Different Experiences, Different Types of Emergence:A-life Sculpture Designer, Interactant, Observer

Melanie Baljko† and Nell Tenhaaf‡†Department of Computer Science and Engineering, ‡Department of Visual Arts

York University4700 Keele St., Toronto, Ontario, Canada, M3J 1P3

mb@cse.yorku.ca, tenhaaf@yorku.ca

Abstract

A-life sculptures are dynamic art works in which hu-man participants function as co-creators. This paperdescribes a collaborative project to develop an A-lifeart work that has as its main component a heteroge-neous population of agents. The human interactant isrepresented within the agent population and their roleas co-creator of the artwork is mediated by their pres-ence in the population of agents. This work is con-cerned with issues of collaborative task performance,artistic co-construction, and the quality of emergence asbeing dependent on the experiencer of the system (i.e.,the system designer, interactant, and/or observer).

IntroductionThe application of social rules by humans when engagedwith computational media is of interest to both scien-tists and to New Media Artists — the phenomenon hasbeen demonstrated scientifically (Reeves & Nass 1996), ex-ploited in interface design by human–computer interactionresearchers (Cassell et al. 2000), and explored artisticallyin several innovative works by New Media artists, suchas Nell Tenhaaf’s UCBM (you could be me), 1999 andSwell, 2003; Elizabeth Vander Zaag’s Talk Nice, 2000; andNorman White’s Helpless Robot, 1987-96 (Tenhaaf 2002;2004).

In this paper, we describe a collaborative project presentlybeing conducted under the auspices of the New Media Ini-tiative program, which has been implemented by the NaturalScience and Engineering Research council (NSERC) and theCanada Council for the Arts. This program is intended topromote collaboration between scientists and/or engineersand artists, since New Media art that develops and appliesnew digital technologies increasingly includes science andengineering-based methodologies, and scientific research isincreasingly making use of artbased practices.

In this project, we are developing public interactive sce-narios. These scenarios are dynamic art works — the be-haviour of the work will depend on the behaviours andchoices made by the human interactant(s). Thus, the au-dience functions as co-creator. The central component of

Copyright c© 2006, American Association for Artificial Intelli-gence (www.aaai.org). All rights reserved.

these scenarios is a community of agents, which is madeobservable to the human interactants through sculptural, ab-stracted visual and auditory displays. A salient aspect ofthis project is the heterogeneity of the agent population. Thepopulation consists of both artificial, low-fidelity agents (fur-ther elaborated below) and of humanrepresentative agents.The human interactant’s role as cocreator of the artwork ismediated by their presence in the population of agents.

This work is guided by both New Media and scientificmotivations. With respect to the former, we are developinga more “true” interactivity that far surpasses the constraintsthat have been experienced in this domain — i.e., when theuser is left with the sense that a pre-arranged media rou-tine has been played back in response to her/his input, as-sembled from a menu of ready-made parts (Manovich 2001,p. 124), (Wilson 2002, p. 344). With respect to the latter,we are developing novel, concrete testbeds for the elicita-tion of interactions between artificial agents and human in-teractants, where the structure and characteristics of theseinteractions will subsequently be analyzed.

Background

System DescriptionSince our discussion below will turn to issues of emergentphenomena, and since conceptions of emergence typicallyrelate to properties that are systemic features of complexsystems, it is worth stating explicitly what is meant by theterm system in our work. In our context, the system includesvarious hardware and software components and the humaninteractant(s). The human is both an interactant and a com-ponent of the A-life sculpture, via their representative agentin the population. The A-life sculpture and the human inter-actant are both actively engaged with one another (attending,and possibly responding, to one another).

The term system should not be used interchangeably withthe term A-life sculpture, as the former refers to the sculp-ture with a particular set of one or more human interactantsengaged with particular behaviours in a particular context,whereas the latter might not. (The distinction is analogousto that made in the field of pragmatics between an utteranceand a sentence.)

