Keywords

1 Introduction

This commentary aims to bring together perspectives on narrative-centered learning and through them to raise questions about how the narrative changes in the area of Artificial Intelligence (AI) when AI is used for learning purposes. The text is a constellation of modalities, as it is based on three interrelated contextual frameworks. One of them includes instances from the keynote speech of Professor James Lester delivered at the AI in Learning conference that took place online in November 2021 (Lester 2021). The second is an interview where Professor Lester further responds to questions posed by Professor Hannele Niemi and Postdoctoral researcher Jenny Niu (the interviewers from now on in this text). The third is this commentary on selected pieces of the keynote and the interview aiming to synthesize these with a focus on the narrative element that underlies the use of AI in Learning.

The keynote was originally in video format and the interview in a similar configuration. Later, the audiovisual texts were transcribed to ease access and the ability to refer to the details of the interactions.

Considering these multiple forms of textuality, it is evident that the two main sources of this chapter constitute expressions of the agencies of the participants of the communicative events of the keynote and the interview.

1.1 The Key Message of the Keynote and Interview

In the keynote speech and interview, Lester’s (2021) overall goal is to discuss ways that AI technologies support education and learning. More particularly, the focus is on AI which, being a megatrend in our era, generates diverse public discourse. As Lester describes it, AI is often thought as a kind of “mysterious force.” This metaphorical linguistic expression does not only present an interesting perspective on AI as a technological entity that is the carrier of a force surrounded by the mystery of the not-yet fully understood. Being a force means that AI has an impact on learning and, as such, is carrier of a certain kind of agency.

The aim of Lester’s (2021) speech is to bring forward key issues in AI-enhanced learning and how it can be promoted through narratives. Lester’s reflections are illustrated with Crystal Island, a game that offers to learners opportunities to develop understanding through storytelling and problem-solving.

The focus of the keynote is on narrative-centered learning that entails the pedagogical use of stories and storytelling for deep learner engagement.

This commentary will focus on the concept of narrative-centered learning and related ones, such as tutorial dialogue and characters.

1.2 Key Concepts and Metaphors

The multimodal texts, therefore, that inform this commentary bring forward powerful contemporary metaphors that refer to AI-enhanced learning in the keynote and the interview.

One such metaphor is narrative-centered learning that runs through the keynote talk and signifies how the agencies of researchers, teachers, and students intertwine with technologies in physical and online environments in the passage of time to construct the metaphors of AI-enhanced learning in the future. The discussion is illustrated with Crystal Island, a game-based learning environment that aims to engage students in story-based activities with believable characters and problem-solving features at the core of the storytelling.

In addition to narrative-centered learning, other key metaphors in the keynote are conveying already established concepts, such as tutorial dialogue and characters. Some others, such as the drama manager, are new and signify ways of supporting the process of learning with AI technologies.

This commentary then aims to bring forward the metaphors associated with the notion of narrative in learning and how they relate with the development of the Crystal Island as a game to support students’ constructing knowledge in science education. To this end, the commentary draws from Paul Ricoeur’s (1978, 1986, 1992) narrative and metaphor theory and introduces aspects from the work of new materialist and post-humanist thinkers (e.g., Barad 2007; Coleman 2020; Stark 2016; Truman 2019) with a focus on the role of technology. In both the narrative and the new materialist/ post-humanist theoretical standpoints, agency is a critical notion. Based on these, the commentary draws from relevant concepts and metaphors in the keynote aiming to take further the narrative of agency in AI-enhanced learning.

According to Coleman (2020), agency becomes visible and understood through temporal, spatial, and material modalities. Modalities signal the ways agency is organized, distributed, and displayed. It is only natural then that in multimodal texts about AI-based environments (like the keynote and the interview are) the multiple modalities convey agency through a multiplicity of metaphors of learning.

