The Artificial Intelligence (AI) Hermeneutic Network: Toward an Approach to Analysis and Design of Intentional Systems

‘I felt that I should be able to get the computer to sound
good more or less on its own, so that someone listening
to it says, “Who is that playing?” But if you get “What’s
that?” instead, you have to go back to the drawing board.’
(Lewis, 2000)
Abstract Digital information technologies are increasingly
being adopted in the humanities as both research
tools and supports for new forms of cultural expression.
Some of these digital technologies, in particular artificial
intelligence (AI) programs, exhibit complex behaviors
usually seen as the territory of intentional human phenomena,
such as creativity, planning and learning. This
paper identifies a prototypical subset of these programs,
which we name intentional systems, and argues that their
seemingly intentional behaviors are not the sole effect of
underlying algorithmic complexity and knowledge engineering
practices from computer science. In contrast, we
argue (paralleling the field of software studies) that intentional
systems, and digital systems at large, need to be
analyzed as a contemporary form of historically, culturally,
socially, and technically situated texts. Perception
of system intentionality arises from a network of continuous
meaning exchange between system authors’ narration
and users’ interpretation processes embedded in a
broader social context. The central contribution of this
paper is a new interdisciplinary analytical framework
called the AI hermeneutic network that is informed by
traditions of hermeneutic analysis, actor-network theory,
cognitive semantics theory, and philosophy of mind. To
illustrate the design implication of the AI hermeneutic
network, we present our recent work Memory, Reverie
Machine, an expressive intentional system that generates
interactive narratives rich with daydreaming sequences.
Intentional Systems
Trombonist and composer George Lewis’s above description of his interactive musical system Voyager exemplifies
a growing number of digital systems, such as
the autonomous painting program AARON (Cohen 2002)
and recent computational narrative works (Harrell 2006;
Mateas & Stern 2002; PŽrez y PŽrez & Aliseda 2006),
that utilize AI techniques in pursuit of cultural expression.
Decades after heated debates about the feasibility
of AI, the question of whether computers may one day
possess human-level intelligence no longer spurs society’s
fear and curiosity. Instead, systems are designed to
encourage users to make sense of them as intentional and
independent entities. Compared to instrumental, production-
oriented programs such as the PhotoShop, these
systems display intentional behaviors related to human
mental phenomena such as planning, learning, narrating,
and creating, as if their actions were about something in
the world (Searle 1983) rather than mere execution of
algorithmic rules. Lewis, for instance, insists that Voyager
‘not [be] treated as a musical instrument, but as an
independent improviser.’ He deliberately designed the
system to display independent behaviors arising from
its own internal processes that even its designer cannot
fully anticipate. The improvisational dialogue between
Voyager and the musicians, Lewis emphasizes, is ‘bi-directional
transfer of intentionality through sound.’ computational
complexity, 2) process opacity, 3) human-like
coherent behaviors, and 4) execution of authorial intention.
The term encompasses not only AI systems but also
AI-like systems that exist either outside of computer science
communities or are not described by their authors
as AI systems for ideological or other reasons. Critical
analysis and design of intentional systems, like information
technologies at large in the digital humanities, calls
for the recognition of these systems as important forms
of cultural production, beyond their traditionally instrumentalized,
productivity oriented roles.
Intentional Systems as Texts
Although generally used to describe written forms of discourse,
the term text as the object of literary theory and
modern hermeneutics is not confined to only linguistic
forms. In his essay on the literary text, German philosopher
Manfred Frank (Frank 1989) criticizes the notion
that meanings that authors encode within texts can be
objectively retrieved without distortion by readers given
appropriate methods of interpretation (Hirsch 1967). Instead,
Frank proposes a complex communication process
in which both author and reader actively create, shape,
and reconstruct meanings. This echoes the even broader
notion of dialogic meaning posited by the Russian philosopher
and critic Mikhail Bakhtin in which language
is understood as dynamic, contextual, intertextual, and
relational (Holquist 1990). Acknowledging the textuality
of intentional systems opens up understanding of system
intentionality to a range of socially situated methods.
