PaperlayerView exact revision on arXiv

Source-linked AI summary

Towards social generative AI for education: theory, practices and ethics

Mike Sharples

arXiv:2306.10063v1cs.CYcs.AI

TL;DR

The paper addresses the gap between prompt-response views of educational AI and a social conception of learning through continual human-AI conversation. It develops this conception through examples of collaborative learning roles and argues that fuller participation requires capabilities such as memory, reflection, and care, while current systems remain limited in these respects.

  • Problem

    Educational GAI is commonly treated as an individual prompt-response system, leaving open how humans and AI could participate in distributed social learning.

  • Method

    The paper develops a social-generative-AI framework through systems theory, learning-process analysis, educational role examples, and ethical design requirements.

  • Results

    The paper identifies current GAI roles in collaborative learning and argues that deeper participation requires acquiring, consolidating, remembering, and transferring knowledge.

  • Takeaways & Limitations

    Social generative AI for education should be designed as a conversational, distributed system that combines machine capabilities with human teachers’ expertise and learner diversity.

  • Takeaways & Limitations

    Current GAI lacks long-term memory, reflection on outputs, knowledge consolidation across conversations, and sensitivity to learners’ emotions and cultural contexts.

Abstract

from arXiv · show

This paper explores educational interactions involving humans and artificial intelligences not as sequences of prompts and responses, but as a social process of conversation and exploration. In this conception, learners continually converse with AI language models within a dynamic computational medium of internet tools and resources. Learning happens when this distributed system sets goals, builds meaning from data, consolidates understanding, reconciles differences, and transfers knowledge to new domains. Building social generative AI for education will require development of powerful AI systems that can converse with each other as well as humans, construct external representations such as knowledge maps, access and contribute to internet resources, and act as teachers, learners, guides and mentors. This raises fundamental problems of ethics. Such systems should be aware of their limitations, their responsibility to learners and the integrity of the internet, and their respect for human teachers and experts. We need to consider how to design and constrain social generative AI for education.

1 Introduction

Generative AI has progressed from narrowly defined language-model tasks toward scaled tools embedded in everyday platforms. The paper proposes social generative AI as the next step and examines its educational possibilities.

  • 1 Introduction: LLMs have followed a trajectory resembling the World Wide Web, moving from narrowly defined tasks toward emergent properties and widespread deployment.The paper compares text completion with early web information retrieval and notes later embedding in tools and social media.
  • 1 Introduction: The next proposed step is social generative AI, examined here for education.The paper frames this shift as following rapid scaling and tool integration.

2 A systems view of generative AI in education

The paper reframes educational AI as a distributed social-learning system in which humans and AI converse through a pervasive computational medium. This perspective extends cooperative learning while raising questions about goals, agreement, and expertise.

  • 2 A systems view of generative AI in education: Emerging tools already integrate GAI with office software, internet search, external plugins, autonomous goal-setting, prompt revision, and long-term memory.The examples include Microsoft Copilot, ChatGPT Plus, AI game characters, and AutoGPT.
  • 2 A systems view of generative AI in education: The proposed educational system treats humans and AI as language agents conversing within a pervasive computational medium.This perspective draws on Pask’s idea of persistent conversations among language processors.
  • 2 A systems view of generative AI in education: Conversation is presented as a continual process through which agents explore differences, gain experiences, and reach agreements.The account links conversation with reflection, questioning assumptions, and mutual understanding.
  • 2 A systems view of generative AI in education: Social learning with AI raises questions about how systems should support learning conversations, mutual agreements, and the roles of teachers and experts.These questions arise because the distributed medium is not grounded straightforwardly in truth and reality.
  • 2 A systems view of generative AI in education: A systems view distributes cognition across humans and AI, using internet tools and the web as a medium for social learning.The paper connects this view to cooperative learning involving shared goals, discussion, and mutual understanding.

3 New roles for GAI in social learning

The paper presents current GAI as a collaborator in several forms of cooperative and social learning. Students can use it to broaden perspectives, argue, design, explore data and games, and create diverse stories.

