Full Terms & Conditions of access and use can be found at
https://www.tandfonline.com/action/journalInformation?journalCode=raag21
Annals of the American Association of Geographers
ISSN: (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/raag21
Evolving Coagency between Artists and AI in the
Spatial Cocreative Process of Artmaking
Paulina Nordström, Riina Lundman & Johanna Hautala
To cite this article: Paulina Nordström, Riina Lundman & Johanna Hautala (2023): Evolving
Coagency between Artists and AI in the Spatial Cocreative Process of Artmaking, Annals of the
American Association of Geographers, DOI: 10.1080/24694452.2023.2210647
To link to this article:  https://doi.org/10.1080/24694452.2023.2210647
© 2023 The Author(s). Published with
license by Taylor & Francis Group, LLC
Published online: 07 Jul 2023.
Submit your article to this journal 
Article views: 230
View related articles 
View Crossmark data
Evolving Coagency between Artists and AI in the
Spatial Cocreative Process of Artmaking
Paulina Nordstr€om,


Riina Lundman,† and Johanna Hautala‡


Department of Global Development and Planning, University of Agder, Norway
†Department of Geography and Geology, University of Turku, Finland
‡School of Management, University of Vaasa, Finland
This article applies theoretical and empirical discussions of emerging human and digital technology relations to
our interest in collaborative artist–artificial intelligence (AI) artmaking processes. Thus far, the theoretical
focus has largely been on mediating (code) and merged (cyborg) human–technology relations, with mutual
(coagency) relations yet to be adequately explored. To address this, we nuance the theoretical discussion and
extend the empirical research, analyzing the spatial cocreative artmaking process through video interviews with
eighteen Finnish artists using AI. Drawing on the work of Barad, we regard humans and AI as fundamentally
entwined, receiving their agencies through intra-action. Building on this, we demonstrate how the agencies of
artists and AI emerge and mutually evolve across three stages of the creative process: (1) coding and data, (2)
learning and training, and (3) curating the outcome. Thus, through our empirical research on how artist and
AI create new material and meaningful artworlds, we are able to nuance understanding of coagency as a spatial
process. Key Words: agency, art, artificial intelligence, cocreativity, digital geographies.
R
apidly developing digital technologies are pro-
foundly transforming our world. Automatic
digital technologies have already come under
extensive scrutiny in digital geographies (Thrift and
French 2002; Dodge and Kitchin 2007; Kitchin and
Dodge 2011; M. Graham, Zook, and Boulton 2013;
Pink and Fors 2017; Thulin, Vilhelmson, and
Schwanen 2020), and autonomous code-embedded
technologies—such as artificial intelligence (AI) and
robots—are eliciting increasing interest (Del Casino
2016; Batty 2018; Chen, Marvin, and While 2020;
Cugurullo 2020; Bissell 2021; Macrorie et al. 2021).
As such, there is a pressing need to investigate the
creative processes taking place between humans and
digital technologies, with the role of AI a particu-
larly novel topic for geographers (Lynch and Del
Casino 2020; Lynch 2022; Wingstr€om, Hautala, and
Lundman 2022; Lundman and Nordstr€om 2023).
Rather than there being a single form of AI, a
variety of AIs are employed for different purposes,
with humans—or as we understand it, human–AI
relations—required for such AIs to participate in
creative processes (see, e.g., Del Casino et al. 2020).
Although artificial can be taken to refer to a nonhu-
man, a technological subject, or both, the term
intelligence has been much debated (Lynch and Del
Casino 2020), and is often considered a human skill
or capacity not yet existent in machines.
Nonetheless, AI involves a display of human-like
behaviors, such as learning, perception, and deci-
sion-making (Bostrom 2017; cf. Cugurullo 2020;
Russell and Norvig 2020). Compared to automation,
where digital technologies simply follow instructions
(see Thrift and French 2002; Dodge and Kitchin
2007, 2009), AI is capable of learning autonomously
(Cugurullo 2020; de Vries 2020; Forbes 2020),
meaning actions can change during the process.
Here, autonomous means human supervision of the
learning process is not required (Cugurullo 2020;
Santos et al. 2021; cf. Mitchell 1997). AI’s capacity
to “exercise choice and make decisions” (Hayles
2017, 31) contains the potential to produce novel
and surprising outputs, including within art and
other creative practices (Miller 2019).
In this article, we study the spatial cocreative art-
making process undertaken between artists and AI.
To do this, we bring digital (Ash, Kitchin, and
Leszczynski 2018a) and creative geographies
(Hawkins 2017a) into contact with the posthuman
thinking of Barad (2003, 2007, 2012, 2015; see also
# 2023 The Author(s). Published with license by Taylor & Francis Group, LLC
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits
unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. The terms on which this article has been published allow the
posting of the Accepted Manuscript in a repository by the author(s) or with their consent.
This article has been corrected with minor changes. These changes do not impact the academic content of the article.
Annals of the American Association of Geographers, 0(0) 2023, pp. 1–16
Initial submission, November 2021; revised submissions, September 2022 and February 2023; final acceptance, April 2023
Barad and Gandorfer 2021). Barad’s strength lies in
how she conceptualizes the relationship between
agencies in the world-forming processes, with her
basic unit of interest the phenomena through which
agencies develop their capacities to act. Although
Barad’s thinking is gaining momentum within
research investigating digital technologies (Draude
2020; Lipp and Dickel 2022) and creative agencies
(A. Harris 2018, 2021), it has yet to be widely
applied in digital geographies (Rose 2017, 780).
Creativity and interaction with humans are central
motivations and challenges when it comes to develop-
ing AI (Boden 2009; Colton and Wiggins 2012). As
such, we propose that the cocreative artmaking pro-
cess between humans and AI is a promising case with
which to study human–technology relations, as well
as to explore how such relations might be conceptual-
ized in digital geographies (see also Wingstr€om,
Hautala, and Lundman 2022; Lundman and
Nordstr€om 2023). We interpret the spatialities of art-
making as artworlds, referring to the two-way relation-
ship whereby artworks make worlds (i.e., how art
performs in the world) and worlds make artworks (i.e.,
how the world in which art is made shapes art;
Hawkins 2017b, xvii–xviii). The concept of cocrea-
tivity is drawn from the field of computational crea-
tivity, which is concerned with how humans and AI
interact, collaborate, and facilitate within a creative
process (Davis 2013; Feldman 2017). By definition, a
creative process involves not just artmaking’s out-
come-oriented practices, which are aimed at generat-
ing meaningful ideas or artifacts (cf. Mace and Ward
2002; Boden 2009; St-Louis and Vallerand 2015), but
the affective, material, and relational events arising
between humans and nonhumans, which hold the
potential for continuous change in the world
(Williams 2016; Nordstr€om 2018; McCormack 2019).
Hence, the cocreative process undertaken between
artists and AI can lead to material, meaningful art-
worlds—a topic of considerable interest in creative
(and) digital geographies, as well as related fields.
In this article, we advance digital geographies in
two key ways. First, we nuance the theoretical discus-
sion of what we call coagency as a mutual relation
through use of Barad’s (2003, 2007, 2012, 2015) con-
cepts of intra-action, agential separability, topological
relationalities, and mattering. In doing so, we step
toward what we see as posthuman cocreative spatial-
ities (Lundman and Nordstr€om 2023). Second, we
enrich the empirical research by shifting focus away
from other collaborative digital technologies (e.g.,
screens, surveillance cameras, smartphones, and
advanced visualization and rendering technologies; cf.
Woodward et al. 2015; Rose 2017, 786) and toward
new AI-based digital techniques. As stated, AI is dif-
ferent from other technologies due to its autonomous
capacities. Despite this, there has been little empirical
research in digital geographies on the creative process
between humans and AI (except for Birtchnell and
Elliott 2018; Wingstr€om, Hautala, and Lundman
2022; Lundman and Nordstr€om 2023) or other digital
technologies (except for Woodward et al. 2015). As
such, we present novel empirical findings about the
cocreative processes of eighteen Finnish artists work-
ing with AI techniques, ranging from neural networks
to generative adversarial networks (GANs). Based on
our previous research (Wingstr€om, Hautala, and
Lundman 2022; Lundman and Nordstr€om 2023) and
the existing literature (e.g., Davis 2013; Kantosalo
and Toivonen 2016), our presumption is that humans
and AI are cocreative together.
In what follows, we first review the literature on
human–technology relations and agencies in digital
geographies, with a focus on technological produc-
tive capacities that can “make things happen” in the
world (Dodge and Kitchin 2005a, 162). For the pur-
poses of the review, we categorize three main
approaches toward human–technology relations in
the digital geography literature, namely (1) code as
mediating relation, (2) cyborg as merged relation,
and (3) coagency as mutual relation. Then, we
develop our theoretical framework by applying the
posthuman thinking of Barad to the cocreative pro-
cess between artists and AI, before going on to pre-
sent our methods and empirical material. Having
done this, we analyze the three phases of the cocrea-
tive process as described by the artists working with
AI, namely (1) entering the cocreative process with
a selection of data and code, (2) training and learn-
ing during cocreation, and (3) curating creative out-
comes. Finally, we summarize our main findings and
offer conclusions, discussing how the creative agen-
cies of artists and AI evolve together in the cocrea-
tive artmaking process through being connected to
various spaces and times when creating new material
and meaningful artworlds. By attending to the coc-
reative spatial process of artists and AI, we present a
processual way of approaching relations between
humans and technology in the era of AI—a topical
issue in contemporary digital geographies.