Note that the human interactants, while being a compo-nent of the system, are also observers of the system. These

Figure 1: Swell [2003]

human interactants, who are experiencers of the artwork,will not necessarily share the same background knowledgeas the system designers — i.e., they are so-called naive ob-servers. Oftentimes in scientific work, the roles of systemobserver and system designer can be conflated or collapsed,since one of the primary purposes in designing a system isto observe and quantify its behaviour, possibly in the serviceof hypothesis testing. In this context, however, the roles ofdesigner and observer may be occupied by different individ-uals. This is a notable point of difference with other sys-tems consisting of heterogeneous populations of agents, forinstance the system described by Kennedy (2000) in whichthe human agent in the heterogeneous population was alsoa designer of the system and thus could be considered anexpert.

As Kennedy (2000) correctly noted, the artificial agentsin the population have a phenomenology, just as the humanones do. Note that the artificial part of the system also be-comes an observer, through processing and analysis of itsown observations of the interactions. This introduces a par-allelism between the human and artificial entities in the pop-ulation, as they each are both agent and observer. Other ob-servers of the system may include co-present visitors to theart space, as well as those such as us who will discuss andanalyze the interactions from video recordings.

Low-Fidelity Embodiment

A goal of this work is to develop artificial agents that havesophisticated behaviours, which are effected through the un-derlying agent architecture, yet have low-fidelity embodi-ments.

Figures 1 and 2 illustrate low-fidelity embodiments, a de-sign technique that was first demonstrated in Tenhaaf’s pre-viously exhibited dynamic A-life sculptures Swell [2003],and, more recently, Flo’nGlo [2005]. In Swell, the artificialentity reacted to the human interactant, though the popula-tion was not represented explicitly. Instead, low-fidelity wasintroduced in the form of low-resolution video of wave mo-

Figure 2: Flo’nGlo [2005]

tion shown in an array of light-emitting diodes (LEDs). Thisdisplay caused the interactant to search for clues as to thenature of the entity. In Flo’nGlo, the population was repre-sented, but was homogeneous. Humans were observers butnot interactants. Again, the emphasis was on low-resolutionvideo used to give character to the entities, although the out-put of an algorithm cued to the timing in Flo and Glo’s “con-versational” exchange was also visible in the LED display.Sound is very important in both of these works, it is a verystrong cue to character and it sets the mood more than anyother element.

In the low-fidelity design technique, agents are embod-ied as composites of electronic components, such as clus-ters of LEDs and 2-channel audio displays (stereo speakers).Such embodiments provide multiple degrees of freedom withwhich the agent can articulate behaviours (which corre-spond to the agent’s modes of articulation) and also providethe physical infrastructure for the mounting of sensors in anon-obvious manner (which provide the agent’s modes ofsensory-perception). Thus, such embodiments afford mul-timodal interaction (multimodal in the sense of having andmaking use of multiple modes of articulation and/or modesof sensory-perception). We consider such embodiments —described here as low-fidelity embodiments — to be prefer-able to high-fidelity embodiments, such as humanoid-like,digitally-rendered characters, because they circumvent thecliches and expectations attached to humanoid characters,avatars, or (even worse) cartoons. We are currently adapt-ing an existing agent architecture module (Baljko 2001a;2001b) that designs multimodal utterances to this new, low-fidelity context.

Low-fidelity embodiment has another advantage: its high-level of abstraction. Both a single agent and a population ofsuch agents can be embodied by the same physical infras-tructure. The human interactant distinguishes between thetwo cases on the basis of the behaviour of the articulators(whether the pixellated lights and audio displays cohere intoperceivable sub-units).

Figure 3: Swarm algorithm-based software prototype forcollaborative herding task.

Shared Task

Another goal of this work is for the public interactive sce-nario to afford to its interactants (both human and artificial)the possibility to perform a task. Of particular interest tous are tasks that require collaboration and must be sharedamong the agent population (to which the human interactantbelongs, via their representative agent). It is expected thatthe particular task embedded in the artwork may be modi-fied in subsequent prototype iterations, but for the currentprototype, a herding task has been implemented. A soft-ware prototype has been developed in which the agents arerepresented simply as two-dimensional circles. In the nextprototype iteration, the agent population will be given low-fidelity embodiments.

The software prototype is shown in figure 3. In this task,the agents in the population must direct the target entity intowhat we term the nest. In the figure, the nest is shown as ared circle, the target is shown as a blue circle and the agentsin the population are shown as cyan circles. The largercircles around each agent indicate each agent’s perceptualrange.

