In deductive domains, three metacognitive knowledge types in ascending order are declarative, procedural, and conditional learning. This work leverages Deep Reinforcement Learning (DRL) in providing adaptive metacognitive interventions to bridge the gap between the three knowledge types and prepare students for future learning across Intelligent Tutoring Systems (ITSs). Students received these interventions that taught how and when to use a backward-chaining (BC) strategy on a logic tutor that supports a default forward-chaining strategy. Six weeks later, we trained students on a probability tutor that only supports BC without interventions. Our results show that on both ITSs, DRL bridged the metacognitive knowledge gap between students and significantly improved their learning performance over their control peers. Furthermore, the DRL policy adapted to the metacognitive development on the logic tutor across declarative, procedural, and conditional students, causing their strategic decisions to be more autonomous.

In this work, we investigate how two factors, metacognitiveskills and motivation, would impact student learning acrossdomains. More specifically, our primary goal is to identify thecritical, yet robust, interaction patterns of these two factors thatwould contribute to students’ performance in learning logicfirst and then their performance on a subsequent new domain,probability. We are concerned with two types of metacognitiveskills: strategy-awareness and time-awareness, that is, whichproblem-solving strategy to use and when to use it. Our datawere collected from 495 participants across three consecutivesemesters, and our results show that the only students who con-sistently outperform their peers across both domains are thosewho are not only highly motivated but also strategy-aware andtime-aware.

In this study we explored the impact of student versus tutorpedagogical decision-making on learning. More specifically,we examined what would happen if we let students decidehow to handle the next task: to view it as a worked exam-ple or to solve it as a problem solving. We examined this im-pact at two levels of task granularity: problem vs. step. This2 × 2 study was conducted on an existing Intelligent TutoringSystem (ITS) called Pyrenees. 279 students were randomlyassigned to four conditions and the domain content and re-quired steps were strictly controlled to be equivalent acrossfour conditions: all students used the same system, followedthe same general procedure, studied the same training materi-als, and worked through the same training problems. The onlysubstantive differences among the four conditions were deci-sion agency {Student vs. Tutor} and granularity {Problem vs.Step}. That is: who decided to present an example or to solvea problem; and was the decision made problem-by-problem orstep-by-step? Our results showed that there was a significantinteraction effect between decision agency and granularity onstudent learning and a significant main effect of granularity ontime on training. That is, step level decisions can be more ef-fective than problem level decisions but the students were morelikely to make effective pedagogical decisions at problem levelthan step level. In general, on both problem and step levels, thestudents were significantly more likely to decide to do problemsolving rather than study it as a worked example.

Based on strategy-awareness (knowing which problem-solving strategy to use) and time-awareness (knowing when to use it), students are categorized into Rote (neither type of awareness), Dabbler (strategy-aware only) or Selective (both types of awareness). It was shown that Selective is often significantly more prepared for future learning than Rote and Dabbler (Abdelshiheed et al., 2020). In this work, we explore the impact of explicit strategy instruction on Rote and Dabbler students across two domains: logic and probability. During the logic instruction, our logic tutor handles both Forward- Chaining (FC) and Backward-Chaining (BC) strategies, with FC being the default; the Experimental condition is taught how to use BC via worked examples and when to use it via prompts. Six weeks later, all students are trained on a probability tutor that supports BC only. Our results show that Experimental significantly outperforms Control in both domains, and Experimental Rote catches up with Selective.

Deductive domains are typical of many cognitive skills in that no single problem-solving strategy is always optimal for solving all problems. It was shown that students who know how and when to use each strategy (StrTime) outperformed those who know neither and stick to the default strategy (Default). In this work, students were trained on a logic tutor that supports a default forward-chaining and a backward-chaining (BC) strategy, then a probability tutor that only supports BC. We investigated three types of interventions on teaching the Default students how and when to use which strategy on the logic tutor: Example, Nudge and Presented. Meanwhile, StrTime students received no interventions. Overall, our results show that Nudge outperformed their Default peers and caught up with StrTime on both tutors.

In this paper, we explore the impact of two types of instructional interventions, worked examples and problem solving, at two levels of granularity: problems and steps. This study drew on an existing Intelligent Tutoring System (ITS) for Probability called Pyrenees and involved 266 students who were randomly assigned to five conditions. All students experienced the same procedure, studied the same training problems in the same order, and used the same ITS. The conditions differed only in how the training problems were presented. Our results show that when the domain content and required steps are strictly equivalent, different granularities of pedagogical decisions can significantly impact students’ time on task. More specifically, the fine-grained step level decisions can have a stronger pedagogical impact than the problem-level ones.