The metaphor of agency, although not explicitly stated in the keynote and the interview, is an all-encompassing one. As multiple modalities converge in the audiovisual display, the scholarship, the background, and interests of the keynote and interview participants are revealed. The video of the keynote, for example, aims to communicate the speaker’s message to researchers, scientists, teachers, and other audiences with whom an interest in the impact of AI on education in the future is shared. As Lester puts it in his keynote,

One is, … I think [we will be] seeing fascinating developments in the upcoming five years or so in AI technologies to support education, which is really the focus of this talk. But it is also the case we are going to see some really interesting developments in ‘AI education’ per se, that is, AI as the subject matter for K-12 education.

The agency of the speaker in the study and research of AI, although outspoken in previous and later parts of the keynote text, here is resting between the lines. Nevertheless, it is underlying the considerations and imaginaries that the keynote expresses. The material dimension of AI will strongly impact the way education takes place in the actual, physical environment of the classroom and school in the future. The narrative of education will, therefore, change in the days to come with the use of AI. It is the directions of the change of narrative that the keynote aims to capture. Similarly, the interview questions target the visualizations of future changes and depart to bring into light their finer shades.

1.3 Modalities, Narrative, and Metaphors in AI for Learning Purposes

The role of the narrative, therefore, is double here. First, there is the narrative that the multimodal texts construct concerning the future of AI in schools. Second, there is the narrative-based approach that is integrated in applications of AI for pedagogical purposes. Indeed, as the work of the philosopher of language Paul Ricoeur has shown, there are diverse forms and modes of narrative. According to Ricoeur (1986, 1992), despite the diversity, all narratives present universal elements. They perform a common function, namely, they mark, organize, and clarify temporal experience (Ricouer 1986).

The temporal experience that is organized and clarified through the conventions of the narrative does not concern the storytellers themselves whose agential knowledge, values, and practices are transferred through the storyline. It mainly concerns the lived experiences of the characters whose actions, events, and relations the stories are telling. The narrative of AI for learning purposes, therefore, emerges out of the agencies of its authors and tells the story of the agencies of its characters.

The plot of the narrative makes it possible to synthesize the experiences of the characters by organizing the story through, for example, expressions of time, descriptions of settings and backgrounds, and so on. In this way, through narrative plot, the meaning of persons, relations, and events that make up life affairs become visible. In this sense, the plot and the characters develop in a dialectical way. The development of the plot cannot happen without the actions, thoughts, decisions etc. of the characters. Neither can the characters grow outside the temporal and spatial configurations of the plot (Ricoeur 1986, 1992).

In this commentary, the plot of the narrative aims to make visible how students and teachers in K-12 education use AI for learning and what meanings emerge out of this use.

To make the multifunctional performance of the narrative possible, speakers and writers use metaphors. Metaphors can be of different types and so are metaphors of learning, multiple and shifting. How metaphors shift, for example, how novel or conventional they become, depends on the era and its socioeconomic and political developments.

As Lester explains,

There are many types, there are many metaphors of learning. I think it is fair to say that for the history of our field one of the most significant and powerful metaphors that we had since the beginning, since the 1970s, 50 years now, is tutorial dialogue. It is a very exciting area, it is an area that our group has worked in, and I know many people in conference, your labs are working on this too. I have seen your program, which looks fantastic. It is such a great metaphor. It is a really interesting development over, roughly the last 2,000 years that we have come to understand that human tutoring, where human tutoring engages with dialogue, in dialogue with the human student is incredibly effective.

It is arguably one of the most effective, if not the most effective approaches that we have. It is curious, we don’t know exactly why this is. Right? It could be for the self-explanation effect. It could be because of very powerful learning mechanisms that are kind of released you might say when students engage in human dialogue with the tutor. There could be a very strong effect on components, for young learners. And likely it is a result of all of these and even more. This is one metaphor out of many, many possible metaphors and I would like to suggest one I think is particularly interesting and one we will be focusing on this morning’s remarks which is known as narrative-centered learning.

