Generative artificial intelligence, co‐evolution, and language education

Kern's (2024, this issue) anchor piece in this issue of The Modern Language Journal offers the opportunity for critical reflection on the role of (human) teachers amidst an ever-widening range of technologies to facilitate second language (L2) development. As Kern so accurately describes it, since March 2020 and the global onset of COVID-19, many language educators accustomed to teaching in face-to-face instructional settings were forced to move to remote instruction, often without the personal experience and infrastructural support that are present in specifically designed distance-learning formats. Language educators met this challenge, and, in my estimation, productive uses of technology across language teaching and learning modalities—including residential, hybrid, and distance contexts—have improved over the past few years. At this point in the history of applications of technology for language learning, there is nothing surprising about uses of video conferencing, social media, language tutorial websites and apps, online textbooks and grammars, translation tools, and video- and audio-based content (among others). As Internet theorist Clay Shirky has described it, "communications tools don't get socially interesting until they get technologically boring" (Shirky, 2008, p. 105), and indeed, there has been considerable interest in more deeply examining humans' now quotidian uses of digital technologies and modalities as they potentially transform language use and trajectories of development. Examples of incisive research in this area include applications of Vygotskian notions of mediation, flat(ish) ontology approaches such as sociomaterialism that reject a strict separation between human and nonhuman entities, and perspectives that expand the locus of human cognition from a brain-local focus and toward cognition and learning as embodied, embedded, enacted, and distributed across changing social, symbolic, and material contexts, including the incorporation of technologies. Let me add that the primary disruptive technology of the moment is generative artificial intelligence (GenAI), which presents many opportunities as well as considerable risks and challenges to conventionally oriented institutionally located instructed language learning. In the following sections, I begin with a discussion of the co-evolutionary dynamics intertwining humans and technology. I then describe some of my own (and others' facilitating) research on uses of technology in language education, address GenAI and its affordances and constraints, and consider the evolving role of language teachers in this era of digital technology ubiquity. The phylogenesis of our species has been closely linked to the invention and utilization of both material and symbolic technologies (systematic knowledge and processes) and tools (instances of broader technological systems and practices). Below, I provide a brief description of two co-evolutionary processes that involve technologies, tools, and the human body and brain. Archaeological evidence dating back to potentially as early as 2.5 million years ago indicates that our first tool-using ancestors, Homo habilis (the "skillful" or "handy" human), developed the Oldowan tool industry (stone axes and rough cutting implements), which the subsequent Homo erectus with their Acheulean stone industry refined into more effective shaped bi-facial tools. Across a few million years, tool creation and use selected for anatomical changes in the hand, in particular the emergence of a bony projection on the third metacarpal of the hand called the styloid process, which increases grip strength, manual dexterity, and precision. This example of co-evolutionary development affecting human anatomy is described by Ward et al. (2014) as "an increased reliance on manipulatory behaviors tool making, refinement, and use indicated by the archaeological record (…) selected for the modern human hand early in the evolution of genus Homo" (p. 121). Among Homo erectus and Homo sapiens populations, spanning from approximately 1.9 million years ago to the present, the contrapuntal emergence of symbolic reasoning and language, accompanied by the development of complex cultural practices, reflects a second phylogenetic co-evolutionary relationship. As described by Deacon (1997), this co-evolutionary process exerted selective pressure resulting in a period of remarkable brain evolution, ultimately culminating in our species' present-day capacity for complex communication and abstract thinking. The process involved a feedback loop where cultural and linguistic evolution influenced the structure of the brain, which in turn enabled more complex cultural and linguistic abilities. This process encompasses early forms of embodied communication, including gestures and bodily deixis, as well as vocalizations in the form of spoken language. More recently, it extends to the subsequent development of graphical communication technologies, such as pictographic, logographic, syllabic, and alphabetic writing systems, among others. In essence, humans, technologies, and tools have been entangled in co-evolutionary processes of mutual adaptation for a very long time. My own work over the past few decades has broadly aligned with a co-evolutionary perspective, if on a microgenetic time scale, focusing particularly on how digital tools in mediational roles with humans evolve diverse identities and uses across user populations. Earlier in my career, I happened upon the biological distinction between the genotype and phenotype of organisms that, through analogy, radically shifted my understanding of contemporary technologies (find the backstory in Thorne, 2016). A genotype is the genetic makeup that an organism possesses for sets of traits (e.g., design features of a digital tool), while a phenotype is the observable characteristics of an organism resulting from the interaction of its genotype with the environment, which can result in different phenotypes depending on environmental conditions (e.g., different ways people routinely utilize the same digital tool). The relationship of genotype–phenotype will be revisited in the discussion of GenAI a bit later. Initially employing a Vygotskian theoretical lens emphasizing mediation, I investigated numerous technology-associated world language education interventions to explore how