  • 3 New roles for GAI in social learning: Current GAI can act as a generator of possibilities, argument opponent, design assistant, exploratory tool, and creative-writing collaborator.These roles are presented as examples of cooperative and social learning applications.
  • 3 New roles for GAI in social learning: Students can broaden perspectives by generating multiple responses, rephrasing prompts, and critically comparing AI output before writing essays.The activity combines group discussion with individual writing based on the explored material.
  • 3 New roles for GAI in social learning: In argumentation, students challenge ChatGPT to clarify or defend a position through a continuing dialogue.The example concerns whether conflict can be fruitful and uses a respectful, constructive culture as the defended position.
  • 3 New roles for GAI in social learning: In collaborative design, students use ChatGPT to research needs, define problems, challenge assumptions, brainstorm, prototype, and test solutions.They can adjust the system’s temperature setting to vary creativity and unexpectedness.
  • 3 New roles for GAI in social learning: Exploratory uses include visualising databases and generating varied language games that help students map principles of game design.Different runs can produce word builders, word chains, sentence swappers, and scavenger hunts.
  • 3 New roles for GAI in social learning: Collaborative storytelling uses multiple generated versions to represent diverse views, cultures, and orientations while avoiding stereotypes.The example prompts ChatGPT to avoid racial and sexual stereotypes and clichéd language.

4 Generative AI as a full participant in social learning

The paper argues that deeper participation in social learning requires GAI to acquire, consolidate, remember, and transfer knowledge. It proposes more capable, transparent systems while identifying missing memory, reflection, and care.

  • 4 Generative AI as a full participant in social learning: Deeper participation in social learning would require GAI to acquire, consolidate, remember, and transfer knowledge.The paper distinguishes this agenda from merely assisting current collaborative activities.
  • 4 Generative AI as a full participant in social learning: The learning model includes goals, new knowledge, consolidation, deeper understanding through research and dialogue, and transfer.The model applies to individual or group learners beginning with prior knowledge, disposition, and motivation.
  • 4 Generative AI as a full participant in social learning: Current GAI acquires surface performance knowledge through pre-training and consolidates it through human fine-tuning and external tools.The paper treats deeper understanding and theory of mind as controversial extensions rather than established capacities.
  • 4 Generative AI as a full participant in social learning: Current GAI lacks long-term memory and cannot reflect on outputs or consolidate knowledge from each conversation.The paper also states that the learning model omits affective and experiential aspects of being a learner and teacher.

5 Embedding care in generative AI

Social learning requires GAI to treat interaction as a responsible, respectful relationship rather than merely a task-completion problem. Current systems lack awareness of learners’ emotions, cultural context, and broader social implications, while affective-computing approaches remain limited.

  • Socially responsible GAI must understand learning contexts, adapt to participants’ needs and preferences, and respect their rights and dignity.Care is framed as a commitment to fulfilling responsibilities respectfully and empathetically, not as an emotion.
  • Current GAI lacks awareness of learners’ emotions, feelings, cultural context, and the broader social consequences of its actions.It also cannot reliably understand social nuances, empathize with learners, or provide emotional support.
  • Affective-computing systems use sentiment analysis to adjust responses, but remain limited in understanding and responding to complex human emotions.

6 Ethical AI for social learning

Ethical social learning requires a systems-level approach in which GAI follows human-rights principles, supports learner agency and diversity, and makes its reasoning and evidence understandable. The paper also identifies cooperation, teacher respect, and transparent hybrid architectures as design requirements.

  • Systems and principles: Ethical social learning requires configuring the entire human-AI system responsibly, beginning with universal human-rights principles.The paper uses Constitutional AI as an example of training a model to evaluate outputs against such principles.
  • Systems and principles: All GAI elements should follow shared principles that support learners, enable personal and cultural diversity, and promote cooperation rather than competitive selfishness.
  • Capabilities and agreement: Learning-oriented GAI should set explicit goals, retain long-term memory, model users, reflect on outputs, learn from mistakes, and explain its reasoning.The proposed systems combine neural AI for media generation with symbolic AI for representing and reasoning about people and the world.
  • Capabilities and agreement: Humans and AIs need verifiable evidence and understandable reasoning about what, how, and why a continuing conversation proceeds.
  • Agreements and roles: Such systems should respect users’ arguments while grounding interactions in human rights, caring for people, protecting learner control, and avoiding manipulation or deception.
  • Agreements and roles: Human teachers and experts remain initiators and arbiters of learning conversations, sources of specific knowledge, and caregivers whose roles AI must recognize and respect.

7 Conclusion

Designing social AI for education requires more than adapting existing language models: it requires human-rights commitments, respect for educators, and care for student diversity and development. The paper proposes interdisciplinary partnership and practitioner involvement to shape such systems.

  • Educational social AI should follow human-rights principles, respect teachers’ expertise, and care for students’ diversity and development.
  • The work should unite neural and symbolic AI researchers with pedagogy and learning scientists, while engaging educators to test, critique, and deploy models.
  • The intended result is an online space for educational dialogue and exploration combining human empathy and experience with networked machine learning.
Loading 2306.10063v1…