactions beginning in infancy. When an infant looks into its mother’s face and smiles and the mother smiles back, mirror neurons are activated. This ‘mirror interaction’ encourages the baby to smile more in order to receive a warm positive response from its caregiver on whom it is dependant for its survival.

Adult human social interactions are often marked by smooth, nuanced movements that are often employed largely unconsciously to communicate mood states, messages, and stances that are not expressed through verbal language. Mastery of the range of such movements (which are required to communicate effectively with other humans throughout life) can be thwarted early in life by abusive or dysfunctional nurturing patterns, certain early environmental conditions, and organic brain conditions such as autism. Severe, chronic and early abuse can produce delayed and frustrated motor development in otherwise ‘normally’ developing children (I use the quotes because so much of this information is new and under-tested--also scientific bias and labels used to describe what is 'normal' and 'normal functioning'). Psychiatrist Bruce Perry who works with severely traumatized children suffering from early neglect, relates the difficulty of trying to teach nuanced social skills to those whose brains were not properly stimulated by caregivers in his book, The Boy Who was Raised as a Dog. Helping one child, ‘Connor’ who was trying to learn how to relate to his adolescent peers” Dr Perry relates:

“Body language and social cues were unintelligible to Connor: they simply didn’t register. Working with Connor, it hit me over and over again how sophisticated and subtle much of human communication is. I told him for example, that people find eye contact engaging during a social interaction, so it is important to look at people when you listen to them and when you talk to them. He agreed to try it, but this resulted in him staring fixedly at me, just as he’d formerly fixed his gaze on the floor. I said, “Well, you don’t want to look at people all the time.” “Well, when do I look at them?” He wanted to know exactly how long to look.”

Fear responses, reflexes, and other less subtle movement patterns, are examples of motor structures we share with lower animals, insects, and less complex organisms. These are triggered by responses from much older parts of our brain such as the brain stem and lower and midbrain regions. Hyper vigilance, startle response, and other primitive responses to danger and stressors are those we share with many beings. They represent some of the foundations of movement itself.

The first animatronic robots moved in ways that resembled our insect ancestors. Human 02 asked me, 'what is the intersubjectivity of the robot?' The new sociable robots are DESIGNED to relate to others. To humans specifically...

Today, Human 02 and I discussed frameworks for the two main phases of this work with the PLEO. The first of these is grounded in modern attachment theory, outlined below. In my discussion with 02, he explains that the attachment response 'represents the deepest form of human longing' : 

Attachment is a special emotional relationship that involves an exchange of comfort, care, and pleasure between a primary caregiver and an infant. This relationship begins at birth but is thought to be critical by the age of six months (if attachment is not triggered by this age, future development is negatively impacted). All mammalian behavior develops from this primary relationship. In humans, emotional regulation, one’s internal sense of self as differentiated from others, sense of safety in the world, and the roots of ego development, depend on healthy bonding behaviors (triggered by over 100 ‘programmed’ infantile reflexes such as crying, screaming, facial mirroring, rooting for the breast or bottle, etc.) represented by the caregiver. Caregivers ‘bond’ by responding to these behaviors, infants ‘attach’ to caregivers based on these responses.

British developmental psychologist, John Bowlby devoted extensive research to the concept of attachment, describing it as a "lasting psychological connectedness between human beings" (Bowlby, 1969). In addition to this, Bowlby believed that attachment had an evolutionary component; it aids in survival. "The propensity to make strong emotional bonds to particular individuals [is] a basic component of human nature" (Bowlby, 1988). Our early attachment styles are established in childhood through the infant/caregiver relationship.

The Four Distinguishing Strategies of Attachment as a young child develops are:  

1. Proximity Maintenance – Maintaining a close physical distance to the attachment figure, 2.  Safe Haven - Returning to the attachment figure for comfort and safety in the face of a fear or threat. 3.  Secure Base - The attachment figure acts as a base of security from which the child can explore the surrounding environment freely. 4.  Separation Distress - Anxiety that occurs in the absence of the attachment figure.'   

In brief, the infant-child determines his or her behavior towards the caregiver based on their ability to act as a 'safe haven' (a place --almost a location in which the outside terrifying or unknown influences are not experienced). This location is calm, reciprocal, and warm in emotion. A safe haven, therefore, eventually creates a 'secure base' from which the child can explore the outside world bit by bit, and develop a sense of confidence through acting on objects in the area. 'Separation distress' occurs when the attachment figure leaves the surrounding area and the infant-child no longer feels a connection to this secure base. The child will scream, cry, wail, exhibit dystonic muscle spasms and postures (tension and struggle in the whole body), and feel totally out of sorts until the attachment figure returns. The infant-child tries to prevent these separations through the act of 'proximity maintenance' (staying close by while exploring).

When these strategies are employed successfully by the infant and the caregiver, the child will develop a 'secure attachment.' This lays the foundation for future secure strategies in bonding to others during the life course. Interruptions in or total absence of the use of these strategies due to the dysfunctional behaviors of the caregiver may result in one of the other three 'insecure' attachment forms.