By Atefeh Ferdosipour
We have become so profoundly immersed in the processing power and rapid response rates of generative artificial intelligence that we have overlooked a vital principle: AI generates responses, but it is the human brain that must perceive them.
To put it more simply, the ultimate consumer of artificial intelligence (AI) is the human being, who is not a passive or static entity but actively participates in receiving and processing information. For this very reason, when external data is received by the human mind and brain as raw materials, it does not remain in its initial state. Instead, it undergoes transformation, leading to distinct and varied outcomes.
Perhaps the deepest contemporary criticism of AI tools, particularly within the learning and education industry, is that while these tools have achieved and will continue to achieve stunning growth in speed and information density, they are not sufficiently human-centered. They remain fundamentally data-driven, built upon complex mathematical laws and numerical computations.
What can bridge this gap, mitigate future risks, and reduce growing anxieties? In my view as a psychologist committed to human-centered AI, the answer lies in applying the principles of psychology and learning sciences to the redesign of these tools. These are principles grounded in experience, rigorous research, and verifiability, rather than personal preference.
I have written extensively about the necessity of this critical shift in various articles. This includes a previous piece where I explored Vygotsky’s theory in the era of artificial intelligence. Due to the profound significance of this topic, this article focuses on the practical application of Gestalt theory in the age of AI.
In my previous article, we discussed how AI requires intentional friction to foster deep learning—a concept rooted in Vygotsky’s theory that prevents superficial and passive learning. However, for this friction to lead to growth and keep the learner within the Zone of Proximal Development (ZPD), artificial intelligence must not act merely as a text generator. It must become a perceptual architect and scaffolder.
‘Holism’: A Search for Structure
The core challenge lies here: Even if AI executes scaffolding correctly, it may still present information without a logical and visual structure. This is precisely where the laws of Gestalt psychology enter the field. Gestalt laws teach us how the human mind instinctively searches for order, patterns, and the whole.
This emphasis on holism within the learner’s perceptual organization was a fundamental critique of schools of thought that practiced extreme reductionism and overemphasized isolated details, such as structuralism and behaviorism. Consequently, Gestalt psychologists maintained that we perceive the world through integrated wholes, rather than through fragmented, analyzed, and reduced parts.
For instance, to comprehend the concept of a tree, we perceive the global image of the tree rather than focusing initially on the individual branches, trunk, leaves, and details. Therefore, learning takes place within the realm of perceiving global relationships, occurring through sudden flashes and insights.
With this understanding of learning, it becomes evident that this school of thought views learning as a deep, meaningful process rather than something superficial and transient.
Deep Learning Within the Gestalt Viewpoint
Given what has been discussed, Gestalt is not limited solely to visual perception. Through these foundational concepts and laws, it directly addresses the processes of deep learning and problem-solving. From this perspective, instead of being the result of a gradual accumulation of responses and habits, learning is the outcome of a cognitive reorganization.
Productive Thinking Versus Reproductive Thinking
This perspective maintains that productive thinking differs fundamentally from reproductive thinking. Reproductive thinking is based on repeating and executing previously learned patterns. Conversely, productive thinking means understanding the internal structure of the problem and reorganizing its elements to reach entirely new solutions. From this viewpoint, deep learning occurs when a person can comprehend the underlying relationships between components, rather than merely reproducing prefabricated answers.
Problem-Solving as Restructuring Cognitive Architecture
The process of problem-solving is defined in Gestalt theory as a form of restructuring the cognitive architecture. Facing a problem typically creates a type of cognitive imbalance or tension. To reduce this tension, the mind strives to reorganize the components of the problem to reach a stable, coherent, and balanced perceptual state, which can be understood as a return to perceptual equilibrium.
The Zeigarnik Effect and Deep Learning
The concept of the Zeigarnik Effect states that incomplete tasks remain in the mind much longer than completed ones. This phenomenon demonstrates the mind’s tendency to retain unresolved cognitive tensions and pursue them until a complete structure is achieved.