Evidently, in the section above, Lester acknowledges the diversity of metaphors that relate to learning and the development of metaphorical language and its meanings as time progresses and technology advances. This reflects Ricoeur’s (1986, 1978) consideration that is built from the claim that metaphor should be grasped not as the substitution of one conventional name for a different one. When it comes to dialogue, tutoring and learning, we should go beyond the conventional meaning of the words by setting the ground for new imaginaries of tutorial dialogue. In this way, the metaphor of tutorial dialogue can evolve and transform into a metaphor of narrative-based learning. The question then arises: Is narrative-centered learning an innovation?

As Lester further elaborates,

Narrative-centred learning is in some ways not a new metaphor at all. The sort of recognition of the importance of story for human learning that sort of episodic memory that it triggers. The deep engagement that often contracept when students engage in it is a sort of hallmark of narrative-centered learning. But what I would like to suggest is that, in fact, because of the very recent developments in AI it is not going to be possible to really create an incredible powerful narrative-centered learning environment.

Narrative-centered learning links, therefore, with different theories of learning that have evolved in time and can have an impact on students’ memory, engagement, and so on.

As Lester continues,

So really two parts of this discussion this morning. First is kind of looking at what you might call narrative-centered learning environments look like today. We look at one, and this is sort of an exemplar. It is kind of like a little case study. And what I would like for you to do when you look at this, think about how this kind of narrative-centered learning environment could in fact be kind of the laboratory for studying narrative-centered learning with ‘AI full on’, with fully supporting learning interactions.

It becomes evident, then, that while narrative-centered learning is not new (i.e., this is a conventional metaphor), the integration of AI system into the narrative approach can be innovative for learning and pedagogical purposes.

2 Crystal Island as a Metaphor for Learning with AI

To illustrate the innovative dimensions of narrative-centered learning with AI, Lester uses the example of Crystal Island in the following section:

Chrystal Island is a narrative-centered learning environment that has been under development by our group over many, many versions over many years. You can think of narrative-centered learning environments as a kind of an intelligent game-based learning environment. … [T]here is a great attraction to having students to participate these story-centered activities that are fundamentally featuring problem solving in a way that fully integrates the story with the problem solving. And the students, …, emerge themselves in these narratives. The narratives can be more or less powerful. They can be more or less well-designed; they can more or less effectively integrate pedagogical [purposes] into the learning experiences.

As it happens with well-organized narratives, there are specific elements that characterize well-designed interactive narratives for learning.

As Lester goes on to argue,

One is believable characters. So of course, enormous amount of work for many years and non-player characters (NPCs), and the words that are very expressive and captivating and then finally rich stories that unfold over time. So, these are core characteristics of narrative-centered learning environments which tend to have certain kinds of effects. One is that unlike many kinds of learning there is actually a very strong elicitation of learner affect in narrative-centered learning environments, and affect has a very strong impact on performance. It can be a positive impact. It can also be negative. Supporting effect is very important as we know in kind of more traditional tutoring, and it is really kind of core characteristic in many forms of learning that can contribute into effective learning. It is kind of particularly amplified in narrative-centered learning.

Indeed, in the Ricoeurian narrative theory, the character is not only an essential element. Most importantly, the character is in dialectical relation with the plot (Ricoeur 1992). This means that as the plot of the narrative evolves, the characters evolve as well. In addition, the events, actions, emotions, and relations that the characters are entangled with move the plot of the story forward. Therefore, beyond the expressiveness of words, the agency of the characters makes them rich and captivating, as stories unfold over time. The characters’ agency is interconnected with whom those characters are. In the Crystal Island game-based learning environment, they represent different genders, racial and ethnic backgrounds. This attributes an innovative element to the game since the referential function of the Island narrative contributes to new imaginaries of AI-based design of games for learning purposes. This means that believable characters reshape the reality of games and display the world as multicultural and diverse.

And yet, Crystal Island represents only a small portion and, no matter how deeply we would wish for it, the world is not an island. How does then the Crystal Island metaphor speak to the rest of the world?

Under this lens, the interviewers ask,

Interviewers: So, … thinking then for the future, now [that] you have this knowledge from creating this wonderful environment... [B]ut... do you think that people in different countries could do something similar, based on what you have done during [these] fifteen years? Or should they do everything just from the beginning?