The behaviour of the target entity is implemented by thesimple rule to evade the agents. The behaviour of the artifi-cial agents is governed by a set of simple rules that expressattraction and repulsions to other entities in the arena (e.g.,attraction to the target and the nest, repulsion from closeproximity to other agents and the arena walls). A singleagent cannot perform this task in isolation (except in spe-cial cases), since the target robot simply evades the agent.However, when multiple agents are placed in the arena, they

collectively are able to herd successfully the target entity tothe nest.

The behaviour of the human-representative agent will beunder the control of the interactant. This mechanism in-volves a phase during which the representative agent en-trains itself to the interactant via tracking by overhead cam-era, and subsequent gesture mimesis (the agent moves whenthe interactant does, and the sound changes to reflect themovement).

Another task we are also presently implementing is onethat requires the population of agents to achieve a particularpattern of communication. The inspiration for this task isthe observation that complex systems of turntaking emergefrom relatively simple sets of rules (such as those identifiedby Sacks, Schegloff and Jefferson (1974)). The task is ofparticular interest as it will allow us to implement an agentarchitecture, which affords the generation and perception ofmultimodal “communicative acts”, and to explore the ideasof multimodal synergies in the low-fidelity embodiments.

Phenomena of InterestCo-ConstructionNew Media Artists have an established practice in the ex-ploration of co-construction. It is implicit in particular whenthey use interactive media, where the viewer has input intohow the work plays out. Interactive artworks are dynamic inthat the behaviour of such works depends on the behavioursand choices made by the human interactant. Examples in-clude: Nell Tenhaaf’s UCBM (you could be me) [1999], aninteractive video installation which explores viewers’ adap-tation to artificial empathy through a verbal “test” and agenetic algorithm that assesses their willingness to relate;Tenhaaf’s work Swell [2003], in which the viewer sets offan electronic sound feedback loop by approaching the pod-like sculpture — if the viewer backs away from it the soundbecomes very distressed, while soft melodic tones are thereward for coming closer; Norman White’s Helpless Robot[1987-96], a sculpture that asks for help in putting itself inmotion, and most often berates anyone who cooperates byletting them know they got it wrong; and, Talk Nice by Eliz-abeth Vander Zaag [2000], a work based on speech recogni-tion in which two teenage protagonists try to get the viewerto talk like they do and when they fail, punish them with ex-clusion from a cool party they are going to. In these works,even scolding messages to the viewer to tell them that theyare not performing well are met with enthusiasm by manyviewers. They elicit a sense of intersubjectivity between theartwork and the participant, and an understanding that co-construction of the experience by the interactant and the in-teractive work results in a unique event for each participant.Although such artworks are shown in traditional artspaces(e.g., galleries, exhibitions), they can also be installed in “ev-eryday” venues.

When new media artwork is explicitly based on princi-ples from A-life, it invokes a tension between control andunpredictability, because the autonomy that is attributed tothe system suggests that unexpected behaviour will emerge.In “Tom Ray’s Hammer: Emergence and Excess in A-Life

Art”, Mitchell Whitelaw (1998) is hesitant to take the term“emergence” at face value because it suggests the ultimatelyfutile quest in technologically-driven artwork for greater andgreater originality, surprise and other features of what hecalls ”emergent excess”. But Whitelaw does build a charac-terization of emergent phenomena by focusing on the expe-rience of emergence (after the theorist Peter Cariani), that is,on the “point of view that constitutes the observer’s model ofthe system”. In Whitelaw’s view, emergence is in the phys-ical system’s deviation from this preconceived model. Fur-ther, this recognition must then be taken into account withinthe further development of the model by its maker. The pro-grammer/ builder of the model and a completely naive ob-server would be two very different kinds of interactants, andthe experience of emergence will always vary depending onwho this is.