2 Nordstr€om, Lundman, and Hautala
Detecting Creativity in Human–
Technology Relations
Human–Technology Relations and Agencies in
Digital Geographies
We have identified three key ways of approaching
human–technology relations within the existing digi-
tal geography literature: (1) code as mediating rela-
tion, (2) cyborg as merged relation, and (3)
coagency as mutual relation. Although these typolo-
gies sometimes overlap and the categorization is not
exhaustive, they nevertheless provide insight into
how this tradition of digital geographies can be
nuanced in the era of AI.
The first approach, code as mediating relation
(M. Graham, Zook, and Boulton 2013; Leszczynski
2015; Pink and Fors 2017; Cugurullo 2020; Thulin,
Vilhelmson, and Schwanen 2020), regards humans
and automated code technologies as having separate
agencies that affect each other, with human life
influenced by “living-with technology” (Leszczynski
2015, 741–42). Digital geographies using this
approach reveal the ways in which automated code
technologies have reproduced (Thrift and French
2002), governed (Amin and Thrift 2002; S. Graham
2005; Dodge and Kitchin 2005a, 2005b; Leszczynski
2020), managed (S. Graham and Marvin 2001;
Dodge and Kitchin 2005a), and augmented (M.
Graham, Zook, and Boulton 2013) everyday life and
tasks. Such technologies can alter human experience
and affective response (Pink and Fors 2017;
Leszczynski 2019) in a variety of environments, from
gaming (Ash 2012, 2013) to urban spaces (Boulton
and Zook 2013; Bissell 2020). The robot provides a
prominent example of code-based agency that can
facilitate human life, and has been increasingly dis-
cussed in human geography (e.g., Del Casino 2016;
Bissell and Del Casino 2017; Lynch 2021).
The theoretical basis for many of these studies
stems from philosophies of science and technology,
including concepts such as “transduction”
(Mackenzie 2002; Simondon 2016), “technicity”
(Mackenzie 2002), and “technical mediation”
(Latour 1994). Automated code technologies pro-
duce “code/spaces” where software (code) and every-
day life are made through each another (Kitchin and
Dodge 2011). Here, the potentiality for creativity
manifests through transduction, which constantly
creates “new means of being and acting” in everyday
life (Dodge and Kitchin 2009, 1363), engendering
new experiences, senses, and intelligences in human
agencies and subjectivities (Thrift 2004, 596; see
also Leszczynski 2015; Pink and Sumartojo 2018;
Bissell 2021). Although human life is in many ways
governed or constrained by automatic code (S.
Graham 2005; Dodge and Kitchin 2005b), human
agency can also be enhanced by these technologies
through the possibilities they offer for movement,
experimentation, and coping with uncertainty (Ash
2012).
The agency of automated code is limited to mil-
lions of calculations proceeding in a specific order
(Thrift 2004, 584; cf. Thrift and French 2002).
Thus, even as such code creates calculative and
accurate actions, it might also contain errors and
instabilities, sometimes careening “out of control”
(Cockayne and Richardson 2017, 1651). Robots,
which are a combination of software (code) and
moving hardware, are a good example of this. Some
of their functions might apply AI, making them
partly autonomous agencies that can escape standard
human–robot relations—for example, when drones
or self-driving cars harm humans in accidents or
wars (e.g., Del Casino 2016; Bissell and Del Casino
2017; Shaw 2017; Bissell 2018). Human–robot rela-
tions are usually considered hierarchical, however,
and are formed in assemblages (Shaw 2017). As
such, robot and human agency remain separate in
the creative process. This is also the case for other
code-based agencies that emphasize technology’s cal-
culative, preprogrammed, and repetitive aspects.
Although it is possible for code-based technologies
to have creative agency—for example, when robots
change their environment (Del Casino 2016, 846)
or affect the creative agency of human beings
(Lynch 2021)—for the most part the calculative per-
spective on code-based agency implicitly undermines
the definition of creativity as something new and
surprising (Boden 2009). Further research is there-
fore needed in digital geographies to understand the
creative potentiality of new technologies, including
autonomously learning AI.
The second approach within digital geographies,
cyborg as merged relation, is influenced by
Haraway’s (1991) cyborg. Although this approach
does not involve pondering the separate agencies of
humans and technology in processes of hybridization
(Kitchin 1998; Schuurman 2002, 2004), some
acknowledge that different subject and object posi-
tions might be formed along the way (Wilson 2009;
Evolving Coagency between Artists and AI 3
Cockayne and Richardson 2017). With the cyborg,
geographers have aimed to break the dichotomy
between what is defined as human and what is
defined as technology, with particular attention paid
to how subject–object positions are altered when
human–nonhuman boundaries are blurred (Wilson
2009). Haraway’s (1991, 180) cyborg, especially, is
both an ontological figure of machine–organism
hybrid—that is, “the machine is us”—and epistemo-
logical figure for creating alternative knowledge.
When human and machine merge to form a hybrid
(Wilson 2009, 504), there is no separate human or
technology agency but only the agency of the
cyborg.
Geographers have applied the idea of a cyborg to
such issues as critical urbanism (e.g., Gandy 2005),
feminist geographic information systems (Schuurman
2002, 2004), and global online education (Sparke
2017). Cyborgs can see, write, and act in new ways,
giving them interesting potentiality for creative
agency. As a methodological tool in geography, the
cyborg is used for writing counterhistories about con-
ventional everyday situations (Henderson et al.
2014), building alternative ways of knowing
(Schuurman 2002, 2004; Wilson 2009), understand-
ing hybrid knowledge-formation processes (Sparke
2017), and “queering” the code/space (Cockayne
and Richardson 2017). Regarding agency, Cockayne
and Richardson (2017, 1652, 1655) claimed that the
potential for “social agency with technology”
emerges from instabilities in the code; that is, sites
where “bodies and machines perform code/space
together.” The cyborg partly liberates humans from
the autonomous self, thereby providing “a means of
becoming ‘posthuman’” (Gandy 2005, 32).
Nonetheless, in terms of researching cocreativity,
the cyborg as merged relation is insufficient when it
comes to understanding how various agencies emerge
and mutually evolve through the process, and there-
fore further investigations on creative relations and
agencies are required in digital geographies.
More recently, a third approach, coagency as
mutual relation, has emerged in digital geographies,
focusing on collaboration between digital technolo-
gies and humans (Woodward et al. 2015; Rose
2017). Epistemologically, mutual coagency exists in
the space between the first (separated) and second
(merged) categorizations of agency. Although pro-
ductive encounters regarding what humans can do or
experience through technology have previously been
endorsed in geography (Leszczynski 2015; see also
Verbeek 2005), this emerging approach emphasizes
human–technology cooperation as the starting point.
In geography, Woodward et al. (2015), drawing on
the philosophies of Simondon, explained how tech-
nical objects can be collaborators capable of trans-
forming human–computer collectives by presenting
unique problems. As they put it, when “technical
objects have the capacity to participate in relations
of invention” (507), new collective individuals are
created. Rose (2017), following Stiegler, put forward
the idea of how a human, when coconstituted with
technologies, becomes a posthuman agency.
Moreover, Pink and Fors (2018) argued that greater
attention should be paid to the coconstitution of
humans and digital technologies. Lynch and Del
Casino (2020), meanwhile, discussed how human
intelligence is augmented by AI, which, according to
them, is based on data “smashed … and parsed” by
algorithms formed partly by humans and partly by
machine-learning technology (609). In addition,
While, Marvin, and Kovacic (2021, 771) recognized
that robotics has become increasingly concerned
with “human robotic co-evolution,” which represents
more than merely replacing or replicating human
actions (Chen, Marvin, and While 2020).
We draw on these pioneering geographical discus-
sions in our research on how human and AI creative
agencies are tied together in cocreative artmaking
processes. Although recent studies in digital geogra-
phies have referred to the productive agency of digi-
tal technology (Dodge and Kitchin 2005b;
Woodward et al. 2015) and changing (post)human
embodiments and subjectivities (Wilson 2009; Pink
and Fors 2017; Rose 2017), they have not examined
in any great depth how the agency of digital tech-
nology changes in these relations, nor how this
affects processes that generate new spatialities and
subjectivities. In particular, we are interested in the
changes both human and AI agencies undergo dur-
ing the cocreative process, and how this evolving
mutuality can transform artworlds.
Human–Technology Relations in Related Fields
Outside geography, several recent studies offer
intriguing findings about relations that could be con-
sidered mutual coagencies between humans and
technology. The first field of interest is human–com-
puter interaction (HCI) research. Guzman and Lewis
4 Nordstr€om, Lundman, and Hautala
(2020) established a research agenda related to AI’s
role as communicator rather than mediator, arguing
that the ontological boundaries of what constitutes
human, machine, and communication are blurred,
and that communication involving AI extends
meaning-making beyond operations “through” the
machine to those “with” the machine (81; see also
Gemeinboeck and Saunders 2022). Draude (2020),
who discussed Barad and Haraway’s critical posthu-
manism in the context of HCI studies, argued that
HCI should be thought of “as a relational, net-
worked activity of matter and meaning inter-
meshing” (23). Both these studies are valuable in
analyzing how artworlds are made sense of in the
cocreative artmaking process between artists and AI.