In response, Lester explains,

There are so many developments in the last, let’s say, five, years or so, that I think is going to make it much, much, much easier to create these environments for everyone. One of the developments is that often at the sort of foundational infrastructure level, there are game technologies and there’s—Finland of course is famous for this—such an enormous investment in the underlying technologies, for game engines that “for free” we researchers are able to leverage all of the 3D worlds, the characters, the game playing mechanics, all kinds of computational capabilities that these game engines offer and that’s our starting point. Rather than starting from nothing, we can start from that, which is very helpful. Then there’s a sort of collection of know-how or maybe best practices that have begun to evolve. So, we start seeing the literature, but we also start seeing in discussions and conferences. Shared interest makes it possible to not only do it kind of more efficiently, because of shared knowledge, but also more effectively. And the third and final thing I mention, which is in my own view the most exciting, is that over the next —let’s say five years, seven years something in this time frame— we’re going to be seeing the emergence of AI technologies that underlie all of it. That will make it amazingly, if not easy, a lot easier to actually create these kinds of game-based learning environments. And that’s the thing, we don’t exactly know how that’s going to happen, but it’s very exciting.

This explanation signifies the need for a transdisciplinary approach to narrative-centered game design for learning. The consideration and integration of theories and practices from the literature is where perspectives from various scientific discourses, including computer science, human-computer interaction, and science education, intertwine. However, Lester’s response in the section above brings forward mainly technological metaphors. These make visible the significance of the role of technology as game changer in the educational discourse of the future. Although the narrative-based game design should consider contextual, social, cultural, economic, historical, and other factors, the technology itself interacts with all of those. Technology, therefore, has an impact on the ways agency is organized, distributed, and displayed in space, time, and materiality.

In this sense, as many new materialist and post-humanist thinkers (e.g., Barad 2007; Truman 2019) would possibly agree, technology itself has agency. As Lester explains, different infrastructure will be needed to serve the needs of Finnish students if narrative game-based learning migrates to, for example, Finland. This would possibly include algorithmic configurations and design that consider the sociocultural dimensions of the learning context.

This speaks to the fact that techno-material (more-than-human or nonhuman) entities interact with the agency of humans. In this sense, techno-materialities bear their own agential qualities.

Moreover, this means that the integrated narrative has an impact on the ways the whole narrative plays and pushes the wider discourse of education and technology-enhanced learning forward.

The role of the characters also shifts, and new agents come into play in order to make possible the integration of Crystal Island in a context other than the one of its origins. For this kind of migration, a labor-intense process takes care the needs of the students on an individual basis and the new role of the drama manager is introduced. The drama manager plays a critical role here. As Lester goes on to explain,

So, in this approach we first create kind of a base line learning environment. It can be like Crystal Island. And then students one by one, typically in a laboratory setting in this approach, will interact with the game. So, they will solve a science mystery, they will talk to the characters, they will fill out diagnosis work sheets, if it’s about sort of diagnostic task. So, sort of that kind of thing. But, unbeknownst to them, so they don’t know this, but sitting often in another room is a kind of ‘expert drama manager.’ So, this is a person who is actually controlling when the character does this, or when a particular event in the world does that, so you can sort of imagine little switches been flipped so that the drama manager is actually the one creating a very personalized interactive narrative for the student. So, when you did that for many students, it’s of course incredibly labor intense because you’re doing it one by one and it’s kind of interesting process.

3 Reversing the Double Narrative Process: The Agency of Students

In addition to adult human (e.g., data manager) and technological characters, the previous section introduces the agency of young students that comes into stage on an ongoing basis during the experimental phase of the game environment.

As the double narrative of actions, reactions, and interactions of technological and human entities unfolds, the agency of students as main characters becomes more visible in the feedback process of the experimentation.

As Lester goes on to argue,

... The long-term effects are having a strong potential for deeply motivating learning experiences and promoting learning characteristics for example like self-advocacy. These learning environments when they are done well have effective characters, and problem-solving guidance. Feedback is context-sensitive. Problems, which you can think of sort of narrative episodes, can be dynamically selected, and explanations can be tailored depending on the needs of students. So, this particular learning environment, Crystal Island, …, is one we have been working on for a very long time. And, in it the student plays a part of protagonist who actually goes to a remote island and finds out that members of their research team are falling ill.

As the integrated narrative process unfolds, the focus reverses into the wider context of the school, where students hold a protagonist (or main) role, as Lester argues. The agency of the students as main characters is manifested through opportunities to explore the environment as well as challenges underlying the learning situation. To deal with them, the students put their reasoning into action to come up with solutions. These actions match the needs and pedagogical objectives of science education. In the process, actions intertwine with the materiality of technology that itself acts to learn from the agency of the students in a situation that constitutes a differential diagnostic test, as Lester describes.

In the following section, Lester offers an account of the characteristics that make these environments attractive poles for thinking about how to integrate AI into learning.

One is that there is exploration of virtual environments. Two is that there are often very knowledge rich components in the environment. They can be sprinkled in to provide […] resources for students and their problem solving. There can be arbitrarily a simple or complex kind of virtual equipment, in this case for science education. They can support very complex reasoning. In this case it is for differential diagnosis. There can be multiple subject matters integrated. In this case it is science and complex informational text comprehension. And then stealth assessment is [a] really important area. I think [what is] promising in this particular metaphor is being able to combine assessment into the narrative.

So really three kinds of promising ways of thinking about interactive narrative. One is that it is a laboratory of investigating learning, super important from a research perspective. Second, it is a great place to study AI learning analytics because of the enormous data that these things produce. Often on very granular levels. And finally, the one I am myself particularly excited about and I imagine you might be as well, which is that it is kind of a lab for investigating new and very, very promising AI learning technologies. I want just to quickly say that there are lots and lots of domains, and lots and lots of student populations and actually lots of settings too that narrative is kind potentially applicable too. I just quickly mention passing, this is a narrative-centered learning environment for middle grade’s computational thinking that we have been working on for many years.

Then, what you have is, you’ve got all this data from student problem-solving interactions, and it’s all captured in the trace data. So, it’s all in the way that the student moves around in the world and manipulates artifacts, interacts with characters, takes these little stealth-assessments and so forth, but you’ve also got the ‘expert drama manager’ as they’re making the decisions about how the narrative should happen. And … that’s a supervised machine learning test.

This account speaks again to the need to pay close attention to the ways students and technology influence one another. In other words, how we relate with technology is a matter that matters, as it constitutes one ethical dimension of the role technology plays in AI-enhanced learning.

4 Agential Cuts in Narrative-Centered Learning Environments

As it was mentioned in previous sections of this chapter, the convergence of modalities in the audiovisual display allows the agency of the participants of the communicative event to emerge. Evidently, different forms and types of agency make an appearance here. As Stark (2016) argues, agency is, rather than constant, a fluid entity that intertwines and intra-acts in material objects and bodies through space and time. Under this lens, agential intra-action is seen as a dynamism of forces, rather than an inherent property of human beings (Barad 2007). It is this dynamism of forces that allows us to experience the world and, therefore, to relate with the world.

In a similar way, the perspectives that emerge though the convergence of modalities in this chapter are associated with a multiplicity of metaphors. These are metaphors of learning that, as Ricoeur has shown, make visible possibilities of reality that can orient agency and contribute to the effort to reshape reality. In this sense, the perspectives of the participants of the communicative events (e.g., keynote, interview, audiovisual, written text, etc.) actually constitute agential cuts.

Agential cuts are forces of bodies, objects etc. (Stark 2016) whose ongoing movement and intra-action transforms the way we understand the world. In AI-enhanced learning with narrative-based learning environments, Crystal Island is an example of agential cut that the agencies of both human (i.e., computer scientists, designers, researchers, teachers, other practitioners, students) and more-than-human (i.e., AI, digital technology, algorithms, etc.) entities both construct and transform.