Gestaltists rely on this concept to explain the motivation required to resolve cognitive imbalances and achieve deep learning. Incomplete content exerts a motivational force on the learner because it generates a natural tension and disequilibrium. The memory naturally strives to complete the information. When the data is successfully completed through cognitive effort and trial, equilibrium is restored, marking the exact moment deep learning occurs and the puzzle is finalized.
Memory & Deep Learning
At the memory level, Gestalt offers a perspective distinct from traditional approaches. Within this framework, perceptual experiences persist as memory traces. However, these traces do not operate independently or in isolation; instead, they are integrated and organized within a comprehensive trace system. Holism governs here as well: The more coherent this trace system is, the easier and more structured the recall of information becomes.
Accordingly, forgetting does not imply the complete deletion of information. Instead, it is the result of a memory trace merging or dissolving into broader cognitive structures, thereby losing its independent identity.
The learner’s mind is constantly and actively reorganizing its informational structure within memory. As previously stated, imbalance creates a motivational effect, prompting the learner to expend effort in reorganizing coherent information within the memory system to resolve cognitive disequilibrium.
The Function of Deep Learning
Within this perspective, the transposition or transfer of learning holds special significance. Gestaltists believed that humans do not merely learn specific responses; rather, they comprehend the relationships between elements. Therefore, if the structure of a situation is understood, it can be easily generalized to new situations.
By understanding a law, a formula, or the internal relationships of a subject, one can easily bypass the need for re-learning similar topics, applying that core underlying principle universally. Consequently, the most important outcome of deep learning is the transfer of learning and the utilization of underlying principles in similar future situations, freeing the learner from the burden of repetitive re-learning.
Applied Principles of Gestalt in Designing AI Tools for Deep Learning
To move from theory to practice, we can examine how core Gestalt mechanisms can be leveraged through a practical lens, transforming AI platforms into highly effective,
human-centered learning environments. These principles are more relevant than ever today, sitting at the very heart of designing artificial intelligence tools. In this section, we examine each Gestalt principle through a practical lens, supported by tangible examples.
1. The Whole Before the Parts
The human mind requires an initial global picture to understand details. When components are presented without a framework, the mind becomes confused. This confusion does not stem from a weakness in the learner, but rather from the lack of an established structure to anchor the incoming information.
For example, imagine a student asking ChatGPT, “What is machine learning?” The system might immediately respond with technical definitions, types of algorithms, and specialized jargon.
Conversely, if the system initially stated, “Machine learning means teaching a computer to learn from examples, just like we learn from experience,” the learner’s mind would possess a framework, making the subsequent details meaningful. A Gestalt-based AI system always begins with a roadmap, not with isolated details.
2. Conceptual Structure Over Scattered Information
AI tools frequently generate vast amounts of information, but the quantity of data does not equate to the quality of learning. What the mind truly requires is the relationships between information, not just the isolated data points.
For example, if a learner inquires about the Industrial Revolution, a typical system might provide a list of dates, names, and events. A structure-oriented system, however, would state, “The Industrial Revolution triggered three massive shifts: in how things were produced, where people lived, and how society interacted. Let us explore each one.” This narrative provides structure rather than a mere data dump.
3. Insightful Learning Over Answer Delivery
One of the greatest paradoxes of AI in education is that a tool designed to aid learning can actually hinder it by delivering answers too quickly. The “Aha!” moment—the very insight Gestaltists emphasize—only occurs after a period of mental engagement. If the answer is provided prematurely, that cognitive milestone is lost.
For example, instead of ChatGPT declaring, “The answer to problem X is this,” it could ask, “Before we proceed, why do you think this happened?” This simple question forces the learner’s mind into active engagement, which is the prerequisite for insight.
4. Active Cognitive Participation Versus Passive Reception
The human mind is not an empty vessel waiting to be filled with information. It is an active processor that requires engagement to function correctly. AI tools that place the learner in a position of pure passive reception run counter to the natural operation of human cognition.
For example, a Gestalt-based system can ask after each explanation, “How does this compare to what you already knew?” or “Where do you think this principle might not apply?” These questions lift the learner out of a passive stance.
5. Meaningful Organization of Information
The Law of Pragnanz states that the mind seeks order. If information is presented chaotically, the mind wastes its energy trying to find order rather than focusing on learning. A well-designed AI tool performs this structural organization in advance.