individuals' experiences of socialization into particular digitally mediated speech communities influenced their framing and use of communication technologies, with corresponding effects on interactional dynamics, volumes of language use, and interpersonal relationship building. This research led to articulating the concept of cultures-of-use of digital tools (Thorne, 2003; Thorne Thorne, 2016). Framed as a question, how do heterogeneous networks of humans, tools, and other entities in the social–material world jointly accomplish activity (Latour, 2005; Thorne et al., 2021)? Contemporary examples of humans interacting with digital tools that appear to make agentive contributions to communicative action include predictive text or autocompletion features that populate Internet search tools and many texting, email, and word processing applications; autocorrect software for spelling and grammar; speech-to-text and text-to-speech software; and voice assistants (Siri, Google Assist), to list only a few. Mapping from Bourdieu's (1979) description of the generative dispositions resulting from socialization processes, these are all "structured structures predisposed to function as structuring structures" (p. 72), which when applied to our focus on the digital technologies mentioned above, describe massive corpora representing prior instances of human language use (structured) to create tools (structures) that subsequently structure (assist, inform, co-create) the communicative and informational actions of people. This leads to the next topic, GenAI and its complicated incorporation into language teaching and learning. The arrival of GenAI in the public sphere, beginning with the release of ChatGPT-3.5 to the public on 30 November 2022, has had a substantial impact on education. Current GenAI tools include OpenAI's ChatGPT, Microsoft's Copilot, Google's Gemini, and Anthropic's Claude, among others. Let me begin by stating the obvious: The most problematic term in GenAI is "intelligence." As of this writing, there is no evidence linking GenAI to intelligence, thinking, or consciousness as these attributes relate to humans. Focusing only on written language, generative large language models, such as GPT-4, are machine learning systems pretrained on massive datasets (the size depends on the language and/or domain). During pretraining, these models infer patterns and associations from language data, enabling them to generate humanlike responses to user prompts. GenAI's genotype is a deep learning and recurrent neural networks algorithmic architecture. Its phenotype can sometimes appear human like, including dropping register, adding emoticons, and the like. But there are (currently) clear limitations. In a recent research project focused in part on uses of ChatGPT for L2 pragmatics, we found that in various conversational role-play scenarios, despite repeated requests for ChatGPT to limit its responses to 20 words, it consistently violated Grice's maxim of quantity and was intensely and consistently too verbose, often producing an extended response to short conversational statements and questions (Sydorenko et al., 2024). GenAI presents much more serious and broad-reaching issues than verbosity. Informational biases are pervasive within AI systems as a function of the biases inherent in (human-generated) large language model training data. The result is that GenAI algorithms perpetuate and amplify societal inequalities in areas such as gender discrimination, criminal sentencing, and ethnic profiling. When queried directly about this, ChatGPT confirmed that "biases within AI systems often result from biases present in the training data, and these biases can indeed be perpetuated and amplified by the algorithms." This is an area where human educators play a crucial role in teaching students how to effectively navigate and assess the information generated by GenAI within, and beyond, educational contexts. A seemingly intractable concern regarding GenAI is academic dishonesty. Students may (and do) misuse GenAI to generate plagiarized content without understanding—or even substantively addressing—the subject matter, undermining the integrity of assessments and devaluing the educational process. Additionally, students may become dependent on these systems, relying on pregenerated answers and texts rather than building their own composition skills as writers. To mitigate this issue, educators need to develop pedagogical approaches, such as those described by Kern (2024, this issue), that encourage students to use GenAI to support learning rather than as a substitute for their own L2 communicative abilities. On the positive side, GenAI has the potential to revolutionize language education in profound ways. With vast knowledge bases and natural language processing capabilities, GenAI empowers students to access information and engage in personalized learning experiences. AI-powered virtual tutors and adaptive learning platforms provide personalized feedback and guidance tailored to individual students' needs, fostering an effective and efficient learning process. GenAI can facilitate collaboration and knowledge-sharing among students by providing a (verbose) conversational partner and answering questions about grammar and usage. For academic writing in many world languages (but not all—attempts to use GenAI with the Manchu language were unsuccessful, suggesting that many low-population languages are not yet represented in large language models), GenAI can be used to assist with editing and stylistic corrections and can offer feedback and pedagogical explanations relating to academic register use, which can support students who may be disadvantaged, minoritized, or who are struggling with academic and other varieties of written language. In these contexts, GenAI has the potential to assist world language students, first-generation university students, and students at all levels for whom English (and other languages of wider communication) are additional languages to create more inclusive, accessible, and responsive learning environments that are adaptive to the needs and learning objectives of diverse learners. While GenAI has transformative potential in language education, it is essential to address these aforementioned challenges proactively to ensure its responsible and effective use to enhance learning. Since GenAI tools can so easily reduce the cognitive load of many tasks, we