In other fields, there is also a growing interest in
comprehending human–digital technology relations
as a process. In anthropology, Hasse (2022, 146)
noted this “human–machine relations as a process”
when studying children’s changing perceptions
about robots. We find such processual relation
underdeveloped, however, as it only acknowledges
changing human agency (which, as shown earlier,
had already been articulated by numerous geogra-
phers). More interesting is Lipp and Dickel’s
(2022) focus on how humans and machines are
both separated and interconnected through
“interfacing,” which for them is a “folded performa-
tivity of human/machine relations” (15), taking
place here and now. We advance this thinking
with our empirical research on cocreative practices
between artists and AI, as well as our considera-
tions of how the cocreative artmaking process can
be interpreted from a spatial perspective.
Elsewhere, A. Harris’s (2018) conceptualization of
“creative ecologies” provides a noteworthy research
opening, as it recognizes how humans and nonhu-
mans can more effectively generate and evolve when
facilitated by creative environments and networks.
Harris put forward “creative agency” as an emergent
assemblage or embodiment of intra-action and the
mutual constitution of entwined agencies. We fur-
ther develop this notion by studying the evolving
agencies of humans and AI, and how this coagency
affects artworlds. In other words, we are not only
interested in how collective agency arises from crea-
tive environments and networks, but also how agen-
cies mutually change in the artmaking process when
connected to other spaces and times (cf. Massey
2005; Nordstr€om 2018).
Developing Digital Geographies through
Research on the Human–AI Cocreativity
Process
The Entwinement of Humans and AI
We recognize the concept of coagency as mutual
relation as being suitable for empirical research on
human–AI cocreativity processes. Whereas Rose
(2017) discussed how (post)human agency is always
sociotechnical, Barad’s concept of intra-action (dis-
cussed further later) helps us understand just how
entwined humans and AI are. The agencies of
humans (in our case, artists) and AI emerge as partly
separated from each other, generating special capaci-
ties that act through what we call a spatial artmak-
ing process. In our study, agency formation is related
to the creative process: Artists (as humans) hold the
potentiality for multiple agencies, identities, and
roles, but when they enter the creative process they
perform as artists, and AI emerges as a creative AI.
We acknowledge, however, that human agency is
different from the agency of AI, which only mimics
human performance (Simonsen 2013). Humans can
feel and experience the newness, surprise, and value
of outcomes. AI, by contrast, can learn to generate
novel outputs, but cannot make sense of them.
Before proceeding, it is necessary to clarify with
reference to some geographical texts why humans
and AI do not exist as independent entities but are
fundamentally entwined. By this, we mean that
humans and AI are already bonded through the ways
in which AI is produced (by humans) and how this
affects everyday human life. Here, we find resonance
with Woodward et al. (2015), who touched on
Simondon’s interpretation of technical objects as
entities that contain human nature in their techni-
cal being. The same applies to AI techniques: AI is
a specific form of code-embedded technology aimed
at replicating human intelligence, and has a capacity
to learn, perceive, make sense, solve problems, and
make decisions (Bostrom 2017; cf. Cugurullo 2020;
Russell and Norvig 2020). AI is formed by algo-
rithms consisting of explicit instructions and models
of ordered action (Berry 2015), enabling the AI to
perform tasks autonomously based on human-set pre-
conditions (cf. Del Casino et al. 2020).
Kitchin and Dodge (2011) wrote about
“ontogenetic” algorithms, which are “always in the
state of becoming” and “teased into being” (Kitchin
Evolving Coagency between Artists and AI 5
2017, 18). Such algorithms are coded by humans as
part of sociotechnological assemblages and then
applied and modified for various aims at multiple
sites (Kitchin 2017; see also Dodge and Kitchin
2009). This idea of ontogenesis is also applicable to
how AI algorithms are created and distributed col-
lectively. Moreover, before AI can perform intelli-
gence, it must be fed with data produced by various
technologies (Janowicz et al. 2020; see also Wilson
2011); to learn autonomously, it must study input
data and training material (e.g., images, sounds,
texts, movements) selected by humans and often
produced for some other initial purpose.
Geographers have also been in interested in view-
ing AI as part of wider constellations that affect
humans in multiple ways, including exercising power
over bodies through governance (Kitchin 2017;
Cugurullo 2020) and surveillance (Del Casino 2016;
Chen, Marvin, and While 2020), restructuring the
urban context and planning (Batty 2018; Macrorie,
Marvin, and While 2021), exacerbating and improv-
ing uneven development (McDuie-Ra and Gulson
2020), and managing services through digital plat-
forms (Bissell 2020; Leszczynski 2020). In addition,
numerous businesses and public organizations have
taken an interest in AI-related methods, funding,
and customers (Lundman and Nordstr€om 2023). To
sum up, from a geographical perspective, humans
and AI cannot be viewed separately—rather, they
are intimately entwined through the production of
AI algorithms and training data, as well as through
AI’s influence on humans via a variety of situated
performances and societal contexts and processes.
The Cocreative Process between Humans and AI
in Artmaking
Our approach to human and AI cocreativity is
processual, meaning we define creativity in terms of
two key aspects. The first involves a context-specific
approach focused on the directional elements of the
creative process, where the aim is to create concrete
outcomes through creative practices (Lundman and
Nordstr€om 2023). The second concerns the continu-
ous changes taking place in the world, and encom-
passes the potentiality for “new relations between
the corporeal and material to emerge” (Williams
2016, 1561) and new ideas and ways of seeing to
develop (Woodward et al. 2015; Nordstr€om 2018).
These might be experienced by the artist as creative
moments or events. Such moments can happen
between humans and nonhumans, and as such, they
are linked to what has been referred to as posthu-
man creativity (Łapinska 2020; Chappell 2022; D.
Harris and Holman Jones 2022; Jagodzinski 2022) or
postanthropologic creativity (Roudavski and
McCormack 2016; McCormack 2019). Interpreted
in Barad’s (2003) terms, we approach cocreativity as
“neither a matter of strict determinism nor uncon-
strained freedom” (826). Accordingly, the cocreative
process in artmaking revolves around directional and
goal-oriented actions that arise from the creative
capacities developed through artist–AI relations
(Boden 2009), as well as uncontrollable experiments
in chance, play, and surprise (Lundman and
Nordstr€om 2023; cf. Williams 2016; Hawkins
2017a).
The cocreative process follows what can be char-
acterized as the general phases of artmaking. This
begins with theme preparation and idea generation
and selection, which might be subconscious activi-
ties, followed by the production and finally comple-
tion of the artwork (Mace and Ward 2002; St-Louis
and Vallerand 2015). Artistic creation does not nec-
essarily, however, proceed linearly from one stage to
the next—instead, the developing artwork might
“return to an earlier developmental phase,” or new
ideas might emerge during the process (Mace and
Ward 2002, 182). This formation of new ideas is
tied to changes in ways of seeing, with artists
experiencing new thoughts, insights, and perspec-
tives throughout the process (Lundman and
Nordstr€om 2023). AI affects this process of artmak-
ing in various ways, as it is capable of autonomous
learning and can generate new material by following
what it has learned (see also Mazzone and Elgammal
2019). In other words, AI changes the artistic crea-
tive process into a cocreative process between
humans and AI.
In traditional algorithmic art, an artist defines the
code’s rules to attain the desired aesthetic. In art
involving AI, the artist works with algorithms that
have the capacity to change, learn, and make deci-
sions based on previous phases of the process
(Mazzone and Elgammal 2019). In particular, neural
networks and deep learning—AI techniques based
on autonomous machine-learning techniques (Santos
et al. 2021; cf. Mitchell 1997)—carry the potential-
ity for AI to hold creative agency of its own.
Although AI has a creative agency similar or close
6 Nordstr€om, Lundman, and Hautala
to that of humans, its creativity is not solely the
product of computational models—rather, it is
affected by various humans in situated encounters
(Avdeeff 2019; Lynch 2022).
Generally, AI and humans affect each other
mutually in the cocreative process. In artmaking, AI
can inspire artists, provide critical knowledge, gener-
ate new ideas and material, and help artists progress
when creation seems obstructed (Kantosalo and
Toivonen 2016; Karimi et al. 2020; Lundman and
Nordstr€om 2023). At the same time, artists affect AI
by supplying it with data such as images, sounds, and
movements. Mazzone and Elgammal (2019) claimed
that the key artist–AI cocreative practices occur pre-
and postcuration. When precurating with visual AI,
for instance, an artist selects visual material, which
the AI starts imitating and changing, forming new
images. In postcuration, the artist selects an AI-pro-
duced image or set of images, developing it into a
final artwork. Between pre- and postcuration, the
artist “tweaks” the code to acquire interesting images
from the AI, which, in our work, relates to how new
artworlds emerge spatially in the intra-action
between artists and AI.
Cocreative Human–AI Agencies in Creating New
Artworlds
Through adopting a processual perspective in the
context of artmaking, we can approach human–AI
cocreativity as a spatial process that stretches
beyond any particular domain to new artworlds. In
speaking of artworlds, we refer to the idea that art
is not in space but is “of space,” and furthermore,
“space is of art” (Hawkins 2017b, xvii). In terms of
our study, this means that artmaking is affected by
worldly materialities and relations (Massey 2005;
Lysgård and Rye 2015) that are not entirely trace-
able, and conversely the created artworks have fur-
ther influence in the world. By focusing on
coagency, we can study how artists and creative AI
evolve mutually, and how this influences the spa-
tialities created—in this case, material and mean-
ingful artworlds. To conceptualize the coagency
from a spatial perspective (discussed but not explic-
itly theorized by Rose 2017 and Woodward et al.
