Does discovery-based instruction enhance learning?

Abstract / Introduction
The paper is a 2 tier meta-analysis of discovery learning approaches in education. The first examines the effects of unassisted discovery learning versus explicit instruction. The second tier examines the effects of enhanced/assisted discovery learning versus other forms of learning.

Details
Alfieri et al. present several counter arguments to discovery learning from previous studies stating that learners might have difficulty arriving at the proper strategy of holding all other variables constant while manipulating only one (in regards to teaching CVS), insisting that there are times when explicit or directive guidance is optimal. Citing studies that present the case for a greater number of cognitive functions when engaging learning by discovery, Alfieri writes that to select, organize and integrate high-level information in a task appropriate way is quite demanding of learners.

The continued meta-analyses of enhanced discovery learning however, yielded superior results compared to that of other learning methods. The outcomes did not yield expected results. First, the adolescent age group was shown to benefit the least form unassisted discovery conditions, as opposed to children. Adults seemed to benefit from both, enhanced-discovery and unassisted discovery tasks more so than children. Probably due to domain specific knowledge and general lack of motivation. Organizing guidance to facilitate discovery requires sensitivity to the learner's zone of proximal development if it is to be maximally useful.

Synthesis

Epistemologically speaking, unassisted discovery learning might be considered epistemic luck if it cannot be reliably replicated nor if it is transferable to other fundamental truths. If the learner has no justification in the belief that learning has occurred, it is difficult to say that knowledge has been acquired at all. As such, we might consider this article from two different contexts.

First, in my personal experience, the implementation of inquiry learning at my school is designed to have teams work on topics that are usually thematically all under the same umbrella, either by finding a wicked problem and designing a solution, or by being given a topic and... designing a solution. In whichever case, learning outcomes are measured by the viability of the solution, or according to some teachers- merely by the fact that the student "experienced" the process, not the actual learning processes in and of itself. The unassisted discovery aspect of these inquiry learning projects varies mainly due to lukewarm educator attitudes towards the initiative itself. Teachers who are willing to put in the extra legwork to create enhanced discovery learning experiences often must do so on their own time, fueled entirely by their own emotional labor. Thus there is great variability in learning outcomes, propelled mainly by motivated individuals who, unsurprisingly produce work with, or without the help of naive or unmotivated teammates.

Second, in considering the role of AI within the purview of such learning methodologies, we quickly spot the harm that having unfettered access to AI during the process of learning results in. In particular, without the process of scaffolding, active engagement and introspective-reflection with AI tools, learners slip into cognitive offloading simply because the cognitive requirement to select, organize and integrate high-level information is too demanding (Aliferi et al., 2011). As Alifieri et al., have stated, unassisted tasks, tasks requiring invention and tasks involving collaboration with a naive peer were in fact found to be equally as detrimental to learning (2011).

Does this mean that AI has no place in learning? Perhaps a better question would be 'What kind of constructs can we build around learner cognition to drive motivation and/or curiosity?' If we consider scaffolds for AI through the SDT framework, we can see that autonomy, competency and relatedness are coaxial aspects of building intrinsic and extrinsic motivation (Self-Determination Theory, n.d.)

Returning to the original discussion prompt, which was how can we design constructivist e-learning activities that require perseverance if AI encourages the path of least resistance, maybe as educators, we must first consider the constructs (environment, task-design, assessment and curriculum development) of learning are optimized for building intrinsic and extrinsic motivation. In doing so, we can consider ways of integrating LLMs (not only GenAI- see Google's NotebookLM, Anthropic's Cowork) as tools to boost motivation and learning.

How do you imagine learning in schools to inculcate intellectual virtue though GenAI tends to encourage cognitive offloading and non-frictive learning processes?