Source-linked AI summary
Promises and challenges of generative artificial intelligence for human learning
Lixiang Yan, Samuel Greiff, Ziwen Teuber, Dragan Gašević
TL;DR
GenAI could transform human learning, but its benefits must be weighed against risks to learning quality, equity, ethics, and assessment. This Perspective synthesizes learning sciences, educational technology, and human-computer interaction to examine how GenAI can support learning while preserving human cognitive and creative capacities.
Problem
Evidence is needed to understand how GenAI can improve learning while addressing model imperfections, ethical dilemmas, assessment disruption, and risks to human agency, critical thinking, and creativity.
Method
The Perspective integrates insights from learning sciences, educational technology, and human-computer interaction to examine GenAI's promises and challenges across human learning.
Results
GenAI can provide personalised support, generate diverse learning resources, deliver timely feedback, support multimedia learning, and enable conversational assessments, but performance with GenAI assistance may not reflect independent learning.
Takeaways & Limitations
Humanity must learn with and about GenAI by cultivating AI literacy, refining research methods, and designing learning and assessment practices that preserve learner agency and genuine skill development.
Takeaways & Limitations
Opaque or biased models can undermine transparency, fairness, and accuracy, while adversarial prompts may bypass alignment measures and facilitate cheating or biased content.
Abstract
from arXiv · showhide
Generative artificial intelligence (GenAI) holds the potential to transform the delivery, cultivation, and evaluation of human learning. This Perspective examines the integration of GenAI as a tool for human learning, addressing its promises and challenges from a holistic viewpoint that integrates insights from learning sciences, educational technology, and human-computer interaction. GenAI promises to enhance learning experiences by scaling personalised support, diversifying learning materials, enabling timely feedback, and innovating assessment methods. However, it also presents critical issues such as model imperfections, ethical dilemmas, and the disruption of traditional assessments. Cultivating AI literacy and adaptive skills is imperative for facilitating informed engagement with GenAI technologies. Rigorous research across learning contexts is essential to evaluate GenAI's impact on human cognition, metacognition, and creativity. Humanity must learn with and about GenAI, ensuring it becomes a powerful ally in the pursuit of knowledge and innovation, rather than a crutch that undermines our intellectual abilities.
1 Main
GenAI offers new ways to support human learning, but its benefits must be balanced against ethical, practical, and human-centred concerns. This perspective examines how GenAI may transform learning while guiding responsible future research and design.
- GenAI can automate learning tasks, provide feedback, outperform average students in reflective writing, create dynamic resources, and support multimedia learning.
- Unequal access to GenAI may exacerbate learning inequalities, while overreliance may weaken learner agency, critical thinking, and creativity.
- The paper integrates learning and instruction theories to examine GenAI’s effects on teaching and learning from a human-centred perspective.
- It also develops a future research agenda for studying human–AI interaction and GenAI adoption as a learning tool.
2 Promises
GenAI promises to transform human learning through scalable personalised support, diversified resources, timely feedback, and adaptive assessment. Realising these benefits requires human oversight, validation, and further research into learning outcomes and assessment validity.
- Overview: GenAI’s learning promises depend on how it interacts with learners and educators across support, resource creation, feedback, and assessment.The overview frames these applications as opportunities whose implementation requires attention to the relevant learning impacts and components.
- 2.1 Learning Support: GenAI can provide personalised, adaptive tutoring at scale across subjects and languages, potentially broadening access to high-quality learning support.Compared with conventional rule-based intelligent tutoring systems, it can generate more naturalistic feedback, questioning, and conversational engagement with less prior knowledge engineering.
- 2.1 Learning Support: Evidence on GenAI-supported learning engagement, agency, and performance remains mixed, with limited evidence on short- and long-term learning outcomes.The passage notes that learning gains may disappear after GenAI support is removed.
- 2.2 Learning Resource: GenAI can help educators co-create accessible instructional materials, including explanations, examples, quizzes, interactive activities, and multimedia content.GPT-4 performs better for lower-level biology questions than for higher-level apply and create questions, leaving educators responsible for accuracy, relevance, and pedagogical soundness.
- 2.3 Learning Feedback: GenAI-generated feedback can be timely, personalised, multimodal, and beneficial across essay writing, programming, and second-language writing contexts.Audio and video modalities may improve engagement and perceived personalisation, but computational demands could widen inequalities in learning opportunities.
- 2.4 Learning Assessment: GenAI can support automated and authentic assessment through multi-agent grading, cognitive assessment, virtual simulations, and multimodal professional scenarios.These systems can model contexts such as medical diagnosis, driving, programming, laboratory safety, virtual internships, and healthcare simulations.
3 Challenges
GenAI introduces model imperfections, ethical dilemmas, and assessment disruptions that complicate the evaluation and support of human learning. These challenges require validation, human-centred safeguards, and assessment strategies that account for human–AI collaboration.
