Exploring the creation of computer game training for attention disorders by using AI computer brains from the end of 2025 to create Famicom and Super Famicom-level pixel art components
In 2024, I was able to create simple games with Claude, similar to those developed by Steve Wozniak and Steve Jobs during the Atari era.
I was able to create games like Breakout, pinball, and Tetris with Claude, but they were at an 8-bit level, similar to the Game Boy or Famicom.
I am exploring AI production for games on the level of "Puyo Puyo by the defunct Compile, Final Fantasy 5 and 6 or Chrono Trigger which were Square at the time, and Dragon Quest 5 which was Enix at the time."
Beautiful pixel art games from about 30 to 35 years ago are still loved all over the world.
It is precisely because they were created on computers with 16-bit limitations and limited CPUs that artistic pixel graphic design games are so highly regarded.
A 2D remastered version of Dragon Quest 3 was released in November 2024, and a 2D remaster of Dragon Quest 1 & 2 was released in October 2025, but they are not very highly rated.
Just as Square Enix released pixel art remasters of Final Fantasy 1-6 for the Famicom and Super Famicom, games from 30-40 years ago are still loved today.
I am exploring whether it is possible to develop games with the sophistication of 30-40 years ago using AI as computer training for higher brain dysfunction or cognitive dysfunction.
As of November 2025, I am exploring whether it is possible to easily develop Super Famicom-level 16-bit games by utilizing the latest AI technology, including Google AI Studio and Gemini.
Games like "Puyo Puyo" from the former Compile, and "Final Fantasy IV, V, VI, and Chrono Trigger" from the former Square were artistic as pixel art games.
It is quite difficult to develop so-called 16-bit machine games that still fascinate many people, such as "Dragon Quest V" from the former Enix, by utilizing AI.
It is possible to do test play and reach the level of "partial development and prototyping."
However, it is difficult to fully automatically generate a "highly complete commercial-level game" from scratch using only AI and complete it through to test play.
It is still difficult without the involvement of human game designers and programmers.
This is due to the balance between the current AI's "generative ability" and "structuring/logical construction ability"and the characteristics of the complex system that is a game.
Evolution and current state of AI-driven game development as of November 2025
Revolutionary changes in the development process
In 2025, generative AI is bringing about a major transformation in the game development field.
Processes that once required years and highly specialized skills are being dramatically shortened by the introduction of AI tools, opening doors to non-engineers and individual developers.
AI is showing particularly notable results in the following areas:
Asset and graphic generation
The design of "sophisticated pixel art"at the 16-bit level, in particular,has seen a dramatic increase in efficiency due to the evolution of image generation AI.
Beyond merely reproducing retro games, AI can generate high-quality pixel art assets (characters, backgrounds, items, etc.) that take into account "color tones, shading, and animation patterns" based on prompts (instructions).
Just as 3D modeling AI tools like Meshy generate 3D models from text or images, specialized pixel art AI can now supply large quantities of consistent design assets in a short time by incorporating style guides and reference images.
This significantly reduces the initial production costs for graphic designs with a vast variety of assets and complex layer structures, such as "Final Fantasy VI" and "Chrono Trigger."
However, 3D models are more like early PlayStation level, and beautiful pixel art can be easily created even with ChatGPT or Google AI Studio if you instruct it to be "pixel art style."
Game logic and code generation
Large Language Models (LLMs) like Gemini and Claudeare also demonstrating high capabilities in code generation, which is the core of games.
Game creation is possible by having a programmer use natural language to instruct game rules and logic, such as "When the player character presses left, move left and do not let them go off-screen" or "When an enemy is attacked, reduce HP based on a damage formula."
AI generates code snippets for Unity, Godot, or pure JavaScript/Python.
This has made it possible to quickly build foundational programs, from simple rules like Breakout or Tetris to complex chain-reaction logic like "Puyo Puyo."
If you use a high-performance model like Gemini 2.5 Pro in Google AI Studio, you can use AI with advanced reasoning capabilities.
It is also possible to propose sophisticated game code that includes more complex state management and class design.
Evolution of storytelling and NPCs
The deep stories and unique NPC conversations that support the appeal of the "Final Fantasy" and "Dragon Quest" series are also areas being innovated by AI.
LLMs can generate the backbone of a game's narrative elements, such as "world settings, character backstories, and scenario plot structures."
What is evolving even further is the NPC function as an interactive agent.
Technologies like DeepMind's "SIMA," an agent that can be instructed in natural language, and the rise of high-performance small LLMs (local LLMs) can be utilized.
