Answers are grounded in this paper and cite their sources.
Generative artificial intelligence (AI) is rapidly entering adult and continuing education, where it can explain, summarise, draft and evaluate on the learner's behalf. The quality of work produced with AI assistance may therefore rise while the learner's own understanding, judgement and critical thinking remain unchanged or decline. This conceptual article examines how AI-assisted adult learning can preserve learner autonomy, independent judgement and critical thinking while still benefiting from generative tools. It builds on Natseva's distinction between functional cognitive offloading and higher-order cognitive outsourcing and integrates it with research on learner autonomy, self-regulated learning, metacognition, automation bias, critical-thinking instruction and the learning–performance distinction, drawing on a purposive, non-systematic literature selection that is reported transparently. Because most empirical studies of generative AI in education involve school or university students, the article distinguishes evidence by learner population and discusses its transferability to adult education. The original contribution is the AI-Assisted Autonomy and Critical Thinking Assessment (AACTA) framework, which specifies six assessable dimensions: prior independent reasoning, purposeful AI engagement, appraisal of AI output, cross-source contrast, decision ownership, and unaided demonstration and transfer. Each dimension is operationalised as a scoring rubric with observable indicators at three levels, from displaced to autonomous, together with a procedure for training two assessors and checking their agreement. AACTA is compared with existing measures of learner autonomy, self-regulated learning and critical thinking. The framework is accompanied by a dual-outcome principle, which requires AI-assisted performance and unaided learning to be measured separately, and by a provisional assistance-gap index. Five testable hypotheses are formulated, and a three-arm randomised study with a preceding pilot and multi-context replication is proposed as future work. The framework has not been empirically validated, and no data were collected for this article.
No takes yet. Share an insight, caveat, or question.
Zheni Markova (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: