Dr Amit Kumar Dwivedi
Research Supervisor
University of Technology
Jaipur

Abstract

Generative artificial intelligence (GenAI) is reshaping the landscape of teaching, learning, and assessment. Its rapid expansion in educational settings has brought forth a set of pressing issues: what will become of social intelligence in education moving forward?

This paper untangles a core set of tensions: the high efficiency of GenAI clashes with the unique, irreplaceable human capacities of empathy, moral judgment, and relational learning. These capacities are the foundation for building genuine connections between people and achieving deep, meaningful growth in educational contexts—no technological tool can replace their role. GenAI can help people quickly organize materials, generate first drafts, and handle cumbersome administrative tasks; these efficiency gains are tangible. But if it is allowed to seep unbound into the core processes of education, these uniquely human core capacities will lose the space to develop. This is precisely why this paper confronts this set of tensions head-on.

Drawing on UNESCO’s human-centered AI guiding principles, trustworthy AI frameworks, and the latest research in the field of social-emotional learning (SEL), this paper argues that the use of GenAI must be constrained by educational and ethical principles to safeguard human autonomy and enable genuine human-AI collaboration. All of these referenced standards and studies point to a single core: technology must not lead education; AI must always serve human growth itself. Only by upholding this baseline can human-AI collaboration avoid devolving into human replacement by AI.

This paper also proposes a theoretical framework to cultivate human social intelligence sustainably while leveraging technological advantages. This framework prioritizes Critical AI Literacy and Ethical Awareness, Dialogic Pedagogy, and Institutional Safeguards. To balance the conveniences brought by technology and the cultivation of core human capacities, we cannot stop at vague advocacy; we must implement concrete actionable pathways: first, ensure all teachers and students can understand AI and recognize its limitations; second, deploy AI through teaching methods that stimulate deep dialogue between people; and finally, rely on rules from schools and the entire education system to prevent misuse of technology from eroding the humanistic core of education.

At the end of the paper, recommendations are put forward for policy development, frontline practice, and future research directions respectively, to ensure that GenAI can ultimately serve an educational future that is more inclusive, equitable, and always preserves its humanistic core.

Keywords: generative AI, social intelligence, social-emotional learning, human-centred AI, education ethics

Generative AI and Social Intelligence: Balancing Technology and Humanity

As generative artificial intelligence (GenAI) rises in the education sector, the field’s reactions have split in two: on one side is excitement over its potential to drive new transformations, and on the other is anxiety over its unforeseen risks. Common generative AI tools such as large language models (LLMs) can draft articles for users, generate complete teaching plans, and simulate one-on-one tutoring conversations. According to UNESCO (2023), these capabilities hold the promise of delivering unprecedented personalized teaching outcomes and elevating the efficiency of the entire education chain to new heights (UNESCO, 2023). Yet it is precisely these seemingly highly practical capabilities that are gradually undermining the foundational bedrock that makes learning possible: the social and emotional core that enables people to learn together—empathy, trust, ethical reasoning to distinguish right from wrong, and the collaborative ability for groups to sort through ambiguous problems and build shared consensus. These capacities are not cold technical metrics; they are the core of education as a warm, shared enterprise for all humanity. Holmes (2023) and Miao & Holmes (2023) both highlighted this core, which is threatened by technology (Holmes, 2023; Miao & Holmes, 2023).

The core question this paper seeks to answer is clear: how can the current education system harness the potential of generative AI, while not only retaining but further strengthening the uniquely human social intelligence? Centered on this question, this paper puts forward three core judgments. First, the fundamental nature of generative AI is to generate content by learning patterns from massive volumes of data; it is inherently incapable of producing genuine, human social understanding, so its use in any scenario must always be overseen by humans, and decision-making power must never be handed over to AI. Second, the social-emotional learning (SEL) framework commonly used in education today is the core metric to assess what role generative AI should play in classrooms and what boundaries it must not cross. Third, a teaching framework that centers human needs from start to finish can align the pace of technological innovation with the goals of talent development, using AI to boost efficiency while systematically cultivating students’ core capacities to build deep relationships, uphold ethical standards, and communicate smoothly with people from diverse cultural backgrounds.

Defining Generative AI and Social Intelligence

Generative AI is a category of specialized artificial intelligence systems, whose core operating logic is to first learn universal patterns of all types of content from massive datasets, then use those patterns to generate entirely new content—such as never-before-seen text, original images, runnable code, or non-existent audio clips. This definition draws on relevant research from UNESCO, 2023. Unlike older generations of artificial intelligence that preceded it, most earlier AI could only perform two types of tasks: either categorize existing content, or predict subsequent developments based on existing information; it was completely incapable of producing output “from nothing.” But generative AI can genuinely “create,” and often uses strikingly fluent expression to mimic human speech and creation, such that many people cannot distinguish at first glance whether content was written by AI or a human.

However, Miao and Holmes (2023) long ago warned that generative AI has an unavoidable core limitation: it “cannot produce truly original ideas, nor solve real-world problems, because it does not understand the actual objects and social relationships that underpin language” (p. 8). All the content it outputs is stitched together based on probability—if certain words or segments of content frequently appear together, it assembles those elements. It has no real-life experience to support its output, nor does it generate content based on its own moral judgment; put simply, it has no understanding of what the content it outputs actually means.

In direct contrast to the logic of generative AI, social intelligence refers to a uniquely human core capacity: the ability to perceive and interpret the social signals others emit, the emotions hidden in their words and actions, and changes in relationships between people, and further to make appropriate responses based on that information. This definition comes from Goleman, 2006. When applied to concrete educational contexts, this capacity manifests in clear, observable ways: a teacher can acutely pick up on a student’s unspoken needs, for example recognizing that a student is distracted in class because of troubles at home, not because they are unwilling to learn; groups of students can collaborate to solve difficult problems that no single person could tackle alone; and the entire school community can engage in inclusive dialogue that excludes no one, so no person is silenced because of their unique identity. The current social-emotional learning (SEL) framework breaks these broad capacities down into five specific competencies that can be taught through instruction: self-awareness to recognize one’s own emotions and state, self-management to regulate one’s own words, actions, and emotions, social awareness to understand others’ circumstances and emotions, relationship skills to get along well with all people, and responsible decision-making to make accountable choices when faced with problems. This framework originates from CASEL, 2020. Even though today’s generative AI can simulate human conversation and appear indistinguishable from a real person, it lacks genuine empathy, the ethical sense of responsibility to answer for its own words and actions, and the sensitivity to grasp the unique specifics of a given context—these three elements are the core of true social intelligence.

Risks of GenAI to Social-Emotional Foundations

The global guideline released by UNESCO in 2023 lists eight core controversies sparked by generative AI, several of which directly impact human social intelligence—the core capacity to perceive emotions and manage interpersonal relationships, which forms the foundation of all human social activity.

First, generative AI may widen the digital divide: users who can access paid premium models will gain overwhelming advantages, while others will be marginalized, unable to access equal learning and development resources (see original source, p. 14). Second, its training data is all crawled from the internet, and these raw datasets already contain various biased judgments; AI will replicate the biases in the original data verbatim, silencing the voices of marginalized groups and reinforcing existing social stereotypes, making prejudice increasingly difficult to dismantle (see original source, p. 17). Third, generative AI will produce “hallucinations”—content that sounds plausible but is completely factually incorrect. If people accept this fabricated information uncritically, it will gradually erode trust between people, and erode humans’ capacity for independent thought and critical judgment to distinguish right from wrong, leaving people unable to verify the truth of information on their own (see original source, p. 16). Fourth, if learners rely excessively on generative AI to write and conduct reasoning, their stamina to sustain long periods of deep thinking, and their capacity for genuine, personal expression shaped by their own experiences and warmth, will gradually degrade. One commentator refers to this phenomenon as “cognitive debt” (Arar & Zhou, 2026).

These risks are not confined to the informational and cognitive levels; when applied to the social-emotional domain, they become far more severe. AI chatbots cannot genuinely care about students; they lack human emotions and empathy, and cannot meet students’ true emotional needs. Yet their human-like design will lead people to form false emotional bonds when interacting with them, crowding out the genuine human relationships that should exist between people—the deep, mutually supportive connections that enable collective growth (Holmes, 2023). One educator noted, “AI might help practice some social-emotional learning (SEL) skills, but it cannot replace adults—adults’ role in teaching these skills and building genuine relationships cannot be taken by AI” (Ed Week, 2026, para. 4). Furthermore, generative AI suffers from a convergence bias, always producing standardized, error-free “safe” content that avoids any unconventional expression, which will stifle the creative dissenting opinions and diverse perspectives that democratic classrooms depend on. A classroom that can hold all voices and let everyone speak honestly relies on those differing perspectives to function (UNESCO, 2023, p. 37).

Opportunities for Enhancing Social Intelligence

Even with all the aforementioned risks, generative AI (GenAI) can still deliver many tangible benefits to social-emotional learning (SEL), provided that we design its application scenarios in a structured manner and use it in full compliance with ethical standards at all times. Recent research has confirmed that AI-powered simulated scenarios, with AI acting as an intermediary, can help people practice perspective-taking, learn to regulate their own emotions, and study to communicate smoothly with people from different cultural backgrounds (Frontiers in Education, 2026; Monash University, 2025).

To give a concrete example, Multilingual Gen-AI can simulate various real-world scenarios for role-play, allowing pre-service teachers to practice difficult, uncomfortable conversations with virtual students generated by the system. AI can also provide real-time feedback to these pre-service teachers, pointing out issues in their tone, inclusivity, and cultural adaptability in their previous interactions, to help them adjust and improve (Frontiers in Education, 2026). In a similar application, AI tutors can create a judgment-free safe space for all learners, letting them voice the emotions they hide in their hearts, repeatedly practice navigating complex social scenarios, and even gradually explore who they are and who they want to become. These functions deliver exceptional value for neurodivergent students or marginalized, vulnerable learners (Atlantis Press, 2026).

The most critical point is that for these AI applications to deliver real, lasting results, there is only one prerequisite: they must only act as auxiliary partners to human-led instruction, and must never replace the core authority of human teachers to oversee the learning process. Holmes (2023) put forward the concept of “social generative AI”: these AI systems, as participants in dialogic learning communities, always recognize human teachers as the ultimate arbiters of meaning-making and emotional care. In this framework, GenAI acts as an “engine of possibility” or a Facilitating Socratic adversary, only posing questions to spark deep reflection among participants, never directly outputting so-called standard answers (Holmes, 2023, Table 1). This mode of use aligns perfectly with the metacognitive and ethical decision-making principles that CASEL’s (2020) has long emphasized, positioning AI as a tool to deepen, rather than replace, connections between people.

A Human-Centered Framework for Balancing Technology and Humanity

To untangle these clashing tensions, this paper proposes a framework built on three core pillars. This framework is not constructed from thin air; its foundations are UNESCO’s (2023) human-centered AI principles, and European Commission’s (2019) ethics guidelines for trustworthy AI.

1: Critical AI Literacy and Ethical Awareness

Teachers and students must all develop Critical AI Literacy and Ethical Awareness, meaning the ability to unpack the applicable scenarios, limitations, and ethical issues of generative AI (NIST, 2024). This set of competencies is not complex to explain; its core is to avoid treating AI as an all-powerful tool, and first grasp its fundamentals: learn to understand the training logic of these models, identify biases hidden in algorithms, and the “hallucinatory” content fabricated out of thin air, and be able to assess what impact they will have on educational equity and individual autonomy.

Arar and Zhou’s (2026) Instructional Model for Human-Centered Generative AI Engagement includes a supporting five-step pedagogical approach to help people build this capacity step by step: (1) build critical and ethical awareness, (2) master prompt literacy, (3) use AI to support learning, (4) reflect on and revise AI output, (5) independently apply the knowledge learned.

Integrating this cyclical process into schools’ formal curricula will help everyone develop a purposeful, reflective habit of using AI, always placing human judgment ahead of automated tools (Arar & Zhou, 2026, p. 5).

2: Dialogic Pedagogy and Human–AI Collaboration

Introducing generative artificial intelligence into teaching scenarios cannot be done through a simplistic, blunt add-on approach; it must use dialogic, constructivist methods whose core is to always prioritize human relationships, never letting tools erode the most critical interpersonal connections in teaching. The “social generative AI” model proposed by Holmes (2023) outlines exactly this kind of classroom scenario: humans and AI can build knowledge piece by piece through sustained, sequential conversations. Throughout the entire process, the teacher retains leading power, on one hand guiding the group’s exploration in the correct direction, and on the other hand upholding ethical boundaries to prevent learning from veering off track. For concrete implementation, these practical methods can be tested:

Treat generative AI as a brainstorming partner, using it to quickly throw out all kinds of unconstrained ideas to open up thinking, but require students to identify the problems and flaws in its output themselves, then revise and improve them one by one, rather than using the content directly.

Design collaborative group projects, allowing AI to produce a first draft to lay a basic framework, but the final content must be refined by the student team, who add context-specific, unique information, before presenting and sharing it to the class.

Organize Socratic debates, having students discuss and challenge the various viewpoints generated by AI, in the process refining their own critical thinking and empathy (Holmes, 2023).

All these teaching practices align with Vygotsky’s social learning theory, positioning AI as a “more experienced guide” whose role is to scaffold and support the pace of interactions between people, rather than replacing genuine human connection. Those face-to-face exchanges, acts of empathy, and intellectual collisions are the irreplaceable core of teaching (Vygotsky & Cole, 1978, as cited in Holmes, 2023).

3: Institutional Safeguards and Policy Alignment

Educational institutions must establish formal governance mechanisms to ensure that the use of generative artificial intelligence on campus complies with ethical standards. Core measures to be implemented include:

Adopting UNESCO’s (2023) recommendations: mandate the implementation of data privacy protections, set age-appropriate usage limits, and verify that AI tools meet teaching requirements.

Implementing NIST’s (2024) AI Risk Management Framework, which outlines 12 GenAI-specific risks and corresponding targeted mitigation strategies.

Enact clear relevant policies covering academic integrity, disclosure requirements for teachers and students using AI, and rules for equitable access to licensed GenAI tools (OECD, 2024).

These safeguards will not only protect the fundamental rights of learners, but also reserve sufficient space for schools to develop compliant, human-centred innovative AI applications, without banning legitimate uses of AI to avoid risks.

Implications for Policy, Practice, and Research

The human-centred framework proposed in this paper can provide actionable, concrete direction for practitioners across all roles in the education sector.

For policymakers, it repeatedly emphasizes that immediate action must be taken to introduce regulations overseeing generative AI service providers, ensuring that all applications of these AI tools in educational settings remain transparent, fair, and aligned with the core values of education (citing UNESCO, 2023).

For frontline educators, it requires schools and districts to deliver professional training on AI literacy and dialogic pedagogy, to help teachers build the relevant capacities to design complete learning activities that use AI to boost learning efficiency while cultivating students’ social intelligence (citing Frontiers in Education, 2026).

For researchers working in the field, it clearly states that long-term empirical research must be launched to track the long-term impacts of generative AI on social-emotional learning, with a particular focus on how these impacts manifest for marginalized learners (citing Monash University, 2025).