2015), we now turn to Barad’s concepts of intra-
action, topological relationalities, agential separa-
bility, and mattering.
Although interaction between humans and digital
technologies has been widely discussed in digital
geographies (e.g., Woodward et al. 2015; Rose 2017;
Ash et al. 2018; Lynch and Del Casino 2020),
Barad’s concept of intra-action allows us to recognize
how different human–technology agencies emerge
and evolve during a cocreative process. In current
digital geographies, humans are seen as constantly
individuating in interactions with digital technolo-
gies referred to as objects (a combination of software,
hardware, and data) that can alter and, in some
cases, actively pose problems (Woodward et al.
2015; see also Rose 2016). The focus in collabora-
tive relations has thus far either been on humans
who are changing with technology (Rose 2017), or
the collective individualization of a team of humans
and changing digital objects (Woodward et al. 2015;
see also Rose 2016). Barad’s (2007) approach zooms
in on the fundamental entanglements and processes
in which material differences develop, creating a
variety of agencies (cf. Rose 2017) with diverse
capacities to act in the world-generating process.
Regarding digital technologies, our interest is in the
active agency of AI and its entwinements with
human agency.
In Barad’s (2012) intra-action, agencies are
“(be)coming together-apart.” Barad (2007) noted
that “since individually determinate entities do not
exist, measurements do not entail an interaction
between separate entities; rather, determinate enti-
ties emerge from their intra-action” (128). In terms
of our study, this translates into humans and AI
gaining their creative agencies as artists and creative
AI through one another in an ongoing, entwined
cocreative process. Although this entwinement of
changing agencies has yet to become a focus of digi-
tal geographies, for Barad every intra-action is thor-
oughly spatial. As such, she described them as
dynamic and topological relationalities, and (re)arti-
culations of agencies, materialities, and meanings of
the world (Barad 2003). Dynamic refers to change
and can be interpreted as the processuality of the
world (see also Williams 2016), whereas topological
refers to the processual connectedness of emerging
agencies and relations to various spaces and times.
Geographers are already familiar with discussions
about multiple space–times (Massey 2005) and topo-
logical spatiality (e.g., Allen 2011, 2016; Lysgård
and Rye 2017). Building on this, our focus on intra-
action draws on Rose’s (2017) interest in the
Evolving Coagency between Artists and AI 7
temporality and spatiality of practices and meanings
used in the reinvention of the posthuman digitally
mediated subject. Here, Barad’s idea of topological
relationalities is useful for understanding how the
emerging agencies of artists and AI are connected to
each other and the world (cf. Lysgård and Rye
2017).
Furthermore, Barad (2003) explained how we
come to understand changing processes, and how
such processes introduce something new to the
world: “We do not obtain knowledge by standing
outside of the world; we know because ‘we’ are of
the world. We are part of the world in its differential
becoming” (829). Thus, rather than occupying an
absolutely external position to the phenomenon
being investigated, the knower (in our case, an artist
giving meaning to an artwork) is in a position of
“exteriority within phenomena,” which Barad (2007,
89, 176–77) called agential separability. This concept
allows us to further identify the moments when,
through meaning-making, artists articulate their cre-
ative process with AI. Barad interpreted meaning-
making (Barad and Gandorfer 2021, 25–27) as
mattering, capturing how concepts are “material
configurations of the world” in the field of
“spacetimemattering.” Similarly, when it comes to
AI-based art, creative outputs interpreted as artworks
emerge from the topological relationalities of intra-
acting humans and nonhumans. Hence, creating
meaningful artworlds can be seen an act of matter-
ing, in which giving meaning is tied to the material-
ities of cocreative processes that stretch beyond any
single location to many spaces and times (Barad and
Gandorfer 2021, 24–25; Massey 2005). To sum up,
not only is the artist’s creativity mediated through
working with AI (cf. Hawkins 2021; Wingstr€om,
Hautala, and Lundman 2022; Lundman and
Nordstr€om 2023), but AI is imbued with its own
creative agency. Hence, artists and AI act as mutual
cocreative agencies that together create new material
and meaningful artworlds.
Material and Methods
The empirical material that forms the basis of our
study consists of recorded and transcribed video
interviews conducted with eighteen Finnish artists
who use AI in their work. We asked the artists to
demonstrate some of their AI-based artworks and, if
possible, the techniques and algorithms involved.
The artists were part of the group interviewed for
our preceding research (Lundman and Nordstr€om
2023), and their pseudonyms (A1–A27) are taken
from this earlier work. For reasons of anonymity, the
videos and transcriptions are not openly accessible,
as the artists could be recognized based on their art-
works and methods. The length of the videos varied
between six and fiftyminutes, in total constituting
six hours and fifteenminutes of material. All but
three of the interviews were conducted online due
to COVID-19 restrictions. Some video recordings
(A12, A13, A25) did not meet our research expecta-
tions, so they were omitted from the analysis.
The interviewees were mainly new media artists
with backgrounds in the visual arts, music, sound
art, or the performing arts. They had a variety of
coding skills and used a wide range of data, codes,
algorithms, and programs in their artworks. Most of
the artists used or had experimented with AI techni-
ques based on so-called neural networks, which
involve deep-learning methods and machine-learning
techniques (Santos et al. 2021). The artists were
given the freedom to present their artworks and
working techniques as they saw fit, meaning the
interviews varied in terms of their content and focus.
As such, we based our analysis on qualitative meth-
ods and illustrative examples of the artists’ narratives
regarding their cocreative processes with AI.
We chose to use videos as our research method
and recording technique to better capture the con-
versations in a multisensory manner (see Garrett
2011). This allowed the artists to mediate their
experience directly (cf. Garrett 2011) and explain
and demonstrate their creative process in a real set-
ting—in this case via a digital interface. As such, we
were able to closely follow how the artists presented
their work with AI, providing us with a deeper
understanding of their artistic practice (cf. Pink and
Sumartojo 2018). We asked the artists to present
their work and talk freely about the creative pro-
cesses that led to the final artwork. This was fol-
lowed by supplementary questions, such as what they
considered the most creative moment in the code or
technique they used. Using qualitative content anal-
ysis (Hsieh and Shannon 2005), we categorized the
interview material according to which phase of the
process was being referred to, after which we pro-
ceeded to form more descriptive categorizations of
the cocreativity between artists and AI. This led to
the following three interlinked categories: (1)
8 Nordstr€om, Lundman, and Hautala
entering the cocreative process with a selection of
data and code, (2) training and learning during coc-
reation, and (3) curating creative outcomes. Using
the concepts of intra-action, topological relational-
ities, agential separability, and mattering from
Barad’s posthuman theory, we analyzed how the
agencies of artists and creative AI emerged and
mutually evolved through these phases.
Analyzing the Cocreative Process
between Artists and AI
Entering into the Cocreative Process with Data and
Code
The cocreative process begins when the creative
agencies of artists and AI start evolving through
data and code. Initially, the agencies are hierarchical
and calculative. An artist codes the AI algorithms
(or selects or alters existing ones), then selects or
produces the input and training data for the AI, and
then initiates the learning process and sets temporal
limits on it. One artist described this phase as the
“months of basic training on the neural network,
that’s the way I cultivate, in the same way that an
artist cultivates his own skills on top of the old”
(A9). Here, we see how posthuman does not mean
entirely detaching from the human—rather, human
agency is reconfigured as part of different material
processes (Barad 2007, 27) with digital technologies
(Rose 2017). AI’s agency forms and evolves within
the limits and possibilities of the selected code and
data. Hence, the AI is not automated but an auton-
omous creative code technology (cf. Cugurullo
2020). Once the AI’s agency has been recognized,
mutual coagencies between humans and AI can
emerge (cf. Barad’s intra-action).
In this first phase of the cocreative process, the
artist’s agency manifests through the selection and
production of data. Many artists considered creating
personal “data sets,” a critical element of the process,
with, in computational terms, a large collection of
data needed to train AI. Here, the data set does not
have determinate boundaries, as the data applied by
the artists—including images (photographs, cartoons,
paintings, drawings, graphics), texts, sound files,
and videos—are entangled in various topological
relationalities. The sources were either the artists’
own productions or ready-made material collected
from newspapers, museum databases, social media
platforms, sounds and audio recordings, and Internet
open data sources. Some artists applied data that
had already been processed by the AI or combined
their own data sets with AI-processed data. A10, for
instance, used this method of accelerating the AI’s
learning process to “always start from somewhere.”
This practice creates distance from the original
source data, enabling the AI to widen its agency.
A17, for example, used sounds created by one AI—
which s/he calls “sequences of broken memories”—as
input data for another AI that creates music in such
a way as to remove any discernable trace of the orig-
inal track. In these examples, the agency of some
artists was directed toward “creative” outcomes,
whereas others wished primarily to play and experi-
ment with AI (see also Lundman and Nordstr€om
2023). Many artists envisioned producing particular
kinds of artworks, although they could not always
know the outcome at this stage of the process.
The artists’ agency related to coding was realized
within the assemblages produced by the coders and
AI. A27 described how the code is not “completely
your own” as “no one did it from start to finish.”