Attribution of AgencyYet another phenomenon of interest in this project is attribu-tion of agency by the human interactant to artificial entitiesin A-life artworks. In this current project, we will focus onthe attribution of agency by the human interactant. This fol-lows from explorations of other types of attributions. Ten-haaf’s integration of A-life research into art practice sincethe mid-1980s has been greatly influenced by Socially In-telligent Agents research that brings in methods from thehumanistic sciences. In particular, the use of psychologicalconcepts such as empathy has informed the approach to in-teractivity. If empathy is attributed to a system, as a functionof its design and through the type of feedback that it givesto an interactant, then the interactant can in turn call uponempathy as a framework for interpreting the exchange thattakes place. Kerstin Dautenhahn’s work on the phenomeno-logical dimensions of social understanding sheds light onhow empathy can be modeled in this way. Her research isgeared toward taking into account “the role of the humanobserver . . . as an active, embodied agent who is biased to-ward interpreting the world in terms of intentionality and ex-planation.” (Dautenhahn 1997). The observer (or interactantif communication is involved) comes with interpretative ex-pectations, and this is why empathy is useful as a paradigmbecause it is largely a function of mind. It sounds like anemotional state because it is activated as a bodily, experien-tial dynamics between two entities, using cues of language,gesture or expression. But empathy is a process of knowingthrough imaginal and mimetic capacities — imagining the“state of mind” of the other. Because its purpose is not todefine the other entity nor to convey a definition of humansubjectivity, but to flesh out a communicative exchange, em-pathy occurs as an instance of social learning in a space thatlies between the beings involved.

Other researchers reiterate the significant role that empa-thy can play. For instance, Claus Emmeche suggests thisas a possible methodology within natural science for bal-ancing objective modes of thought with an “internalist wayof having emergent explanations” for non-human living or-ganisms, that is, as a way to express the nonhuman agent’spoint of view (Tanhaaf 2001, p. 117). In general, the needto articulate psychological or cognitive concepts that can be

perceived as common to many types of entities is key tohow attribution within a human/non-human interaction canbe made and understood.

Task Performance: Autonomous vs. CollectiveIn his “exteriorized particle swarm” experiment, Kennedyused a particle swarm algorithm that he introduced in 1995for function optimization (PSO), as the basis of a learn-ing environment in which a human performed a problem-solving task together with the computational agents of theswarm (Kennedy 2000). The experiment showed that thehuman performs as well as the artificial agents on a simpleproblem, but worse than them on a harder problem. Kennedyqualifies his use of the term agent, since the particles are notautonomous. Nonetheless, the particles seem to exhibit asocial behaviour as they interact with each other, and in thisexperiment with the human as well, to find the best solutionto the problem given them.

The particles in the algorithm are problem solution vec-tors in a search space, and they learn from experience byadjusting their velocity vector elements toward each newbest solution that is generated. They also take advantageof social learning, in that individuals adjust their velocitiestowards the best solutions found by their neighbours. Thehuman who is introduced as an actor in the “exteriorized”world (i.e., outside the electronic world of the computer butin conjunction with it) is shown the most successful solu-tions of the computational agents at each iteration and is in-vited to propose one of his or her own. This person has thesame information that the particles receive from each otherand produces the same kind of input into the system. Inhis own first-person account of experiencing the exterior-ized particle swarm, Kennedy notes that his feelings abouthis relation to the particles’ successes tended toward com-petitiveness, while his analysis of the results (resenting theirbetter results and enjoying their emulation of his good re-sults) is that the overall collective behaviour of the system isan example of a social drive for improvement. The experi-ment provides insight into sociocognitive behaviour throughsimulating such behaviour, and allowing the human partici-pant to study their own involvement.

Expert vs. Novice User/InteractantOur system will also be a heterogeneous collection of humaninteractant and artificial entities. But the interactant will be anovice user. Kennedy is the ideal user of his system, since hefills all three roles of designer, interactant and observer andis therefore easily able to interpret the emergent behaviourof his system. The novice user can be readily guided bylanguage or other overt cues, as in UCBM (you could be me),Swell, Talk Nice, and the Helpless Robot described above.We propose that low fidelity cues such as light, colour andmovement can be used to enable interactants to recognizeemergent behaviours in the system, within the context of acomplex, multimodal artwork.

In Kennedy’s account, he as an expert could readily un-derstand that what at first appeared to be competitive be-haviour of the group led to an end result of mutual accom-plishment. We propose that the naive user can be made

Figure 4: Autopoiesis [2000]. Image uses with permissionof artist.

aware of such a shift in perception through the artistic pro-cess. That is, all aspects of the A-life sculpture (including,and perhaps especially, its more intuitive and subtle features)can be put into the service of enabling an interactant to graspthe emergent features of their experience with the work.

