Technical Principles of Generative AI and an Examination of Its Educational Applicability

Fengchun Miao’s “Technical Principles of Generative AI and an Examination of Its Educational Applicability” appeared in Modern Educational Technology, issue 11 of 2023, as the first in a series interpreting UNESCO’s Guidance for generative AI in education and research. The following introduces the abstract’s analysis of technical principles and educational suitability.

Technical principles and educational potential

The abstract defines generative AI as technology that processes across symbolic representations of human thought and generates new content. The author argues that limitations in understanding semantics and the real world constrained its potential to transform education at that time.

Foundation models, values and language bias

The author identifies security concerns for digital infrastructure associated with monopolistic foundation models, alongside risks of projecting Western values and linguistic and cultural bias. He therefore advocates independently controllable foundation models and models more suitable for education as a safety baseline for educational applications.

Minors, chat access and educational content

The abstract notes the predominant use of internet data generated by adults in training and questions the suitability of chat-based content services for minors, recommending a lower age limit for independent conversations. It also questions using paid content generated from broad, heterogeneous training data as the main source of educational content, arguing that this can conflict with both the technology’s design purpose and teaching aims. These are the paper’s analyses and recommendations.

Knowledge construction and learner agency

Regarding the educational process, the author compares automated outputs to “ready-made meals” of content and warns of an “anti-intellectual” risk: weakening learners’ knowledge construction and skill acquisition through unpacking content. Learner agency is central to this argument.

Five teaching-application contexts

The abstract proposes designing applications around learners and subjects in five main contexts:

• Curriculum-content production.
• Learning foundational knowledge and basic skills.
• Special learning needs.
• Higher-order thinking.
• Inquiry and practical exploration.

Keywords

The original keywords are:

• Generative artificial intelligence
• ChatGPT
• Symbolic representations of human thought
• Values and language bias
• Educational applicability
• Foundation models

Sources and reading

The Cankaowang article-information page identifies the journal and author and is dated 28 November 2023; the original site archive is dated 21 November. The dates are retained separately. This page presents the abstract; consult the journal for the full paper. Background: UNESCO guidance; related framework: iSTAR human–machine synergy; learning resources: special and inclusive education resources.