Exposing critical gap in AI education systems: How machines teach vs how humans learn

AI-powered educational assistants are becoming more widespread in classrooms, but still failing to clearly define how real learning actually happens. A new systematic review suggests the answer lies in a fundamental gap in how artificial intelligence supports human learning.

The study, "Explicit and Implicit Learning Mechanisms in AI Educational Assistants: A Systematic Review," published in AI, analyzes decades of research to show that while AI tools are rapidly evolving, their learning mechanisms remain fragmented, unevenly evaluated, and often poorly understood.

It introduces a structured framework that distinguishes between explicit and implicit learning in AI systems, offering one of the most comprehensive attempts to map how AI educational assistants actually influence user learning across domains.

How AI systems teach: The divide between explicit and implicit learning

The study identifies two primary ways in which AI educational assistants support learning: explicit and implicit mechanisms. This distinction forms the backbone of the research and highlights a critical imbalance in how current systems are designed.

Explicit learning, which dominates the field, involves direct instruction through structured interactions such as feedback, guided questioning, task generation, and step-by-step explanations. According to the review, nearly 79 percent of analyzed systems rely on explicit learning strategies, reflecting a strong emphasis on controlled, instructor-like engagement.

These systems are commonly deployed in programming education, language learning, and technical training, where learners are expected to actively engage with content through quizzes, problem-solving tasks, and structured feedback loops. AI assistants in this category function much like digital tutors, offering clear instructions and corrective feedback to guide users toward specific outcomes.

However, the study warns that this heavy reliance on explicit mechanisms may limit deeper learning. Systems that provide constant guidance can reduce independent reasoning, potentially creating over-reliance on AI-driven instruction rather than fostering critical thinking.

On the other hand, implicit learning remains significantly underutilized, accounting for only 21 percent of the reviewed systems. This form of learning occurs indirectly through interaction, experimentation, and observation rather than deliberate instruction. Users acquire knowledge by engaging with tasks, exploring system outputs, or participating in conversational exchanges without consciously intending to learn.

Examples include activity-based learning environments, conversational agents that adapt dynamically, and systems that allow users to experiment with inputs and observe outcomes. These approaches align more closely with real-world learning processes but remain less developed in current AI implementations.

The study identifies a growing shift in recent years, particularly after 2019, toward integrating implicit learning features through voice assistants, conversational AI, and generative models. However, this transition is still in its early stages, with most systems continuing to prioritize structured, explicit instruction.

From chatbots to intelligent systems: How AI learning tools are built

The research provides a detailed examination of how AI educational assistants are implemented. It identifies four dominant approaches: conversational AI, intelligent systems, AI assistance tools, and standalone AI design tools.

• Conversational AI, including chatbots and virtual assistants, represents one of the most widely used approaches. These systems rely on natural language processing to interact with users through text or voice, enabling real-time responses and personalized guidance. While rule-based chatbots offer reliability through predefined responses, AI-driven conversational agents provide greater flexibility by learning from user interactions.
• Intelligent systems, such as tutoring platforms and recommendation engines, extend this capability by analyzing user behavior and tailoring learning experiences. These systems often incorporate machine learning algorithms to track progress, generate personalized recommendations, and adapt content delivery based on user performance.
• AI assistance tools and design platforms further expand the scope of AI in education by supporting creative and technical workflows. From programming assistants to design ideation tools, these systems enable users to interact with AI in more complex ways, blending learning with real-world application.

The study also traces the evolution of underlying technologies. Early systems relied heavily on rule-based logic and static knowledge bases, limiting their adaptability. Over time, machine learning and neural networks introduced greater flexibility, while recent advancements in large language models have transformed how AI systems generate responses and support learning.

The research identifies persistent limitations in knowledge representation and system architecture. Many systems still rely on static or poorly documented knowledge bases, restricting their ability to adapt to new information or evolving user needs. This lack of transparency and adaptability remains a major barrier to effective AI-driven learning.

Interaction, evaluation, and the limits of AI learning systems

While AI tools are designed to enhance learning, their effectiveness often depends on how users interact with them and how outcomes are measured. The research shows that written interaction dominates AI learning environments, accounting for 45 percent of interaction types, followed by visual interfaces at 34 percent. Action-based interactions and voice-based systems remain less common, highlighting a gap in multimodal learning experiences.

This imbalance suggests that many AI systems still rely on traditional text-based interfaces, limiting their ability to engage users in more dynamic and immersive ways. Voice assistants and interactive simulations, which could support more natural learning processes, remain underdeveloped in comparison.

Evaluation practices present another critical challenge. Most studies rely on short-term experimental methods, often involving limited sample sizes and controlled environments. While these approaches provide initial insights, they fail to capture long-term learning outcomes or real-world impact.

The study calls for more robust evaluation frameworks, including longitudinal studies that track user progress over extended periods. It also emphasizes the need for more diverse participant groups to ensure findings are generalizable across different learning contexts.

User experience issues further complicate the effectiveness of AI systems. The research identifies common challenges such as user dissatisfaction with chatbot responses, difficulty understanding AI-generated suggestions, and cognitive overload in complex systems. These issues highlight the importance of user-centered design and iterative testing in developing effective AI tools.

Technological limitations also persist, particularly in handling complex tasks, supporting multi-user interactions, and integrating advanced features. Many systems struggle to balance flexibility with reliability, leading to inconsistent performance and reduced user trust.

Generative AI and the future of learning systems

The integration of generative AI and large language models marks a significant turning point in the evolution of AI educational assistants. Unlike traditional systems, these models generate dynamic responses based on large-scale training data, enabling more adaptive and conversational interactions.

The study identifies several emerging approaches, including direct integration of generative APIs, hybrid systems that combine structured knowledge bases with generative models, and multimodal platforms that incorporate text, images, and interactive inputs.

These advancements have the potential to bridge the gap between explicit and implicit learning. Generative systems can provide detailed explanations and structured guidance while also supporting exploratory, interaction-driven learning. This dual capability represents a major step forward in creating more holistic learning environments.

However, the research also highlights new risks associated with generative AI, including inaccuracies, hallucinated responses, and reduced critical thinking among users who may overly rely on AI-generated outputs. These challenges are particularly concerning in high-stakes domains such as healthcare and security, where incorrect information can have serious consequences.

To address these risks, robust evaluation metrics that go beyond accuracy and usability are critical. Future systems must incorporate mechanisms for verifying information, ensuring transparency, and maintaining user engagement without compromising learning quality.