Intentional systems are not simply the result of clever
algorithmic and data structural innovations. The AI
practitioner and theorist Philip Agre cogently points out
that the ‘the purpose of AI is to build computer systems
whose operation can be narrated using intentional vocabulary.’
(Agre 1997) Michael Mateas, co-developer of Façade,
further deconstructs the codes invoked in AI practice
by computation, and definitions of system progress)
and the co-existing ‘code machine’ (including physical
processes, computational processes, and complex causal
flow), in order to pin down the long-neglected social and
discursive aspect of AI systems (Mateas 2002). In addition
to considering actual computer programs, analysis
of intentional systems should not omit the authors’
publications, presentations, and interpersonal communication
about the system. Such narrative outputs situate
the system in AI research communities and frame users’
interpretation, and therefore must be considered as part
of the intentional system.
The AI Hermeneutic Network
The central contribution of this paper is the AI hermeneutic
network model, enabled by theorizing intentional
systems as texts. The interdisciplinary framework analyzes
system intentionality as a result of a hermeneutic
communication process that involves both authors’ narrations
and users’ interpretations through interaction
with both actual systems and authors’ narrative output.
In addition, this paper recognizes that intentional systems
exist in broader social contexts that involve more
than just authors and users. Animate and inanimate actors,
called ‘actants’ in actor-network theory (Callon
1986; Latour 1996), participate in the network through
multi-directional communication. Government and military
funding, for instance, often plays a prominent role in
determining direction and validity of different approaches
of AI research.
Historically, hermeneutic studies developed interpretative
theories and methods in order to recover the meanings
of sacred texts intended by the (divine) author(s).
Modern hermeneutics, influenced by Schleiermacher,
recognizes that everything calls for the work of interpretation
and broadens itself to the philosophical interrogation
of interpretation (Virkler 1981). This paper
highlights discursive ‘elasticity’ of the AI key words,
such as planning (Agre 1997). He observes that these
key terminologies are simultaneously precise (formal)
and vague (vernacular), which allows AI practitioners
to seamlessly integrate their everyday experience as embodied
intentional being in the algorithmic research, and
to narrate computation with popularly accessible vernacular vocabulary.
One relatively unexplored aspect of this continuous negotiation
of values and meanings between both human
and computational actors (Latour 1996) is users’ readings
and interpretations of intentionality from systems
that are clearly inanimate. For example, human coperformers
and their audiences’ interpretations of Voyager’s
behaviors as intentional are central to construe the
systems’ status as an independent performer in its own
right, as intended by its designer. Frank argues that ‘[i]
n the understanding of its readers the text … acquires a
meaning which exceeds the memory of its origin.’(Frank
1989) Any analysis of system intentionality then is not
adequate without considering participation of users and
audiences.
This paper emphasizes the discursive strategy and semantic
interpretation from a cognitive linguistics perspective.
Conceptual blending theory (Fauconnier 2001;
Fauconnier & Turner 2002; Turner 1996) offers a cognitive
foundation for understanding system intentionality
as actively (re)constructed by users via integrating concepts
of intentionality based on encounters with animate
agents, and conceptualization of algorithmic operation
of inanimate computer systems. Thus, users compress
the behavior of unfamiliar computational systems to human
scale by constructing conceptual blends of systems
with human-like intentionality, through semantic hooks
that facilitate such blends in the various discourses surrounding
the systems.
Conclusion: Design Implications of the AI
Hermeneutic Network
The novel framework of the hermeneutic network suggests
new design approaches for intentional systems in
digital humanities. Our current interactive narrative work
Memory, Reverie Machine generates stories in which the
main character varies dynamically along a scale between
a user-controlled avatar with low intentionality and an
autonomous non-player character with high intentionality.
By algorithmically controlling the semantic hooks
for interpreting system behavior as intentional in the narrative
discourse (Zhu & Harrell 2008), the authors turn
system intentionality into a scalable expressive dimension
in interactive storytelling (Harrell & Zhu 2009).