Similarly, it is impossible to trace the origin of the
code to the data set, as rather than being produced
in any one location, both code and data are assem-
bled in many digital and physical sites through
complex human–technology relations. The AI algo-
rithms and programs applied by the artists had often
been retrieved from open-source code development
platforms (e.g., GitHub) or companies developing
AI software (e.g., DeepMind, OpenAI, Nvidia,
Microsoft, IMB, Google). These AIs had already
been created for a particular purpose and trained
through specific data sets, meaning that neither the
artist nor the AI entered their relation as preexisting
entities—rather, their agencies emerged in different
situated entanglements (cf. Barad 2012). A17,
touching on the entanglement of the AI and artist,
stated, “A human … has coded … all the AIs. …
It is a combination of AI and human, they are not
separated.” A1, meanwhile, described the coding
process of neural networks as “we walk together, but
… that machine is made by me. I do it with my
materials.” Some artists built new codes and func-
tions for the AI, whereas a few had entirely gener-
ated their own neural networks. All the interviewed
artists had the skills to “tweak,” or as A19 expressed
it, “poke” the AI’s code. As such, the artists could
limit or enable the AI’s creative agency by changing
Evolving Coagency between Artists and AI 9
parameters that would “make small changes in the
program itself” (A2), altering the code (A9) or pur-
posefully coding mistakes (A22, A23). A14
described this playing with the parameters as
“pulling the lever.”
Although AI agency is mostly calculative and
subordinate to human agency at the beginning of
the cocreative process, it nevertheless holds consid-
erable creative potentiality. In fact, rather than AI
having a single agency, various AIs are equipped
with different agential potentialities. The AI techni-
ques applied by the artists were mostly based on neu-
ral networks, with GANs and neural style transfers
(NSTs) being two of the most common AIs applied
by the artists (see Forbes 2020). Although many of
the artists questioned the creative agency of AI
alone, some considered GANs to be an exception,
because they were seen to produce genuinely new
and surprising (and thus creative) outcomes through
their generative capacities (the technique we explain
in the next section). A9 explicitly stated that GANs
can be creative on their own, mimicking humans’
unplanned biological processes, and A2 gave an
example of a GAN starting with a random noise and
proceeding to create a completely new sound.
Training and Learning during the Cocreation
Process
Gradually, the cocreative agencies of the artists
and AI begin to evolve. We detected this in the vid-
eos not only when an artist trained the AI, but
when the AI started to train itself. Here, the agen-
cies of the artist and creative AI evolve together-
apart, with their capacities to act changing through
the cocreative process of making art (cf. Barad and
Gandorfer 2021, 15, 59). An AI’s training process
represents a shared action on the part of both the
artist who teaches the AI using code and data, and
the AI itself, which learns to perceive images and
sounds—a process we call “learning to see.” This
form of seeing involves multisensuous activity (cf.
Woodward et al. 2015; Nordstr€om 2018) that in the
case of AI is algorithmic in nature.
The training and learning of GANs, in particular,
is intriguing in terms of cocreativity. Here, the
evolving agency of AI arises from two competing
neural networks: the generator and the discriminator
(see Karras, Laine, and Aila 2020). Whereas the
generator learns to produce outcomes that are
increasingly “creative” in the eyes of the discrimina-
tor, the discriminator increasingly learns to recognize
“faults” in these outcomes (Karras, Laine, and Aila
2020; Santos et al. 2021). The aim of the generator
is to create original material that could have been in
the input images or sounds. The work of A1—in
which the AI (GAN) started learning by “watching
TV” to create faces—provides a good example. The
generator created images, and the discriminator tried
to recognize the “real images” among them. Such
training and learning are entwined with processes of
change and iteration. A9, who had studied how AI
“learns to see” animals and their body parts (e.g.,
hair, ears, eyes, nose), demonstrated how the learn-
ing and seeing occurs only gradually: “It is chaoti-
cally searching. … It is not correct and a little bit
in that direction. Then it starts slowly to form. …
It shapes, it starts to find those ears.” As A1
described, the AI’s iterative process of learning to
see is slow: “A neural network may take days before
it begins to understand anything because it has an
awful lot of layers that it needs to learn to perceive.”
This process is far from perfect: “It [AI] tries to
learn. It does not always succeed very well. … It
gets lost, interrupts, and tries something else” (A1).
This incomplete learning includes “mistakes” (A12,
A22). Moreover, AI can “get confused” (A27),
which makes it more human-like and connects it
with human agencies.
The artist’s agency evolves through observing the
AI and intervening in its internal processes. This
includes paying close attention to how an AI’s see-
ing changes, with the artist also “learning to see” dif-
ferently alongside the AI. A1 documented this
process carefully, including which data and codes
limited the AI’s actions, how the artist tweaked
these, and at what point s/he saw intriguing images:
“I taught it [AI] to scratch the images and it became
really exciting. … This image here. Where did it
come from? … There! Now it is there and then it
vanishes as the learning progresses!” A17, mean-
while, observed how AI learns to see with sounds:
“It makes the composition by itself, re-organizes the
sounds, loops … like this sounds really exciting, it
tries to get into the swing of it. There emerges this
characteristic … squeaking sound.” A1 found the
new material introduced by the AI in the middle of
the learning process intriguing: “In the beginning,
it’s very interesting, when it [AI] has one way to
perceive and then you teach it … and it starts
10 Nordstr€om, Lundman, and Hautala
learning. … The effect of both [human and AI] is
combined, and that’s where something interesting
can be found.” Understanding how the AI behaves
requires a considerable amount of experimental
work, as the artists need to understand both the
structure of the code and the visuality involved. To
do this, the artist must follow the code on the
machine, observing how the neural network changes
during the learning process.
Some interviewees experienced feelings of stress
when intra-acting with AI in its learning process,
with A27, for example, commenting that,
“Sometimes it is really frustrating when you try to
spin that code for many days, like ‘start working
now, start working now!’—and nothing happens.”
Sometimes the artists had to adjust their behavior
and actions to suit the AI. This was the case for A7,
who explained, “It’s been frustrating that the AI
does not obey our interpretation.” The imperfection
of working with AI also holds potential, though: “I
thought it was a mistake. But now, seven years later,
I realized that I can do art with this” (A10).
Although the artists’ agency evolved simply by pay-
ing attention to the AI’s learning process, they also
wanted to find and create interesting outcomes.
Despite this, the results were often unexpected: “I
give the prompt … ‘Tell me about x’ … And then
the text comes. … It’s so terribly unpredictable
what’s going to come from there” (A15).
In general, these examples highlight a profound
experience of creativity that would not be possible
without AI, which is why these creative moments
during AI training and learning are fundamentally
intra-active. Intra-action means both artist and AI
evolve as coagencies responsive to each other’s
actions: “The artist does something and then AI
catches it and takes input … and creates something
new from it or processes it in some direction … and
then the artist responds to it” (A2).
Curating the Outcome of the Cocreative Process
The final stage of the cocreative process involves
curating the creative outcome, for instance, paper
prints, digital proofs, videos, sound, shows, narratives,
or installations. Here, again, the issue of the artist’s
agency is evident: “I feel more like a curator … .
I don’t directly accept the answer that comes [from
the AI]” (A15). In this phase, artists select and make
sense of the creative outputs as meaningful works of
art. In doing so, new material artworlds emerge, with
creative outputs finalized and made available for fur-
ther exploration by audiences. For Barad, concepts do
not exist as free-floating ideas; rather, they are only
imbued with meaning through mattering (cf. Barad
and Gandorfer 2021, 24–25). Similarly, making sense
of a creative output as an artwork can only be done
through the dynamics of the cocreative process and
the mattering involved. Barad also argued that sense-
making is not an individual affair but iterative and
collaborative (Barad and Gandorfer 2021, 28), mean-
ing that when it comes to a work of art, the audience
and other actors (e.g., art critics) continue meaning-
making after it has been presented or performed for
the public.
In the curating stage, the artist’s agency evolves
through the materialities of working with AI and in
selecting the outcomes generated by the AI (cf.
Mazzone and Elgammal 2019). In practice, the artist
is the one who makes sense of the cocreative pro-
cess, finalizing the outcome as an artwork presented
to an audience. Although AI can generate novel
outputs, the artists in our study observed that it lacks
the capacity to judge outcomes (see also Avdeeff
2019). For instance, while A17 recognized that AI
could create good, original music—perhaps even
blurring the boundaries of what constitutes “an
artist”—s/he nonetheless considered the presence of
human agency crucial given the artist’s role in
directing the generation, selection, and curation pro-
cesses. A22 commented on the limitations of AI: “It
is capable of finding many different variations, but it
often lacks taste with which to select the best …
creative or good variation.” Moreover, only the artist
can imagine what it is like to be a human spectator
watching the final artwork (A27). A3 was interested
in finding creative symbols in AI outputs, which,
according to the artist, the audience could then
attach meaning to. Although meaning-making is a
strong factor in the artist’s agency, it is firmly con-
nected with the ebbs and flows of the AI: “I’m find-
ing more meaning in it all the time. It kind of takes
me deeper and deeper into a rabbit’s hole. I look for
meanings in things where there are no meanings,
because for AI, they don’t exist” (A27). There can
also be uncertainty and fluidity when selecting the
final outcomes: “It is kind of curating, curating the
visual material … I don’t know if these are art-
works, though” (A9). Hence, the artist does not
occupy an absolutely external position when it
Evolving Coagency between Artists and AI 11
comes to judging the cocreative process—rather, an
agential separability is generated within the artmak-
ing (cf. Barad 2003, 828).