need to foster a culture shift in education that emphasizes and supports a desire to learn with these powerful new tools and not to use them to complete assignments and learning activities with minimal effort. This is a tall order, but one that must be central to our concerns and practical uses of GenAI as we develop critical and developmentally focused pedagogies for utilizing these tools for discrete purposes, for example, refining written texts, brainstorming ideas, gathering contrasting points of view, and searching for regional information on localized variants of named languages. As a broad generalization, accepting that there are personal rewards to individually oriented discovery and learning, and for many a pleasure emerging from the activation of neural reward circuits associated with gamified language-learning technologies (i.e., Duolingo and related services), the ultimate goal of additional language learning is to develop an enhanced capacity for communication that initiates or strengthens social bonds and feelings of connection and belonging. With numerous others in the field, I have long espoused the idea that language learning is accelerated when framed as a resource for developing, maintaining, and deepening social relationships of significance that foster experiences of intersubjective alignment, empathy, and emotional resonance. Linguistic, intercultural, and interactional competencies are necessary dimensions of language learning, each usefully investigated in analytically bounded L2 research and expressed in pedagogical approaches. And we know that learning of any kind, especially the case with complex objectives such as language learning, requires large volumes of effortful engagement, totaling thousands of hours, to result in robust learning outcomes. For the vast majority of language learners, implicit and nebulously framed future-self ideations involving work, travel, and aspirations for interpersonal relationships—or explicit and narrowly framed ambitions, such as becoming a supply chain consultant in a Mandarin-speaking environment or more closely aligning with (actual or potential) romantic partner(s)—are goals that fuel the thousands of hours of commitment that most students need to attain advanced levels of linguistic, intercultural, and interactional abilities in their language of study. Human teachers are positioned to enhance and effectuate these goals through their abilities to encourage, motivate, provide corrective feedback, describe and model pragmatic and interactional norms, and situate the language-learning process among the relevant domains of culture, history, sociology, and literary expression. These abilities extend far beyond explanations of the "aboutness" of linguistic competence, which, while necessary, is a thin and ultimately not-satisfactory-enough intellectual enterprise to drive continued language study for most students. How can we integrate GenAI into language education to develop abilities that are relevant to social and work–life motivations? How do we ensure that GenAI does not exacerbate existing inequities? And further, how can we create conditions that couple human intelligence and creativity with GenAI tools in order to support full participation in society by a diversity of individuals and communities? While GenAI has transformative potential in higher education, it is essential to address such questions to ensure its responsible and effective use for enhancing equity and learning. A new kind of work needs to be done to create conditions for engaged language learning in instructional contexts. I fully agree with Kern (2024, this issue) that human teachers motivated by an inquiry-based approach, who are interactionally attuned and pedagogically skilled, have a vital and continuing role in language learning. Like many students, I still recall a large number of transformational teachers and professors from my past (and among collaborators, continuing into the present). Personally, I am pleased that GenAI also has become a surprisingly useful interlocutor. Fully acknowledging the limitations and risks, we have an opportunity for the co-evolutionary advancement of instructed language learning through the joint accomplishments possible with GenAI. With the plethora of digital tools now available, I am reminded of Apple co-founder Steve Jobs's 2007 keynote address introducing the then-nascent iPhone. He began by stating that "every once in a while, a revolutionary product comes along that changes everything." Many commentators have made similar claims about GenAI and the future of work, knowledge production, and formal education. We currently do not know if these prognoses of radical transformation are accurately predictive of our collective futures. In the same keynote address 17 years ago, Jobs also quipped, to applause and laughter, that "we want to reinvent the phone. Now, what's the killer app? The killer app is making calls!"1 He moved to then describe the relative difficulty of making calls with then-current cell phones and the iPhone's improved contact list feature. But what I want to emphasize is that amidst unveiling the many iPhone apps for informational and computational feats, his self-described "killer app" is the high social presence, foundationally primordial modality of human-to-human voice communication (this from my generational standpoint, accepting that a great deal of voice-related phone use has been displaced by texting). With GenAI and other digital tools framed as critically appraised allies, I remain committed to the idea that language learning is ultimately driven by the human relationships that it makes possible. For many, living is now intertwined with information and communication that propagates and refracts across multiple media. How can instructional and language learning practices co-evolve with GenIA to articulate preferred futures that moderate problems (e.g., academic integrity) while also amplifying affordances for personalized learning, dialogic interaction, and editorial assistance and feedback that, through an equity lens, has the potential to support emergent bilinguals (García, 2009; here in application to all who wish to learn additional languages)? It is imperative to collectively forge these likely multiple futures as education co-evolves with emerging technologies. In my view, teachers remain a keystone species in this human–nonhuman relational ecology.