- 3.1 GenAI’s Imperfections: GenAI hallucinations can become more frequent with complex, specific queries, potentially introducing inaccuracies into learning content, feedback, and assessments.The lack of transparency in model decision-making further complicates identifying when and why errors occur.
- 3.2 Ethical Dilemmas: 92% of GenAI learning-support tools are transparent only to AI experts, limiting educators’ and students’ understanding of their functions and flaws.The passage attributes this gap primarily to the absence of human-in-the-loop involvement in development and evaluation.
- 3.2 Ethical Dilemmas: Personalisation requires learner data, but unclear consent, inadequate protection, and possible data breaches create substantial privacy risks.The passage notes that private information can be difficult to remove once incorporated into large proprietary models.
- 3.2 Ethical Dilemmas: English-dominant and unevenly accessible GenAI systems may disadvantage non-Western populations and intensify inequalities in learning opportunities.The concern is framed around unequal language representation, accessibility, global applicability, and fairness.
- 3.2 Ethical Dilemmas: Biased, underperforming, or adversarially manipulated models can facilitate cheating, expose learners to harmful content, and compromise safety and inclusivity.Proposed responses include balanced sampling, cautionary labelling, alignment, and stronger regulation, although opaque models make fairness difficult to ensure.
- 3.3 Disruption of Assessment: AI-generated responses challenge product-focused assessments by making it difficult to distinguish learners’ work from generated output.AI can also imitate learning processes and produce digital traces that complicate learning analytics.
- 3.3 Disruption of Assessment: GenAI assistance can improve task performance while sharply reducing performance when removed, creating an illusion of learning without essential skill development.This performance paradox raises questions about whether assessments measure learners’ abilities or AI-supported outputs.
- 3.3 Disruption of Assessment: Assessment priorities may need to differ by educational stage, emphasizing human cognition and metacognition in K–12 and hybrid cognition in higher education.This shift reflects the collaborative nature of learning when AI is integrated into educational settings.
4 Needs
Effective integration of GenAI into human learning requires AI literacy, evidence-based decision-making, and methodological rigour. These needs support responsible use while protecting cognitive, metacognitive, and creative development.
- 4.1 AI Literacy: AI literacy should extend across learners, educators, policymakers, and administrators to support effective, responsible, and ethical GenAI use.It includes understanding how AI systems work, their impacts, ethical considerations, and limitations.
- 4.1 AI Literacy: Users may prefer comprehensive, fluent AI-generated content despite inaccuracies, making recognition of hallucinations and model limitations essential.The passage notes that hallucination remains difficult to address at the foundational-model level.
- 4.2 Evidence-Based Decision Making: GenAI can improve information processing and retrieval, but fluency bias may cause learners to overestimate their understanding.Reliance on GenAI for creative and problem-solving tasks may also weaken critical skills and foster dependency.
- 4.2 Evidence-Based Decision Making: Researchers, practitioners, and policymakers should collaborate to generate evidence that aligns GenAI deployment with learning goals and cognitive development.The proposed partnership is intended to help GenAI enhance rather than replace human cognitive, metacognitive, and creative processes.
- 4.3 Methodological Rigour: Methodological rigour is necessary because apparently exceptional GenAI performance can be overstated by dataset contamination, weak accuracy assessment, and ambiguous verification.A widely publicized GPT-4 study claiming perfect scores in an MIT curriculum was later retracted for such concerns.
- 4.3 Methodological Rigour: Evidence-quality standards are needed to produce reliable, valid, and generalisable findings about GenAI’s effects on learning processes, outcomes, and experiences.The paper points to risk-of-bias tools in medicine as an example for developing appraisal frameworks.
5 Conclusion and Future Directions
GenAI is expected to become deeply embedded in learning, requiring people to learn both with and about these systems while preserving cognitive and metacognitive autonomy. Realising this potential depends on human-centred integration, educator preparation, ethical accountability, and research on effects on cognition, metacognition, and creativity.
- GenAI may become integral to society, transforming how people learn, work, and live.
- Learning with and about GenAI should continue alongside development of critical thinking, problem-solving, self-regulation, and reflective thinking skills.These skills are described as crucial for maintaining cognitive and metacognitive autonomy as AI becomes embedded in daily life.
- Research should examine how GenAI affects human cognition, metacognition, and creativity while identifying indicators of these processes and strategies that prevent over-reliance.The paper highlights learner agency and continued human responsibility for critical thinking and problem-solving as key research concerns.
- Educators will need AI literacy, new pedagogical paradigms, and support systems to use GenAI while preserving creativity, critical thinking, social interaction, and human mentorship.Institutions are urged to address techno-stress and workload burdens associated with adopting new technologies.
- Policymakers and technology companies must address accountability, ethical guidelines, equality, and inclusivity in educational AI tools.
- Human-centred learning theories should guide GenAI integration so the technology supports human cognitive capacities and becomes an ally rather than a crutch.The paper calls for coordinated effort among researchers, policymakers, technology companies, and educators.