Future research should explore:

Three specific directions require attention in future research. The first is how simulation scenarios mediated by generative AI will impact humans’ capacity for empathy and cross-cultural competence. The second is to identify concrete methods to resolve the “cognitive debt” spawned by AI, while preserving the deep thinking required to work through challenges during learning. The third is to explore human-AI collaborative teaching models that strengthen, rather than weaken, teacher-student relationships.

Conclusion

Generative AI is neither a panacea that can solve all longstanding problems in education, nor an irreversible threat that will completely destroy core teaching models. It is simply a tool, and what impact it ultimately delivers depends entirely on the choices humans make. This conclusion is not baseless; applying it to real educational contexts makes it clear: it cannot on its own fill resource gaps in education or eliminate all inequities, nor will its introduction completely destroy the teacher-student relationship in education. Its role and impact have always been in the hands of the people who use it.

We must root the rollout of generative AI in social-emotional learning, critical literacy, and ethical governance. Only by anchoring ourselves to these three directions can the education system leverage its unique technological advantages while preserving the core capacities that define human intelligence—the ability to build connections between people, empathy, and the democratic capacity to support public discourse. These capacities are the fundamental bedrock only humans can uphold; we must never abandon these core elements simply because we use a tool.

As UNESCO (2023) emphasized: “AI must not usurp human intelligence. Rather, it invites us to reconsider our established understandings of knowledge and human learning” (p. 7).

The path forward must not follow the unexamined logic that “technological development is inevitable,” assuming that because AI has arrived, we can only follow its rhythm and cede all decision-making power to technological iteration. Instead, we must actively choose values-centred design solutions that prioritize the human growth and the essence of education that we cherish, always placing humans at the centre of AI’s educational applications.

References

Arar, K., & Zhou, L. (2026). Prompts to practice: A pedagogical framework for human-centered          generative AI engagement. Open Praxis, 18(3), 1191.             https://doi.org/10.55982/openpraxis.18.3.1191

Atlantis Press. (2026). Generative AI-driven emotional education for students.             https://www.atlantis-press.com/article/126022367.pdf

CASEL. (2020). What is SEL? Collaborative for Academic, Social, and Emotional Learning.         https://casel.org/what-is-sel/

EdWeek. (2026, February 23). AI is poised to reshape social-emotional learning. But for better         or worse? https://www.edweek.org/technology/ai-is-poised-to-reshape-social-emotional-       learning-but-for-better-or-worse/2026/02

European Commission. (2019). Ethics guidelines for trustworthy AI. https://digital-            strategy.ec.europa.eu/en/library/ethics-guidelines-trustworthy-ai

Frontiers in Education. (2026). Multilingual Gen-AI simulations for social-emotional learning in       teacher education. Frontiers in Education, 11, 1874639.             https://doi.org/10.3389/feduc.2026.1874639

Goleman, D. (2006). Social intelligence: The new science of human relationships. Bantam            Books.

Holmes, W. (2023). Towards social generative AI for education: Theory, practices and ethics.       Learning and Teaching in Higher Education, 1(1), 1–12.             https://doi.org/10.1080/23735082.2023.2261131

Miao, F., & Holmes, W. (2023). Guidance for generative AI in education and research.          UNESCO. https://doi.org/10.54675/EWZM9535

Monash University. (2025). Social-emotional learning and generative AI: A critical literature         review. https://research.monash.edu/en/publications/social-emotional-learning-and-  generative-ai-a-critical-literature/

NIST. (2024). Artificial Intelligence Risk Management Framework: Generative AI profile (NIST AI 600-1). National Institute of Standards and Technology.             https://nvlpubs.nist.gov/nistpubs/ai/NIST.AI.600-1.pdf

OECD. (2024). The potential impact of artificial intelligence on equity and inclusion in         education. OECD Publishing. https://doi.org/10.1787/15df715b-en

UNESCO. (2023). Guidance for generative AI in education and research. United Nations            Educational, Scientific and Cultural Organization. https://doi.org/10.54675/EWZM9535