Recent research suggests AI systems can support learning through questioning, scaffolding, feedback, and knowledge organization (Kim & Baylor, 2006; Rogers & Carbonaro, 2025, Zhou et al., 2025). During testing, NotebookLM demonstrated several of these characteristics, particularly through source-grounded dialogue and adaptive support.

From an EVA perspective, an important limitation is that effective learning partnerships typically involve more than instructional support. Human learning partners build relationships over time, remember previous interactions, and adapt to evolving learning needs (Kim & Baylor, 2006; Zhou et al., 2025). While NotebookLM demonstrated strong support within individual sessions, it lacks persistent memory, evolving learner models, and long-term relational continuity. These limitations suggest that AI learning partners remain an emerging rather than fully realized category of educational technology.

Google. (2026).  Cognitive Strengthening vs. Cognitive Substitution [AI-generated code]. Gemini. https://gemini.google.com.

García-Barrios (2025) argues that AI systems can function either as epistemic partners, helping learners develop understanding, or cognitive crutches, reducing the need for independent thinking.

From an investment perspective, this distinction represents both an opportunity and a risk. Research suggests that these concerns should be taken seriously. Stadler et al. (2024) found that AI support reduced mental effort during inquiry tasks, while Zhai et al., (2024) identified growing concerns regarding overreliance and critical thinking. Fan et al. (2025) describe this risk as “metacognitive laziness” (p. 489), where learners complete tasks successfully while engaging in fewer self-regulated learning processes.

However, earlier research on pedagogical agents found that thoughtfully designed learning companions can improve learner engagement and support knowledge construction (Kim & Baylor, 2006). More recently, Gerlich (2025) found that intentional prompting and interaction design can improve critical reasoning when working with generative AI systems.