The Future of Machine Learning and Tomiki Aikido
Gabriel Adalem
The first goal of this essay is to very briefly introduce Machine Learning (ML) and the future utility of ML in an Aikido context. Second is describing the goals of these teaching tools and how they might be used, from live video feedback to tactile learning with a humanoid machine. Third is the eventual goal of using ML as a tool for scaling education and organizational growth in Aikido. Last is a call to action for what can be done in the meantime, primarily data collection and its value.
Introduction to Machine Learning and Aikido
The current cutting edge machine learning model is GPT-4o. It is a multimodal model, built using multiple sources of information: visual, audio, and text data. The specialty of GPT-4o is that it can observe a person visually, hear what they’ve said or done, all in real time, in-context while giving guidance. The main difference between these models and ML prior to 2019 is that it scales linearly with compute. This implies there is a large amount of potential growth and capability (as evidenced by Nvidia’s stock value). A use-case in an Aikido context would be a student proofing their skills in front of an AI that gives useful feedback. Be it after seeing solo warm up drills, or as far as maintaining the Three Principles and Six Concepts when observing Junonahon.
Utilizing Tactile-Responsive Humanoid Robots
Beyond visual feedback from an ML model, would be the utilization of tactile-responsive, humanoid robots. They would be more than a demonstrative humanoid: motion flexibility and strength, functioning as an uke or tori. The long-term implication is that these types of tools could expand the capability of aikidoka by generating novel methods in randori and enbu, just as ML has with the notoriously complex board game Go. Sadly there are no visual teaching models that exist for high motion visual feedback yet. But as time goes on, the known utility and capability will likely increase. In addition, there will be competitive, accessible alternatives to construct or redesign an ML model for the specific needs of any group or skill set.
Addressing Population Dynamics and Knowledge Transfer
Taking a step back, these tools are increasingly important as the population growth-rate shifts down due to a correlation between birth rate and overall increased planetary wealth. With a smaller youth population and a larger retired population, demand for specialty experts in any field is becoming more intense. The risk of lost knowledge transfer will become increasingly high as time continues. Consider Boeing’s airplanes in 2024, which did not succeed at transferring vital skills and quality assurance to newer factories. But the most important reason for utilizing these potential tools, is personal. As a spiritual Aikido principle, giving back is more important than receiving. As my physical conditions have shifted over the past 4 years, my understanding of goals and long-term vision have as well. Where the primary source of eudemonia (action that provides fulfillment and purpose) in my Aikido is not so much in personal achievement as it is in preserving knowledge and maintaining fidelity for future generations.
Scaling Educational and Organizational Growth
To paraphrase the late Merrit Stevens, “it takes approximately 400 students coming in the door to produce one shodan.” The primary bottleneck for teaching students is having enough teachers. Having a tool that allows scalability can increase the intake of students; and hopefully that will translate to an increase in human teachers, as well as an increase to the organization overall. Imagine a future hypothetical, where adding on another 50 teachers is as easy as copying software to the new humanoid robots.
The Current State and Future Potential of AI Tools
Unfortunately these opportunities are still beyond the cusp of current technological innovation. The reason the bulk of AI tools exist in their current, primarily-textual form is because of the availability and simplicity of text. More complex AI will mean a higher demand for higher-accuracy, complex information. The aikidoka sensei will still exist, even in a hypothetical age of high-utility AI. Because most computational energy will be directed toward the most financially valuable ventures (see the economic concept “comparative advantage”). In this scenario, transferring Aikido into multiple recorded mediums is paramount. Data collection can be simple: video footage, multiple cameras. Or an added challenge is securely attaching a smartphone to a body part and using it’s IMU (inertial measuring unit) along with the footage.
Data Collection for Future AI Models
The more granular this information the better, and the larger amount of the same videos collected with good form, the better. As building any decent model will require hundreds if not thousands of iterations of video. In addition to videos of varying sizes and style of students, including bad form that is corrected in a very detailed way. This is not an immense task when every aikidoka can set up a camera tripod or two (the more cameras the better), and throw a partner on the mat. Any preserved information is a teaching tool for the future, and has lasting value that would otherwise be lost.
Conclusion
This future of highly capable tools will greatly benefit from having silently observed the corpus of recorded aikido, as if it were a silent observer in infant stage, finally acquiring motion years later. This vision is not only for preserving the art, but expanding the practice, principles, and values of Tomiki’s aikido.
Comments are closed