An excellent example of such a shift in perception broughtabout by an A-life artwork is Ken Rinaldo’s robotic sculp-ture installation Autopoiesis [2000], shown in figure 4. Inter-actants walk among a group of fifteen robotic sound sculp-tures whose behaviour collectively changes over time. Eachsculpture, suspended from the ceiling, can individually de-tect and respond to an interactant through smart sensor or-ganization (passive infrared sensors), moving its tip towardthe person but stopping just short of touching them. At thesame time the entire group sends its data to a central statecontroller for coordination of group behaviour. The fasci-nating shift in understanding that occurs for an interactantis from defensiveness at the initial “probing” of a sculpturetoward them, to feeling part of the overall gorgeous balleticmovement of the group. This is a direct, intuitive experienceof emergence.

Current Challenges and DiscussionElicitation without Overt ExplanationRules are built into the system by its designers, but they arediscovered by the human participants as they interact withthe A-life sculpture, rather than being made explicit. Thisprocess results in a co-construction of the artwork, in thesense that experience of the work is different for each par-ticipant, and many facets of the work are not immediatelyavailable but appear during the time spent with it. Arisingfrom this is one of the central challenges of this project:the need to devise means by which interactants can recog-nize and interpret emergent phenomena, without recourseto overt explanations using language. The means for inter-pretation should be interesting to both artistic and scientificquestions. In particular, can we use low-fidelity means?

Status of Embodiment as an Experimental Factor

Methodological Issues for Evaluation One of the aspectsof this project is scientific hypothesis testing. Of interestis our hypothesis that even low-fidelity embodiments canafford sophisticated behaviours (behaviours that afford theattribution of agency). A challenge for us is to devise anappropriate methodology for testing this hypothesis. Con-sider the experimental paradigm, for which we would de-vise two conditions for the system (low-fidelity and high-fidelity embodiments), and suppose we were to allow thesystem to function under each. Under each condition, thepatterns of behaviour by the human interactant and A-lifesculpture will emerge; this is the emergent behaviour of thesystem. Assume that we have induced an adequate numberof interactions under each condition so as to characterize thevariation in the system’s behaviour under each condition. Isit methodologically sound to attribute differences in the re-sulting patterns of interaction to the differences in the con-ditions?

The answer boils down to whether one believes that thesystem behaviour can be explained as a function of the an-tecedent factors. This would fly in the face of both the“predictive” and “irreducible-pattern” views of the emer-gent behaviour of the system (i.e., the emergent patternsof behaviour by the human interactant and A-life sculp-ture) (O’Connor & Wong 2005). Both views hold that emer-gent properties are systemic features of complex systems.How they differ is that the former holds that these systemicfeatures cannot “be predicted from the standpoint of a pre-emergent stage” (even if “given a thorough knowledge ofthe features of, and laws governing, their parts”), whereasthe latter holds that these systemic features are “governedby true, lawlike generalizations within a special science”(but where this special science is “irreducible to funda-mental physical theory” and the features in question “can-not be captured in terms of the concepts and dynamics ofphysics”) (O’Connor & Wong 2005).

Our challenge is determining how to do scientific eval-uation without falling prey to the fallacy that the emergentinteraction is a systematic product of the components (suchas the type of embodiment and other features such as attribu-tion or turn-taking that have been built in). These can be de-scribed as follows: if the novice human user imitates or oth-erwise gets in synch with the computational agent, then thereis strong attribution of agency, especially in the form of in-tentionality. But because the human interactant’s behaviourhas an effect on, but does not solely determine the computa-tional agents’, and vice versa, isn’t there ipso facto a recog-nition of emergent social dynamic — which is not a com-ponent per se, even if Kennedy says that social behaviours(especially imitating) informed the design of the PSO. Per-haps the results of applying the experimental methodologyitself can serve as the criteria for emergence.