When talking about the selection of outcomes,
some interviewees reflected on the entire cocreative
artmaking process, including how their experiences of
intra-action affected the articulation and expression
of the final artwork. A11, for example, was interested
in the visual input data’s influence on the final out-
put: “A collection of images with a certain kind of
light or a certain kind of feeling, content; then it is
possible to bring that out in implementation.”
Similarly, A17 paid attention to both the input and
output material when curating the sounds made with
AI. Many of those using GAN as a creative method
found the data produced by the generator during the
training of AI interesting, with the result that the
training process itself could affect curation. For
instance, both A1 and A9 described how fascinating
(but imperfect) images emerge when two different
data sets meet, with A9 recounting an intriguing
example discovered when exploring animals through
the eyes of AI: “I am not interested in when it looks
perfect, but when it looks interesting.”
In the curating stage, the cocreative process
between artists and AI is realized as experimental
and open to possibilities, laden with potential for
the creation of unforeseen artworlds (cf. Lundman
and Nordstr€om 2023). Although many artistic pro-
cesses can result in artworks that cannot be fully
envisaged beforehand, the element of surprise is
even more pronounced when working with AI. As
the interviews revealed, the artists could not foresee
what kind of artwork they would end up finalizing.
Conclusion
As a rapidly developing technology, it is crucial
to understand AI’s potential as a cocreative agency.
Through bringing digital and creative geographies
into contact with Barad’s posthuman digital think-
ing, we have spotlighted mutual coagency as a way
of researching human–AI relations and related spa-
tialities. In doing so, we have demonstrated that
mutual coagency and cocreative artmaking process
can be spatially explained. Here, Barad’s inherently
spatial posthuman theory enriches the digital geogra-
phies endeavor, guiding geographers toward better
understanding of how human and digital technolo-
gies are fundamentally entwined. Moreover, Barad’s
thinking helps us discern how different agencies
receive their capacities to act and evolve as part of
processes “of the world.”
Our empirical research demonstrates how the coc-
reative process between artists and AI begins with the
selection of materials (data and code) for artmaking
(or precuration, according to Mazzone and Elgammal
2019). Here, the agencies of artist and AI are hierar-
chical and calculative, and the directional element of
creativity is present. The artist’s agency primarily
emerges in the fact that it is they who launch the coc-
reative process and define the temporal limits govern-
ing the training of the AI. Whereas the AI only exists
because it has been coded in different sociotechnolog-
ical assemblages (Ash, Kitchin, and Leszczynski 2018,
36–37; see also McLean 2020), the artist creates,
applies, or tweaks the code. Moreover, the data intro-
duced into the cocreative process by the artist have
often been created in some other context for some
other purpose. Hence, rather than being confined to
the here and now (Lipp and Dickel 2022), the cocrea-
tive process is linked to various spaces and times (cf.
Massey 2005; Nordstr€om 2018). New artworlds
develop within the topological relationalities of the
selected data and code. Although less dominant in
this early phase, AI agency emerges in the potential
to create art alongside a human.
After data and code have been selected, the crea-
tive agencies of artists and AI begin to evolve mutu-
ally through processes of training and learning,
which involve creative moments and adjustments in
ways of seeing. Training and learning take place
through intra-action, with the capacities to act of
the artist and AI changing in relation to each other.
Here, AI agency evolves from potentiality to actual
creative agency, as AI is able to create novel outputs
through a process of “learning to see.” This ability to
modify its own actions sets AI apart from merely
being an artist’s tool, such as a paintbrush or ham-
mer (Mazzone and Elgammal 2019), temporarily
bestowing it with autonomous creative agency (cf.
Pink and Sumartojo 2018; Cugurullo 2020). The
creative agency of AI alone is, however, restricted,
as its capacity to differentiate is limited (de Vries
2020, 2113). In turn, the artist’s agency evolves
through close collaboration with AI, allowing them
to “learn to see” differently as well (cf. Woodward
et al. 2015; Nordstr€om 2018). Creative potential can
also be found in the mistakes made by AI when it
does not act as anticipated (cf. Cockayne and
12 Nordstr€om, Lundman, and Hautala
Richardson 2017), with—from the point of view
of artist agency—feelings of joy and frustration a
crucial part of the creative process (Simonsen 2013;
St-Louis and Vallerand 2015). This phase of the
cocreative process (i.e., AI creating novel outputs
and artists finding them intriguing) is also when the
agencies of humans and AI most clearly evolve
mutually as coagencies. Such intra-actions between
artist and AI hold the potential for new and surpris-
ing artworlds to emerge.
In curating the outcomes, it is the artist who
selects and articulates the final creative output (an
act that could also be called postcurating; cf.
Mazzone and Elgammal 2019). Here, our understand-
ing differs from the posthuman approach, where
humans and nonhumans hold equal weight in the
creative process, without constraints (A. Harris
2021). Rather, by emphasizing artist agency in this
phase, we touch on notions of agential separability
and mattering (Barad and Gandorfer 2021). The
artist’s agency emerges in giving meaning to the cre-
ative output as works of art, a process that happens
“together apart” with AI. This is not a return to
humanism (Simonsen 2013), as meaning-making is
closely tied to the materialities and mattering of the
cocreative process with AI (see also Draude 2020).
Nonetheless, meaning-making is also iterative and
social, continuing into when the artwork is pre-
sented to the audience. The artworlds created as
part of the cocreative process of artmaking become
not only meaningful and material, but unforeseen
and experimental.
Although Mazzone and Elgammal (2019) stated
that artists and AI are creative together in pre- and
postcurating, our research shows that the artmaking
process is much more interlinked, with the creative
agencies of both artists and AI evolving during coc-
reation. Moreover, drawing on Barad’s thinking, we
have developed understanding of the spatialities of
the cocreative process, revealing how the resultant
artworlds are linked to the different spaces and times
where data and code were created; how they are
formed in the intra-action between artists and AI;
and how they become meaningful and material in
the process of curating the final outcomes.
With our focus on AI as an evolving agency, we
have brought forward the geographical discussion
about the role of digital objects as active agents in
creative processes (Woodward et al. 2015). In doing
so, our research on actual cocreative practices
between artists and AI provides crucial insights into
how creative agencies mutually emerge and evolve
in artmaking. All in all, our contribution to digital
geographies spotlights the role of other-than-humans
in creating and understanding our world(s) in the
making. Even so, the topic is not straightforward, as
AI is not (yet) fully autonomous, and humans can
develop different capacities as agencies (Rose 2017).
Once digital technologies develop further, other
(non)human creative agencies could emerge and
new topological relationalities could be formed.
Digital geographies should follow this development
and be open to the novel human–technology rela-
tions and agencies yet to come.
Acknowledgments
We would like to thank the anonymous reviewers
for their valuable comments. We are also grateful for
insights and discussions with our colleagues Astrid
Huopalainen and Rosa Wingstr€om, as well as the
organizers and participants in the digital session
“Actually Existing Digital Geographies in the
Antipodes (and Elsewhere)” at IAGNZGS 2021. Our
special gratitude goes to the interviewed artists.
Funding
This work was supported by Kone Foundation
(grant number 201902784) and Academy of Finland
(grant number 319872).
References
Allen, J. 2011. Topological twists: Power’s shifting geogra-
phies. Dialogues in Human Geography 1 (3):283–98.
doi: 10.1177/2043820611421546.
Allen, J. 2016. Topologies of power. London and New
York: Routledge.
Amin, A., and N. J. Thrift. 2002. Cities: Reimagining the
urban. Cambridge: Polity Press.
Ash, J. 2012. Technology, technicity, and emerging prac-
tices of temporal sensitivity in videogames.
Environment and Planning A: Economy and Space 44
(1):187–203. doi: 10.1068/a44171.
Ash, J. 2013. Technologies of captivation: Videogames
and the attunement of affect. Body & Society 19
(1):27–51. doi: 10.1177/1357034X11411737.
Ash, J., B. Anderson, R. Gordon, and P. Langley. 2018.
Digital interface design and power. Environment and
Planning D: Society and Space 36 (6):1136–53. doi: 10.
1177/0263775818767426.
Evolving Coagency between Artists and AI 13
Ash, J., R. Kitchin, and A. Leszczynski. 2018. Digital turn,
digital geographies? Progress in Human Geography 42
(1):25–43. doi: 10.1177/0309132516664800.
Avdeeff, M. 2019. Artificial intelligence & popular music.
Arts 8 (4):130. doi: 10.3390/arts8040130.
Barad, K. 2003. Posthumanist performativity. Signs:
Journal of Women in Culture and Society 28 (3):801–
31. doi: 10.1086/345321.
Barad, K. 2007. Meeting the universe halfway. Durham,
NC: Duke University Press.
Barad, K. 2012. Interview with Karen Barad. In New
materialism: Interviews and cartographies, 48–70. Ann
Arbor, MI: Open Humanities Press.
Barad, K. 2015. On touching—The inhuman that therefore
I am (version 1.1). https://www.academia.edu/7375696/
On_Touching_The_Inhuman_That_Therefore_I_Am_
v1_1_
Barad, K., and D. Gandorfer. 2021. Political desirings:
Yearnings for mattering (,) differently. Theory &
Event 24 (1):14–66. doi: 10.1353/tae.2021.0002.
Batty, M. 2018. Artificial intelligence and smart cities.
Environment and Planning B: Urban Analytics and City
Science 45 (1):3–6. doi: 10.1177/2399808317751169.
Berry, D. 2015. The philosophy of software. 2nd ed.