Most certainly, there are issues for consideration here. It is debatable, for instance, whether the agency of technology is a valid notion. This discussion goes beyond the limits of this brief commentary. However, it might be worth mentioning here the example of the brittle fish (adapted from Barad 2007) as a response to the long-held belief that agency is associated with human consciousness only. The brittle star, a relative to the starfish, manages to develop a visual system to avoid ocean predators without the aid of actual eyes and brain. Without the brain organ, it is hard to imagine that survival is possible. Despite this, the brittle star survives thanks to the spherical calcite crystals covering its limbs and central body, functioning as micro-lenses that collect and focus light directly onto its diffuse nervous system. In this way, even without a nervous system, the brittle star manages to escape its predators and survive. Agency, therefore, seems to not link with brain function and consciousness necessarily.

5 Summarizing Remarks

This brief commentary aims to bring together perspectives that arise from the double narrative of multimodal texts (keynote and interview) and the agential cuts that emerge from metaphors of AI-enhanced learning and the entanglement of experiences and actions of scholars, researchers, teachers, and students. In this way, it touches upon the ethical dimensions of technology in AI-enhanced learning.

The multiplicity of metaphors includes both older and newer, conventional and novel ones. The tutorial dialogue, a metaphor that Lester (2021) introduces early in his keynote, is not new. The tutorial dialogue is traced back in history with the Socratic dialogues being possibly the first notable example of teacher-student interaction, where Socrates teaches his students logic, reasoning, argumentation, and ethics. Later, the narrative-centered learning environment emerges through practice, in time. Even this is not a new metaphor at all, it acquires new meanings when associated with AI-enhanced learning metaphors.

Some metaphors seem to be in fluidity, as their meaning transforms in time. The discussion in this commentary shows that these are mainly metaphors associated with agential cuts, that is, dynamic forces of human and more-than-human entities that move, intra-act, and transform in time and space.

Another conventional metaphor is the student being a protagonist in school environments with student-centered orientations. However, how the narrative of student-centeredness becomes believable remains an issue when it comes to AI-based learning. The example of Crystal Island seems to offer possibilities for engaged learning with the spaces it opens for exploration, experimentation, and the new roles it generates in its experimental process. The role of drama manager, as Lester describes it, resembles that of the tutor. It could be the basis for an agential cut in the future.

Its current orientation, however, seems to be targeting the improvement of technology exclusively. In ancient Greece, “drama” is a type of narrative and, as such, refers to the actions, events, and relations of the characters that shape its plot. The noun “drama” is associated with the verb δρω (Greek for /dro/ meaning “act”). The drama manager should then take care of how students relate with the world rather than how they interact with technology only.

The visualization of the future, as eloquently expressed by Lester (2021), takes the metaphor of AI for learning forward with the multicultural, inclusive Crystal Island. And yet, the technological metaphors are not enough. More thought and transdisciplinary discussions and collaborations are needed with scientists and pedagogues to articulate clearly how the agential roles of students and teachers are redefined in AI-enhanced learning.

As Lester rightly puts it, AI-enhanced learning will always be built on pedagogies of care and therefore the employment status of teachers is not threatened. Indeed, although the employment of workers has been the object of heated discussions since the 1950s brought up by the rapid advances of systems of automatization (Arendt 1998), the world will always need teachers who care.

In an era that is shaken by the COVID-19 pandemic and the larger questions concerning the sustainability of the planet, the role of teachers cannot be confined in the teaching of how technology functions. Teachers should be able to teach, among others, what environmental crisis means, what climate injustices are really about and who are most inflicted by them, what indigenous knowledges are and how they are downplayed. These could be part of science education curricula on the one hand. On the other hand, computer scientists need to think deeper how to integrate these realities into their algorithmic configurations. After all, science education does not happen in a vacuum.

And these can be some ways to move the narrative forward, having considered the crucial ethical questions that Lester poses at the very beginning of his talk, about «[W]hat happens when the AI, which this very powerful force, is kind of unleashed on the world».