Dynamic storytelling that does not require cloud connection and changes reactions and narrative developments in real-time according to player actions and conversations is becoming a reality.
This enables the development of NPCs that respond to the player's free input rather than monotonous 'choice-based' conversations, providing a more human-like and personalized experience.
The Superiority of Google AI Studio and Gemini
Google AI Studio is a platform that provides access to the Gemini model family and accelerates AI-powered prototyping.
Multimodal Capabilities
Gemini 2.5 Pro and Flash have multimodal capabilities that allow them to understand not only text but also images.
Developers can present rough sketches of pixel art or screenshots of existing game screens to the AI and instruct it to 'draw a new enemy character in the same style as this character and generate the animation code corresponding to its movements.'
This is a very powerful advantage in game development where art and code are handled in an integrated manner.
High-Performance Reasoning and Logical Construction (Deep Think Mode)
Advanced reasoning functions like the 'Deep Think' mode in Gemini 2.5 Pro are particularly useful in designing complex game systems.
It has the ability to partially address structural and logical challenges that were difficult for conventional AI, such as 'adjusting game balance, simulating diverse playstyles with AI testers, and checking the consistency of logic that manipulates time axes like in Chrono Trigger.'
Developer Ecosystem
The strengthening of Google's ecosystem, such as the provision of the Gemini API and integration into native code editors, dramatically improves the speed of prototyping.
'Rapid realization of ideas, iterative testing, and immediate generation of necessary code' become possible, shortening prototype production that used to take weeks to just a few days.
Challenges in 16-bit Level Game Development and the Importance of 'Human Contribution'
Even with the latest AI technology, several essential challenges remain in fully automating the development of a 'refined game'.
Realizing 'Refined Game Design'
The refinement found in 'Final Fantasy IV/V/VI' or 'Chrono Trigger' is not just about the beauty of the graphics.
It relies on extremely human creativity in designing 'game balance, level design, pacing of storytelling, and the fun that keeps players from getting bored.'
While AI can 'generate' materials and code, its cognitive functional abilities—like the human brain's 'intent and judgment' required to 'make them function as a game and provide an emotional experience'—have not yet reached the human level.
Game Balance
The meticulous work of optimally adjusting the parameters of a large number of items and enemies generated by AI to match the player's growth curve.
Level Design:
Spatial design to control player emotions, such as dungeon structure, placement of puzzles, and encounter rates with enemies.
Consistency in Art Direction
The 'sense' and 'consistency' to integrate assets generated individually by AI into a single worldview and provide a seamless visual experience.
While the above is an area where an AI computer brain can propose and assist, the final 'creative editing and selection' remains the role of the professional designer's human brain.
Maintaining Structure and Consistency
Maintaining the large-scale logical structure and data consistency that make up the entire game remains a difficult challenge for AI.
A long story and generation-spanning system like 'Dragon Quest V' involves complex interweaving of 'countless flags, events, and relationships between characters.'
AI can generate partial code or scenarios.
However, the task of managing and debugging the entire codebase spanning tens of thousands of lines to ensure that everything maintains consistency and operates without bugs currently requires intervention by the human brain's cognitive functions or human engineer capabilities.
In particular, since AI output involves uncertainty, human time may actually be required for quality checking and correcting generated assets and code.
Legal and Ethical Aspects
The issues of copyright and commercial use licenses for pixel art and music assets generated by AI cannot be ignored.
When developing commercial-level games like 'Puyo Puyo' or 'Final Fantasy,' all materials used must clear legal risks.
A hybrid model that adds 'sufficient human contribution (creative editing and correction)' to initial AI-generated products to make them eligible for copyright protection is becoming the standard legal strategy in the game industry as of 2025.
This is based on the idea that AI functions as a 'tool' and final creativity belongs to humans.
Possibilities for Game Development in the AI Era
AI technology as of November 2025 enables 'high-speed prototype development' and 'material generation' for 16-bit level pixel graphic games.By combining 'Google AI Studio, Gemini, and other generative AI tools,' one can prepare the main systems and high-quality assets of a game in a few weeks.
Creating components and bringing them into test play is fully achievable.
However, it is difficult to adjust game balance and optimize the pacing of the story based on the feedback obtained from that test play.
The final process of creating 'refinement' by removing bugs from the entire system still requires deep insight and editing capabilities from the cognitive functions of the human brain of 'professional game designers, writers, and engineers.'
AI isrealizing an 'era where anyone can create programs,' and no-code programming is also a possibility.