Hampshire, UK: Palgrave MacMillan.
Birtchnell, T., and A. Elliott. 2018. Automating the black
art. Geoforum 96:77–86. doi: 10.1016/j.geoforum.
2018.08.005.
Bissell, D. 2018. Automation interrupted: How autono-
mous vehicle accidents transform the material politics
of automation. Political Geography 65:57–66. doi: 10.
1016/j.polgeo.2018.05.003.
Bissell, D. 2020. Affective platform urbanism: Changing
habits of digital on-demand consumption. Geoforum
115:102–10. doi: 10.1016/j.geoforum.2020.06.026.
Bissell, D. 2021. Encountering automation: Redefining
bodies through stories of technological change.
Environment and Planning D: Society and Space 39
(2):366–84. doi: 10.1177/0263775820963128.
Bissell, D., and V. Del Casino. 2017. Whither labor geog-
raphy and the rise of the robots? Social & Cultural
Geography 18 (3):435–42. doi: 10.1080/14649365.
2016.1273380.
Boden, M. A. 2009. Computer models of creativity. AI
Magazine 30 (3):23. doi: 10.1609/aimag.v30i3.2254.
Bostrom, N. 2017. Superintelligence. Oxford, UK: Oxford
University Press.
Boulton, A., and M. Zook. 2013. Landscape, locative
media, and the duplicity of code. In The Wiley-
Blackwell companion to cultural geography, ed. N. C.
Johnson, R. H. Schein, and J. Winders, 437–51.
Hoboken, NJ: Wiley.
Chappell, K. 2022. Researching posthumanizing creativity.
Qualitative Inquiry 28 (5):496–506. doi: 10.1177/
10778004211065802.
Chen, B., S. Marvin, and A. While. 2020. Containing
COVID-19 in China: AI and the robotic restructur-
ing of future cities. Dialogues in Human Geography 10
(2):238–41. doi: 10.1177/2043820620934267.
Cockayne, D., and L. Richardson. 2017. Queering code/-
space: The co-production of socio-sexual codes and
digital technologies. Gender, Place & Culture 24
(11):1642–58. doi: 10.1080/0966369X.2017.1339672.
Colton, S., and G. Wiggins. 2012. Computational creativ-
ity: The final frontier? In Proceedings of the 20th
European Conference on Artificial Intelligence (ECAI),
21–26. Amsterdam: IOS Press.
Cugurullo, F. 2020. Urban artificial intelligence: From
automation to autonomy in the smart city. Frontiers
in Sustainable Cities 2(38):1–14. doi: 10.3389/frsc.
2020.00038.
Davis, N. 2013. Human–computer co-creativity. In The
doctoral consortium at AIIDE, ed. G. Smith and A.
Smith, 9–12. Palo Alto, CA: AAAI Press.
Del Casino, V. 2016. Social geographies II: Robots.
Progress in Human Geography 40 (6):846–55. doi: 10.
1177/0309132515618807.
Del Casino, V. J., Jr., L. House-Peters, J. W. Crampton,
and H. Gerhardt. 2020. The social life of robots.
Antipode 52 (3):605–18. doi: 10.1111/anti.12616.
de Vries, K. 2020. You never fake alone. Creative AI in
action. Information, Communication & Society 23
(14):2110–27. doi: 10.1080/1369118X.2020.1754877.
Dodge, M., and R. Kitchin. 2005a. Code and the trans-
duction of space. Annals of the Association of American
Geographers 95 (1):162–80. doi: 10.1111/j.1467-8306.
2005.00454.x.
Dodge, M., and R. Kitchin. 2005b. Codes of life.
Environment and Planning D: Society and Space 23
(6):851–81. doi: 10.1068/d378t.
Dodge, M., and R. Kitchin. 2007. The automatic manage-
ment of drivers and driving spaces. Geoforum 38
(2):264–75. doi: 10.1016/j.geoforum.2006.08.004.
Dodge, M., and R. Kitchin. 2009. Software, objects, and
home space. Environment and Planning A: Economy
and Space 41 (6):1344–65. doi: 10.1068/a4138.
Draude, C. 2020. “Boundaries do not sit still”: From interac-
tion to agential intra-action in HCI. InHuman–computer
interaction, ed. M. Kurosu, Vol. 12181, 20–32. Cham:
Springer.
Feldman, S. 2017. Co-creation: Human and AI collabora-
tion in creative expression. Proceedings of EVA, UK,
422–29. doi: 10.14236/ewic/EVA2017.84.
Forbes, A. 2020. Creative AI: From expressive mimicry to
critical inquiry. Artnodes 26 (26):1–10. doi: 10.7238/
a.v0i26.3370.
Gandy, M. 2005. Cyborg urbanization. International
Journal of Urban and Regional Research 29 (1):26–49.
doi: 10.1111/j.1468-2427.2005.00568.x.
Garrett, B. 2011. Videographic geographies. Progress in
Human Geography 35 (4):521–41. doi: 10.1177/
0309132510388337.
Gemeinboeck, P., and R. Saunders. 2022. Moving beyond
the mirror: Relational and performative meaning mak-
ing in human–robot communication. AI & Society 37
(2):549–63. doi: 10.1007/s00146-021-01212-1.
Graham, M., M. Zook, and A. Boulton. 2013. Augmented
reality in urban places: Contested content and the
duplicity of code. Transactions of the Institute of British
Geographers 38 (3):464–79. doi: 10.1111/j.1475-5661.
2012.00539.x.
14 Nordstr€om, Lundman, and Hautala
Graham, S. D. N. 2005. Software-sorted geographies.
Progress in Human Geography 29 (5):562–80. doi: 10.
1191/0309132505ph568oa.
Graham, S., and S. Marvin. 2001. Splintering urbanism:
Networked infrastructures, technological mobilities and the
urban condition. London and New York: Routledge.
Guzman, A., and S. Lewis. 2020. Artificial intelligence
and communication. New Media & Society 22 (1):70–
86. doi: 10.1177/1461444819858691.
Haraway, D. 1991. Simians, cyborgs and women. New York:
Routledge.
Harris, A. 2018. Creative agency/creative ecologies. In
Creativity policy, partnerships and practice in education,
ed. K. Snepvangers, P. Thomson and A. Harris, 65–
87. London: Palgrave Macmillan.
Harris, A. 2021.Creative agency. London: Palgrave Macmillan.
Harris, D., and S. Holman Jones. 2022. A manifesto for
posthuman creativity studies. Qualitative Inquiry 28
(5):522–30. doi: 10.1177/10778004211066632.
Hasse, C. 2022. Humanism, posthumanism, and new human-
ism. In The Palgrave handbook of the anthropology of
technology, ed. M. Hojer Bruun, A. Walhberg, R. Douglas-
Jones, C. Hasse, K. Hoeyer, D. B. Kristensen, and B. R.
Winthereik, 145–64. London: Palgrave Macmillan.
Hawkins, H. 2017a. Creativity. London and New York:
Routledge.
Hawkins, H. 2017b. Foreword: Art works: Art worlds—
Scales, spaces, spaces. In Art and the city, ed. J. Luger
and J. Ren, xvi–ix. London and New York: Routledge.
Hawkins, H. 2021. Cultural geography I: Mediums.
Progress in Human Geography 45 (6):1709–20. doi: 10.
1177/03091325211000827.
Hayles, K. 2017. Unthought: The power of the cognitive non-
conscious. Chicago: University of Chicago Press.
Henderson, V., J. Davidson, K. Hemsworth, and S. Edwards.
2014. Hacking the master code: Cyborg stories and the
boundaries of autism. Social & Cultural Geography 15
(5):504–24. doi: 10.1080/14649365.2014.898781.
Hsieh, H., and S. Shannon. 2005. Three approaches to
qualitative content analysis. Qualitative Health Research
15 (9):1277–88. doi: 10.1177/1049732305276687.
Jagodzinski, J. 2022. Quantum creativity. Qualitative Inquiry
28 (5):586–95. doi: 10.1177/10778004211066633.
Janowicz, K., S. Gao, G. McKenzie, Y. Hu, and B.
Bhaduri. 2020. GeoAI: Spatially explicit artificial
intelligence techniques for geographic knowledge dis-
covery and beyond. International Journal of
Geographical Information Science 34 (4):625–36. doi:
10.1080/13658816.2019.1684500.
Kantosalo, A., and H. Toivonen. 2016. Modes for creative
human–computer collaboration. In Proceedings of the
seventh international conference on computational creativ-
ity, ed. F. Pachet, A. Cardoso, V. Corruble, and
F. Ghedini, 77–84. Paris: Sony CSL.
Karimi, P., J. Rezwana, S. Siddiqui, M. Maher, and N.
Dehbozorgi. 2020. Creative sketching partner: An
analysis of human–AI co-creativity. In Proceedings of
the 25th international conference on intelligent user inter-
faces, ed. F. Paterno and N. Oliver, 221–30. New
York: Association for Computing Machinery.
Karras, T., S. Laine, and T. Aila. 2020. A style-based gen-
erator architecture for generative adversarial net-
works. IEEE Transactions on Pattern Analysis and
Machine Intelligence 43:4217–28.
Kitchin, R. 1998. Towards geographies of cyberspace.
Progress in Human Geography 223 (3):385–406. doi:
10.1191/030913298668331585.
Kitchin, R. 2017. Thinking critically about and research-
ing algorithms. Information, Communication & Society
20 (1):14–29. doi: 10.1080/1369118X.2016.1154087.