In conclusion, this paper proposes a new interdisciplinary
framework to analyze intentional systems as social
and cultural productions, as opposed to construing them
as the domain of purely technical practices. It underlines
authors’ narrative and users’ interpretative strategies, in a
socially situated network of meaning exchange. Finally,
through our own computational work we suggest new
design implications for intentional systems, such as the
scale of intentionality (Zhu & Harrell 2008) that potentially
can add new forms of expressivity to intentional
systems in digital humanities.
References
Agre, P. E. (1997). ‘Toward a Critical Technical Practice:
Lessons Learned in Trying to Reform AI’, in Social
Science, Technical Systems, and Cooperative Work:
Beyond the Great Divide, eds G. C. Bowker, S. L. Star,
W. Turner, L. Gasser & G. Bowker, Lawrence Erlbaum
Associates, pp. 131-58.
Callon, M. (1986). ‘Some Elements of a Sociology of
Translation: Domestication of the Scallops and the Fishermen
of St Brieuc Bay’, in Power, Action and Belief: A
New Sociology of Knowledge, ed. J. Law, Routledge &
Kegan Paul, London.
Cohen, H. (2002). ‘A Self-Defining Game for One
Player: On the Nature of Creativity and the Possibility
of Creative Computer Programs’, Leonardo, vol. 35, no.
1, pp. 59-64.
Fauconnier, G. (2001). ‘Conceptual Blending and Analogy’,
in The Analogical Mind: Perspectives from Cognitive
Science, eds D. Gentner, K. J. Holyoak & B. N.
Kokino, MIT Press, Cambridge, MA.
Frank, M. (1989). The Subject and the Text: Essays on
literary theory and philosophy, Cambridge University
Press, Cambridge.
Harrell, D. F. (2006). Walking Blues Changes Undersea:
Imaginative Narrative in Interactive Poetry Generation
with the GRIOT System, paper presented to AAAI
2006 Workshop in Computational Aesthetics: Artificial
Intelligence Approaches to Happiness and Beauty, Boston,
MA, AAAI Press.
Hirsch, E. D. (1967). Validity in Interpretation Yale
University Press, New Haven and London.
Holquist, M. (1990). Dialogism: Bakhtin and His
World, 2 edn.
Latour, B. (1996). Aramis, or the Love of Technology,
Harvard University Press, Cambridge.
Mateas, M. (2002). Interactive Drama, Art, and Artificial
Intelligence, Carnegie Mellon University.
Searle, J. (1983). Intentionality: An Essay in the Philosophy of Mind, Cambridge University Press, Cambridge.
Turner, M. (1996). The Literary Mind: The Origins of
Thought and Language, Oxford UP, New York; Oxford.
Virkler, H. A. (1981). Hermeneutics: Principles and
Processes of Biblical Interpretation, Baker Book House,
Grand Rapids, MI.
Fauconnier, G. & Turner, M. (2002). The Way We
Think: Conceptual Blending and the Mind’s Hidden
Complexities, Basic Books.
Harrell, D. F. & Zhu, J. (2009). Agency Play: Dimensions
of Agency for Interactive Narrative Design, paper
presented to to appear in the Proceeding of AAAI Spring
2009 Symposium on Interactive Narrative Technologies
II.
Mateas, M. & Stern, A. (2002). ‘A Behavior Language
for Story-Based Believable Agents’, IEEE Intelligent
Systems, vol. 17, no. 4, pp. 39-47.
Pérez y Pérez, R. & Aliseda, A. (2006). The Role of
Abduction in Automatic Storytelling, paper presented to
Proceedings of the AAAI workshop in Computational
Aesthetics: AI Approaches to Beauty & Happiness, Boston,
MA, AAAI Press, pp. 53-60.
Zhu, J. & Harrell, D. F. (2008). Daydreaming with Intention:
Scalable Blending-Based Imagining and Agency
in Generative Interactive Narrative, paper presented to
AAAI 2008 Spring Symposium on Creative Intelligent
Systems, Stanford, CA, AAAI Press, pp. 156-62.