It is a powerful co-creator that embodies ideas at astonishing speed.However, the 'philosophy' and 'emotional depth' that form the core of historical masterpieces
have not yet been produced by an AI computer brain on its own.If you utilize AI
not as a 'fully automatic development machine' but as a 'super-high-performance development assistant' that collaborates with the human brain,it is possible to create games that are incomparable to the 8-bit Game Boy level games created with Claude in 2024.
You will be able to produce masterpieces with dense and refined pixel graphics in a surprisingly short period and conduct test plays with beautiful materials.
This signifies the dawn of a new era where the human brain's creative overall command ability—'how to utilize the computer brain called AI'—is questioned, rather than technical skill.
As of November 2025, it is partially possible to create computer games with 16-bit and 8-bit retro game level structures, such as those of the Super Famicom, Family Computer, and Game Boy.
Some people are also using AI to create games for the purpose of training individuals with higher brain dysfunction or cognitive impairment, or for cognitive rehabilitation.
Technically, it is possible to perform effective playtesting at a very high level.
This is because the powerful "generative, personalization, and analytical/evaluative capabilities" of AI enable the development of rehabilitation games (rehabilitation gamification).
The three core requirements for computer game training, also known as therapeutic games—"repetition, individual optimization, and objective evaluation"——require the cognitive functions of the human brain.
Partial game creation has become possible due to the evolution of AI.
This is because we have reached a partial stage where it can be realized without human intervention, or with minimal intervention.
The latest multi-functional AI models, such as "Google AI Studio, Gemini, and Claude," demonstrate their power not only in game development but also in the integration of medical and psychological knowledge.
Going beyond mere game creation,it enables the construction of "personalized therapeutic intervention systems."
Affinity between cognitive rehabilitation and retro game structures
Rehabilitation for higher brain dysfunction and cognitive impairment can be developed by targeting specific cognitive domains (attention, memory, executive function, spatial awareness, etc.).
The goal is to train these in a repetitive and step-by-step manner.
The latest stroke treatment guidelines state that there is evidence for the treatment of attentional disorders using computer training.
There is an affinity between the requirements for attentional disorder training and the game structures of the Super Famicom (SFC) and Family Computer (FC) generations.
Structural characteristics of 8-bit and 16-bit games
Simple and clear rules
Compared to modern complex 3D open-world games, the rules and control systems of FC and SFC games are very simple and clear.
Super Mario Bros. involves "moving right and jumping," and Tetris involves "stacking falling blocks without gaps," which are program contents with low cognitive resource load.
The computer brain known as AI has clarified the tasks that need to be focused on.
This holds the potential to provide an ideal environment for early cognitive rehabilitation, where patients can easily understand the purpose of the task and focus on training without being distracted by unnecessary information (noise).
Low visual noise
Dot graphics have less visual information compared to high-resolution photorealistic graphics.
This has the advantage of making it easier for people with attentional disorders to accurately extract necessary information (items, enemies, the path to follow next, etc.).
The clear contrast and simple color schemes of the FC and SFC levels are extremely effective in reducing cognitive load.
Repetition and immediate feedback
Games of this era, such as "action games, puzzle games, and RPG battle systems," are designed based on "repetition and immediate results."Even if you fail, you can retry immediately, and there were many games with immediate feedback where the results of actions were reflected on the screen right away.
These are essential elements for maintaining learning motivation and drive in training.
Therapeutic transformation of retro game structures by AI
The latest AI, such as LLMs like Gemini and generative AI, has the partial ability to transform these retro game structures not just as tools for game creation butas "therapeutic intervention tools."
Advanced generation and personalization of cognitive training games by AI
AI realizes ultimate personalization by generating and adjusting "game difficulty, speed, and visual elements" in real-time to match the individual symptoms and rehabilitation plans of higher brain dysfunction.
Automatic generation and customization of training content
By utilizing Google AI Studio and Gemini, it becomes possible to automatically generate advanced game content such as the following:
Task generation for executive function training
Rehabilitation for executive dysfunction (decline in planning, goal setting, and problem-solving abilities) requires complex, multi-stage tasks.
AI can learn the quest structures of SFC-era RPGs (like Final Fantasy or Chrono Trigger) as templates, and potentially partially auto-generate quests that adjust "the number of subtasks, the complexity of required information, and the presence or absence of time limits" according to the patient's progress.For example, a series of tasks such as "deliver an item from villager A to B, exchange that item for material C, and use it to repair D" might be created.
AI generates these in real-time and gradually increases the difficulty.
Adjustments for Attention and Working Memory Training
Training for attentional functions (sustained attention, selective attention) and working memory depends heavily on the speed and complexity of information presentation in a game.