For example, when a system presents information in logical groups, with clear headings, and explicitly demonstrates the connections between them, the learner’s mind is liberated to focus entirely on comprehension rather than categorization.
6. Keeping Curiosity Alive
The Zeigarnik Effect teaches us that the human mind remains engaged with incomplete tasks. This implies that an AI tool delivering everything perfectly and completely can inadvertently stifle the learner’s curiosity.
For example, instead of an exhaustive answer, the system can leave a gap: “This topic has two conflicting viewpoints. We have explored one. Would you like to see the opposing view and form your own judgment?” This open loop keeps the mind actively engaged.
7. Human Cognition as a Design Model
Perhaps the most vital message Gestalt holds for AI developers is this: Before asking “What can this system do?” ask “How does the human mind work?” Truly human-centered AI is not merely designed for humans; it is engineered based on the laws governing the human mind. Gestalt psychology provides a roadmap for this architecture—a map derived from decades of empirical observation and experience rather than pure abstract theorizing.
8. Unified Perception Versus Isolated Elements
Learners process information through interconnected networks rather than disconnected particles. Tools that present every concept in an isolated bubble conflict with this cognitive reality.
For example, when a system links a new concept to what the learner has previously mastered—stating, “This is similar to what we discussed earlier, but with one critical distinction”—deeper, more resilient learning occurs.
9. Balancing Speed & Depth
AI is unrivaled in its speed. However, speed and depth do not always align. Sometimes, slowing down means learning deeper. A Gestalt-oriented system understands when to introduce a pause.
For example, following a complex explanation, the system might suggest, “Pause for a moment before moving on. If you had to explain this concept to a friend in one simple sentence, what would you say?” This simple prompt stimulates deep cognitive processing.
10. Meaningful Learning & Intrinsic Motivation
The final and perhaps most critical principle is that when learning is meaningful—when the learner feels they have comprehended a whole rather than just receiving information—intrinsic motivation emerges. This motivation is neither injected from the outside nor driven by external rewards; it arises from the experience of understanding itself. Gestalt reminds us that humans are naturally meaning-seeking creatures. An artificial intelligence that respects this need in its design is not only more effective, but fundamentally more human.
Conclusion: Why Applying Theories Like Gestalt Is Highly Effective in AI Redesign
The established frameworks of learning sciences and psychology serve as the ultimate blueprints for educational technology stakeholders and AI developers. Because these cognitive models have withstood empirical trial and rest on solid scientific foundations, they offer a highly reliable guide for designing diverse AI interactions.
Ultimately, if Vygotsky’s principles guide us on when to introduce challenge and how to scaffold the learner, Gestalt theory ensures that the final destination of that challenge is a sudden, meaningful cognitive reorganization. AI must evolve beyond a simple information dispenser; it must design structured environments that respect human perceptual organization, guiding the learner seamlessly toward a comprehensive and lasting understanding of the whole.
Given the brevity required for this article, many of these foundational concepts have been highly condensed. A much more comprehensive exploration of these psychological mechanics, empirical evidence, and their advanced architectural integration into human-centered AI
frameworks will be featured in my upcoming book, currently in development.
References
Ferdosipour, A. (2026). Why AI needs Vygotsky: The Case for AI-Based Intentional Friction. Learning Solutions Magazine.
Hergenhahn, B. R., & Olson, M. H. (2013). An introduction to theories of learning (9th ed.). Routledge.
Koffka, K. (1935). Principles of Gestalt psychology. Harcourt, Brace & World.
Köhler, W. (1925). The mentality of apes. Harcourt, Brace & World.
Norman, D. A. (2013). The design of everyday things (Revised and expanded ed.). Basic Books.
Schunk, D. H. (2020). Learning theories: An educational perspective (8th ed.). Pearson.
Wertheimer, M. (1945). Productive thinking. Harper & Brothers. Zhang, J., & Lu, Y. (2021). Human-centered artificial intelligence in education: A cognitive architecture approach. Computers and Education: Artificial Intelligence, 2, 100027.
Image credits: The author created the top image and content diagram using Google Gemini and are based on the author’s own original concepts and prompts.