The Role of Embodiment In a recent article, the differentideas of embodiment that operate in cultural discourse andA-life modeling were contrasted (Tenhaaf 2000, pp. 241–242): in the former it refers to bridging the mind-body splitby reintegrating emotion that arises from senses other than

the “disembodied eye”, and in the latter it means buildingagents that are autonomous and responsive to their environ-ment. The article goes on to discuss how A-life art can leadto a kind of selfhood endowed upon an artificial agent viaexchanges that allow emotional states to enter for the humaninteractant. While art has historically focused on emotionalreaction as a key aspect of the viewing experience and itsinterpretation, art that involves A-life or AI systems wouldseem to rely on their more functional definition of embod-iment (autonomy) — even if emotions are part of the art-work’s theme. The artist Van der Zaag discussed in this ar-ticle builds an embodied model that calls on both domains,art and A-life. By having the interactant submit to the ra-tionalized authority of the technological system through thephysical setup of the work and through chiding language,the work prompts the interactant to let go of the ego-self andlet in the emotional realm that hovers outside any orderedmodel. Evaluating this work as an artwork is based on thebodily set of responses invoked, rather than on assessing theautonomy of the system, which is attributed by the interac-tant.

Relationship among Embodiment, Attribution ofIntelligence, and Attribution of AgencySeveral have investigated the idea of embodiment being anecessary condition for intelligence (an artifact must havean embodiment in order to be intelligent). For instance,what kind of artifact could be considered to be embodied (cfwhat kind of embodiment does a particular cognizer have?).Ziemke’s (2005) discussion of the different types of embod-iment seems to be motivated by the need to determine mini-mal criteria for an artifact to be intelligent (have cognition).This follows from the assertion that embodiment is an indis-pensable and essential condition of intelligence (natural orartificial). Of interest to us is the degree to which human in-teractants perceive embodiment being a sufficient conditionfor agency (which could be construed as an extremely weakform of intelligence).

The proponents of organismic embodiment would assertthat low-fidelity agents do not meet the minimum criteriafor intelligence. These proponents hold that cognition iswhat living systems do in interaction with their environment,where living systems are necessarily autonomous and au-topoietic. Low-fidelity agents, on the other hand, have aninteraction with their environment, but are acting accordingto mechanisms that are heteronomous (the rules that guidetheir operation are not their own’ but human rules, whichhave been built into the machine, and therefore also can bechanged only by humans). Moreover, they are allopoietic(made by a system other than itself; a human system, in thiscase). Our project presents an interesting test-bed for obtain-ing empirical data for questions that are typically pursued intheoretical frameworks.

Different Conceptions of Emergent BehaviourJudging from the philosophical and artistic literature, theterm emergence has not converged to a single notion (andmay never do so). We have not yet established a connec-tion between the literature’s characterizations of the various

types of emergence (epistemological, ontological, aesthetic,even metaphysical) and the behaviours we have described.One point of consensus, however, if that emergence is some-thing more than the mere emersion or appearance of be-haviours/properties.

We have discussed two sorts of behaviours that might beconsidered emergent: the behaviour of the system, from theperspective of an observer engaged in the process of co-construction (an artistic process), and the behaviour of thesystem as it arrives at a collaborative solution to a giventask. We have described how the artistic process of co-construction is predicated on the human interactant’s recog-nition of their representative in the agent population. (Actu-ally, the realization by the human interactant that he or sheis a member of the agent population alone is not enough, asthat would be an internal, mental state of the human interac-tant. Once the human takes action that is predicated on sucha mental state, this, in turn, becomes system behaviour). Itdoes not necessarily follow that co-construction will occuronly if the human interactant understands the effect of hisor her behaviour on the population of agents, his or her rolein the emergent behaviour, or even that he or she has a rolein the art work. Co-construction may yet emerge for otherobservers of the system.

In our work, we have found utility in the following logi-cal test in order to characterize emergence — does a fallacyof division arise? By fallacy of division, we mean the situ-ation in which one cannot reason logically that “somethingtrue of a thing must also be true of at least some of its con-stituents”. In other words, for a property or behaviour to beemergent of the system, it cannot also be found to be a prop-erty or behaviour of its constituents. Consider the propertyof a system evoking a realization by the observer that someresults are emersive behaviours of agent populations and notattributable to complex sets of rules. If the system evokesthis, this is surely an emergent behaviour; none of the sys-tem components on their own could do so.

Artistically, this could be the “moral” of the story or whatan observer of the A-life sculpture might take away fromthe interaction. Do not be waylaid that the realization itselfconcerns emergence; the artistic component of this projectconcerns, in a sense, second-order emergence.

A challenge posed to the participants of this symposium isto take this opportunity “to theorize hybridity formed at thejunction of human- and non-human”. The “space between”that our A-life artwork presents to all types of particpants isthe locus for the phenomena that we have characterized asemergent. Each participant applies interpretive biases, andif there is flexibility toward the unknown (which artificialentities tend to be, almost by definition), we come to knowthe other. Note that for both the making and analysis of art-works, making both the designer and observer of a systemaware of their pre-existing biases is an interesting accom-plishment in itself.

Acknowledgements The authors would like to expressthanks to the reviewers for Natural Sciences and Engineer-ing Research Council (NSERC) and Canada Council forthe Arts for their detailed and insightful analysis of theproject proposal “A-Life Sculpture: Eliciting Complex In-teractions”. The authors also gratefully acknowledge thejoint support of these two funding bodies for the workdescribed here. The sound artist for the project is JohnKamevaar. Thanks also to graduate student David Jacobfor his work developing the software simulation of the agentpopulation performing the herding task. The authors also arethankful for the feedback and comments of the reviewers.

ReferencesBaljko, M. 2001a. Articulatory adaptation in multi-modal communicative action. In Thompson, C.; Paek, T.;and Horvitz, E., eds., Proceedings of the North AmericanChapter of the Association for Computational Linguistics(NAACL) Workshop on Adaptation in Dialogue Systems.Pittsburgh, PA.

Baljko, M. 2001b. The evaluation of microplanning andsurface realization in the generation of multimodal acts ofcommunication. In Pelachaud, C., and Poggi, I., eds., Pro-ceedings of the Workshop on Multimodal Communicationand Context in Embodied Agents, Fifth International Con-ference on Autonomous Agents, 89–94. Montreal, Canada.

Cassell, J.; Sullivan, J.; Prevost, S.; and Churchill, E., eds.2000. Embodied Conversational Agents. MIT Press.

Dautenhahn, K. 1997. I could be you: The phenomenolog-ical dimension of social understanding. Cybernetics andSystems 28(5).

Kennedy, J. 2000. Human and computer learning togetherin the exteriorized particle swarm. In Dautenhahn, K., ed.,Proceedings of the 2000 AAAI Fall Symposium on SociallyIntelligent Agents: The Human in the Loop. North Fal-mouth, MA: AAAI Press. Available as Technical ReportFS-00-04.

Manovich, L. 2001. The Language of New Media. Cam-bridge, MA; London, England: MIT Press.

O’Connor, T., and Wong, H. Y. 2005. In Zalta, E. N., ed.,The Stanford Encyclopedia of Philosophy. MetaphysicsResearch Lab, Center for the Study of Language and In-formation, Stanford University.

Reeves, B., and Nass, C. 1996. The Media Equation: HowPeople Treat Computers, Television and New Media LikeReal People and Places. Cambridge University Press.

Sacks, H.; Schegloff, E. A.; and Jefferson, G. A. 1974. Asimplest systematics for the organization of turn-taking inconversation. Language 50:696–735.

Tanhaaf, N. 2001. Where surfaces meet: Interviews withStuart Kauffman, Clause Emmeche and Arantza Etxeber-ria. Leonardo 34(2):115–120.

Tenhaaf, N. 2000. Perceptions of self in art and intelligentagents. In Dautenhahn, K., ed., Proceedings of the 2000AAAI Fall Symposium on Socially Intelligent Agents: TheHuman in the Loop. North Falmouth, MA: AAAI Press.

Tenhaaf, N. 2002. Perceptions of self in art and intelligentagents. In Dautenhahn, K.; Bond, A. H.; Canamero, L.; andEdmonds, B., eds., Socially Intelligent Agents: CreatingRelationships with Computers and Robots. Kluwer. 235–242.Tenhaaf, N. 2004. Artificial agency. In Proceedings of theArt and Biotechnology Colloquium. Concordia Universityand UQAM, Montreal, PQ: Musee d’Art Contemporain.Publication in French, English to follow.Whitelaw, M. 1998. Tom Ray’s Hammer: Emergence andexcess in a-life art. Leonardo 31(4):345–348. Special Dig-ital Salon Issue.Wilson, S. 2002. Information Arts: Intersections of Art,Science, and Technology. MIT Press.Ziemke, T. 2005. What’s that thing called embodiment. InProceedings of the 25th Annual Meeting of the CognitiveScience Society. Lawrence Erlbaum.