Kitchin, R., and M. Dodge. 2011. Code/space: Software
and everyday life. Cambridge, MA: MIT Press.
Łapinska, J. 2020. Creativity of human and non-human
matter interwoven: Autonomous sensory meridian
response videos in a posthuman perspective. Creativity
Studies 13 (2):336–50. doi: 10.3846/cs.2020.11703.
Latour, B. 1994. On technical mediation. Common
Knowledge 3:29–64.
Leszczynski, A. 2015. Spatial media/tion. Progress in
Human Geography 39 (6):729–51. doi: 10.1177/
0309132514558443.
Leszczynski, A. 2019. Platform affects of geolocation.Geoforum
107:207–15. doi: 10.1016/j.geoforum.2019.05.011.
Leszczynski, A. 2020. Glitchy vignettes of platform urban-
ism. Environment and Planning D: Society and Space 38
(2):189–208. doi: 10.1177/0263775819878721.
Lipp, B., and S. Dickel. 2022. Interfacing the human/machine.
Distinktion: Journal of Social Theory. Advance online pub-
lication. doi: 10.1080/1600910X.2021.2012709.
Lundman, R., and P. Nordstr€om. 2023. Creative geogra-
phies in the age of AI: Co-creative spatiality and the
emerging techno-material relations between artists
and artificial intelligence. Transactions of the Institute
of British Geographers. doi: 10.1111/tran.12608.
Lynch, C. 2021. Critical geographies of social robotics.
Digital Geography and Society 2:100010. doi: 10.1016/j.
diggeo.2021.100010.
Lynch, C. R. 2022. Glitch epistemology and the question
of (artificial) intelligence. Dialogues in Human Geography
12 (3):379–83. doi: 10.1177/20438206221102952.
Lynch, C., and V. Del Casino. 2020. Smart spaces, infor-
mation processing, and the question of intelligence.
Annals of the American Association of Geographers 110
(2):382–90. doi: 10.1080/24694452.2019.1617103.
Lysgård, H. K., and S. Rye. 2017. Between striated and smooth
space: Exploring the topology of transnational student
mobility. Environment and Planning A: Economy and Space
49 (9):2116–34. doi: 10.1177/0308518X17711945.
Mace, M., and T. Ward. 2002. Modeling the creative pro-
cess. Creativity Research Journal 14 (2):179–92. doi:
10.1207/S15326934CRJ1402_5.
Mackenzie, A. 2002. Transductions: Bodies and machines at
speed. London: Continuum Press.
Macrorie, R., S. Marvin, and A. While. 2021. Robotics
and automation in the city. Urban Geography 42
(2):197–217. doi: 10.1080/02723638.2019.1698868.
Massey, D. 2005. For space. London: Sage.
Mazzone, M., and A. Elgammal. 2019. Art, creativity, and
the potential of artificial intelligence. Arts 8 (1):26.
doi: 10.3390/arts8010026.
Evolving Coagency between Artists and AI 15
McCormack, J. 2019. Creative systems. In Computational
creativity, ed. T. Veale and F. Cardoso, 327–52. Cham:
Springer.
McDuie-Ra, D., and K. Gulson. 2020. The backroads of AI:
The uneven geographies of artificial intelligence and devel-
opment.Area 52 (3):626–33. doi: 10.1111/area.12602.
McLean, J. 2020. Changing digital geographies. Cham,
Switzerland: Palgrave Macmillan.
Miller, A. I. 2019. The artist in the machine: The world of
AI-powered creativity. Cambridge, MA: MIT Press.
Mitchell, T. M. 1997. Machine learning. New York:
McGraw-Hill.
Nordstr€om, P. 2018. Creative landscapes: Events at the
sites of encounter. Doctoral diss., University of Turku.
Pink, S., and V. Fors. 2017. Being in a mediated world.
Cultural Geographies 24 (3):375–88. doi: 10.1177/
1474474016684127.
Pink, S., and S. Sumartojo. 2018. The lit world: Living with
everyday urban automation. Social and Cultural Geography
19 (7):833–52. doi: 10.1080/14649365.2017.1312698.
Rose, G. 2016. Rethinking the geographies of cultural
“objects” through digital technologies. Progress in
Human Geography 40 (3):334–51. doi: 10.1177/
0309132515580493.
Rose, G. 2017. Posthuman agency in the digitally mediated
city. Annals of the American Association of Geographers
107 (4):779–93. doi: 10.1080/24694452.2016.1270195.
Roudavski, S., and J. McCormack. 2016. Post-anthropo-
centric creativity. Digital Creativity 27 (1):3–6. doi:
10.1080/14626268.2016.1151442.
Russell, S. J., and P. Norvig. 2020.Artificial intelligence: A mod-
ern approach. 4th ed. Harlow, UK: Pearson Education.
Santos, I., L. Castro, N. Rodriguez-Fernandez, A.
Torrente-Pati~no, and A. Carballal. 2021. Artificial
neural networks and deep learning in the visual arts:
A review. Neural Computing and Applications 33
(1):121–57. doi: 10.1007/s00521-020-05565-4.
Schuurman, N. 2002. Women and technology in geogra-
phy: A cyborg manifesto for GIS. The Canadian
Geographer/Le Geographe Canadien 46 (3):258–65. doi:
10.1111/j.1541-0064.2002.tb00748.x.
Schuurman, N. 2004. Databases and bodies: A cyborg
update. Environment and Planning A: Economy and
Space 36 (8):1337–40. doi: 10.1068/a3608_b.
Shaw, I. 2017. Robot wars: US empire and geopolitics in
the robotic age. Security Dialogue 48 (5):451–70. doi:
10.1177/0967010617713157.
Simondon, G. 2016. On the mode of existence of technical
objects. Vol. 39. Minneapolis: Univocal Publishing.
Simonsen, K. 2013. In quest of a new humanism. Progress
in Human Geography 37 (1):10–26. doi: 10.1177/
0309132512467573.
Sparke, M. 2017. Situated cyborg knowledge in not so
borderless online global education. Geopolitics 22
(1):51–72. doi: 10.1080/14650045.2016.1204601.
Stiegler, B. 1998. Technics and time I. Palo Alto, CA:
Stanford University Press.
St-Louis, A., and R. Vallerand. 2015. A successful crea-
tive process. Creativity Research Journal 27 (2):175–87.
doi: 10.1080/10400419.2015.1030314.
Thrift, N. 2004. Movement-space: The changing domain
of thinking resulting from the development of new
kinds of spatial awareness. Economy and Society 33
(4):582–604. doi: 10.1080/0308514042000285305.
Thrift, N., and S. French. 2002. The automatic production
of space. Transactions of the Institute of British Geographers
27 (3):309–35. doi: 10.1111/1475-5661.00057.
Thulin, E., B. Vilhelmson, and T. Schwanen. 2020.
Absent friends? Smartphones, mediated presence, and
the recoupling of online social contact in everyday
life. Annals of the American Association of Geographers
110 (1):166–83. doi: 10.1080/24694452.2019.1629868.
Verbeek, P. P. 2005. Artifacts and attachment: A post-
script philosophy of mediation. In Inside the politics of
technology: Agency and normativity in the co-production
of technology and society, ed. H. Harbers, 126–46.
Amsterdam: Amsterdam University Press.
While, A., S. Marvin, and M. Kovacic. 2021. Urban
robotic experimentation. Urban Studies 58 (4):769–
86. doi: 10.1177/0042098020917790.
Williams, N. 2016. Creative processes: From interventions
in art to intervallic experiments through Bergson.
Environment and Planning A: Economy and Space 48
(8):1549–64. doi: 10.1177/0308518X16642769.
Wilson, M. 2009. Cyborg geographies: Towards hybrid
epistemologies. Gender, Place and Culture 16 (5):499–
516. doi: 10.1080/09663690903148390.
Wilson, M. W. 2011. Data matter(s): Legitimacy, coding,
and qualifications-of-life. Environment and Planning D:
Society and Space 29 (5):857–72. doi: 10.1068/d7910.
Wingstr€om, R., J. Hautala, and R. Lundman. 2022.
Redefining creativity in the era of AI? Creativity
Research Journal. Advance online publication. doi: 10.
1080/10400419.2022.2107850.
Woodward, K., J. P. Jones, L. Vigdor, S. Marston, H.
Hawkins, and D. Dixon. 2015. One sinister hurricane:
Simondon and collaborative visualization. Annals of
the Association of American Geographers 105 (3):496–
511. doi: 10.1080/00045608.2015.1018788.
PAULINA NORDSTR€OM is an Associate Professor in
the Department of Global Development and Planning,
University of Agder, PB 422, 4604 Kristiansand,
Norway. E-mail: paulina.e.nordstrom@uia.no. Her
research interests include spatialities of creativity, crea-
tive processes, and art in various contexts.
RIINA LUNDMAN is a Postdoctoral Researcher in
the Unit of Social Research, Tampere University,
Tampere 33100, Finland. E-mail: riina.lundman@tuni.fi.
Her research interests are various, including art, play,
and creative geographies.
JOHANNA HAUTALA is an Associate Professor
(Regional Studies), University of Vaasa, PB 700, 65101
Vaasa, Finland. E-mail: johanna.hautala@uwasa.fi. Her
research interests focus on spatiotemporal creativity and
knowledge creation processes in the contexts of
advanced technologies.
16 Nordstr€om, Lundman, and Hautala