AI analyzes the player's past play data (reaction speed, error rate, patterns of mistakes) and dynamically adjusts the following "therapeutic parameters."
Increase or decrease of visual noise
Increase or decrease the number of moving objects in the background, the number of colors, and the frequency of flash effects.
Task execution speed
Fine-tune the falling speed in games like Puyo Puyo or Tetris to match the patient's "Zone of Proximal Development."
Information retention capacity
Adjust the number of times important information is presented in RPG conversation scenes or the number of items to be remembered according to working memory capacity.
Automatic generation of code and assets
Retro game-level structures are relatively simple code structures for modern AI, and this is an area where LLM-based code generation is highly accurate.
Models like Gemini have the potential to instantly generate 8-bit/16-bit game logic (collision detection, simple AI movement, scroll control, etc.) for game engines like Unity or Godot, at least in part.
Pixel art generation AI can supply large quantities of simple, highly visible pixel art assets tailored to the goals of cognitive rehabilitation, significantly reducing design costs and time.
This gives developers the flexibility to quickly change game training content when treatment protocols for higher brain dysfunction are modified.
AI-based objective evaluation and feedback systems
The success of rehabilitation depends on objective effectiveness measurement and the continuous review of training based on that data.
AI functions in this evaluation process with a level of detail and accuracy that surpasses human therapists.
Integrated analysis of biological data and play data
Advanced AI systems can obtain data not only from in-game scores and clear times, but also from wearable devices worn by the patient (if available).
It is also possible to integratively analyze biological data (heart rate variability, skin conductance, etc.) and behavioral data during gameplay (eye tracking, hesitation during input, time spent in specific areas, etc.). This allows for the quantitative grasp of
"real-time cognitive load states" and "timing of frustration," which are deep psychological and physiological reactions difficult for the human eye to capture, enabling objective evaluation of the "effectiveness and adaptability" of the training.
Automatic generation of training reports
Based on this integrated analysis, AI automatically generates detailed training reports for cognitive rehabilitation specialists.
In report creation, it can document correlations between specific cognitive functional areas and behavior, such as "selective attention function improved by X% over the past week, which was particularly notable in the high-speed mode of the Tetris-style task," or "planning function still shows a high error rate in quests with three or more stages."
Reports created by AI can serve as powerful evidence for therapists to quickly and scientifically review treatment plans, potentially improving the quality of treatment dramatically.
Possibilities for AI therapist agents
As a more advanced application of AI, incorporating an "AI therapist agent" into the game can be considered.
NPCs equipped with LLMs like Gemini can infer the patient's emotional state during gameplay.
They can provide empathetic and specific feedback and motivational conversation in natural language, such as "Don't worry about that mistake, let's calm down and try thinking from the bottom row again."
This allows patients to receive psychological support while progressing through training, contributing to a reduction in training dropout rates.
NPCs in game terminology are an abbreviation for "non-player character," referring to computer-controlled characters that the player does not operate, such as "villagers, shop clerks, or enemy characters".
Ethical and clinical challenges and prospects
While the development of AI-based cognitive rehabilitation games holds great potential, clinical validity and ethical considerations are essential for their practical application.
Verification of clinical validity
The effectiveness of AI-generated games must be verified through rigorous clinical trials.
It is not enough for the game to simply be "fun"; there is a need for the accumulation of evidence that "game training has led to improvements in cognitive function in real life (e.g., improved ability to plan shopping or work)."
This verification requires close cooperation between neuropsychology experts and AI developers to clarify the correlation between the parameters output by the AI and cognitive function evaluation indices.
Data privacy and ethics
AI systems require the collection and analysis of sensitive health information, biological data, and detailed behavioral logs of people with higher brain dysfunction.
"Data anonymization, security, and privacy protection" become the top priority.
Strict ethical guidelines and sufficient informed consent from patients and their families are essential to ensure that AI systems do not collect excessive personal information under the guise of "treatment."
Optimization of role division between humans and AI
The AI computer brain is an "extension tool," not a "substitute" for the human brain.
By having AI handle "game generation, difficulty adjustment, and objective evaluation," therapists with human brains can focus on "building trust with patients, emotional support, and bridging cognitive functions to real life based on individual factors."
By utilizing AI, more time and resources can be concentrated on areas that require the creativity and empathy unique to the human brain.
The evolution of AI in November 2025 may make it possible to personalize rehabilitation for people with higher brain dysfunction while basing it on the structures of retro games, which are like relics of the past.
Computer training for attentional disorders that approaches individual factors has the potential to become a decisive turning point in evolving into something more effective and accessible.