Asilla, which specializes in security systems featuring state-of-the-art behavior recognition AI, has established a new research team focused on human science. The team, known as the Human Science AI Research Team (HSAR), was launched in February 2023 with the goal of enhancing AI technology even further. 

In this article, we spoke with Mr. Masahiro Wakasa, the CTO and founder of HSAR, about research efforts and future developments related to Generative AI.

The potential for research expands through the use of innovative "generative AI."

ー HSAR's current research focus is on generative AI. But what exactly is generative AI?

‍Generative AI," also known as "generative artificial intelligence," refers to artificial intelligence that generates data based on certain input. For example, through its own learning based on past data or human instructions, this AI can generate images, text, music, and other creative outputs. It is truly a creative AI.
 Since the introduction of the image-generating GAN in 2014, research on generative AI has continued to advance rapidly. In recent years, "generative AI" has attracted further attention in the AI industry, and with the availability of image-generating AI for general use in Japan, AI-generated images, including "AI Picasso," have become a hot topic.
Generative AI has matured significantly in terms of technology and is currently being applied in various fields. We have reached a stage where the challenge is how to connect what AI has created to new values. Our company is currently utilizing this technology from the perspective of action recognition AI.


‍ー What specific research is HSAR conducting and how is it being progressed?

‍At HSAR, we are attempting to use Generative AI to automatically generate behavior data based on human actions. Behavior data refers to data related to human behavior, such as "fighting" or "collapsed and stopped moving." We conduct research and development of AI based on such behavior data, but traditionally, data acquisition has relied heavily on analog methods, such as us physically performing the behaviors and recording them. We installed cameras and had people perform behaviors hundreds or thousands of times in front of them to patiently accumulate data as behavior data.
The accuracy of AI is directly proportional to the amount of data it learns from. Currently, we are considering using Generative AI to generate new behavioral data based on the vast amount of data accumulated in the past for the purpose of further data collection. Then, the AI can learn by using the generated data. We are currently conducting research to achieve such a positive cycle.

‍

Virtuous circle with New AI

‍
ー When it comes to advancing the research, what do you feel is the most difficult thing?

‍The most difficult thing we feel in advancing our research is how to correctly distinguish whether the "human-like" behavior data generated by AI is truly human-like. It is challenging to accurately determine whether a certain behavior data is "human-like" or rather a mechanical movement based solely on computer science techniques. Therefore, we believe that research in humanities and social sciences is essential for this kind of discrimination.

The behavior data is fundamental to our AI research, so by using only truly human-like behavior data, the accuracy of AI predictive models can be further improved, leading to new learning. We hope that this will lead to a cycle where an AI that can truly infer human behavior is created.

The current stage is still in the research phase, but we have begun to see some promising results. Leading up to the use of generative AI, we have spent many years painstakingly accumulating human behavior data. I feel that it is precisely because our company has a vast amount of behavior data that pursues human-like behavior, we are entering the next phase.

‍

The proper utilization of AI will break down the "traditional barrier" and lead to the promotion of research.

ー Compared to conventional data acquisition methods, what are the benefits of using generative AI?"

‍Although it hasn't been precisely quantified, there has been a significant improvement in the speed of data generation.
To illustrate, let's assume that over 10,000 pieces of data are required for accurately detecting a single action. Traditional methods of data acquisition can take up to two weeks to complete this task. Conversely, utilizing generative AI can generate the necessary data within just 1-2 hours.
In addition, unlike traditional methods that involve human labor and hence entail labor costs, generative AI only requires the user to press a keyboard and wait for the process to complete. This means that data generation can be achieved without any human labor.

I feel that there is a merit not only in dramatically improving the speed of data generation, but also in reducing the cost.


ー Will experts be able to focus on further research if AI can generate highly accurate data on its own, without human intervention?

‍We see the use of generative AI as a turning point in our research going forward. The HSAR team will continue to study the use of generative AI to further improve AI accuracy.
To improve the accuracy of AI, the underlying data is always necessary. However, not only quantity, but also quality of data used in AI is required. Therefore, data acquisition requires both time and cost.
Regarding data acquisition, ethical issues related to cheap labor have recently become a concern. If AI can generate accurate data on its own, it could also lead to a solution to these ethical issues.

‍

Realizing solutions to social issues with "Action Recognition AI x Generative AI

‍
ー The utilization of generative AI, which is gaining worldwide attention, is likely to have a significant impact on future research. Could you please tell me about your future efforts in this regard?

‍Generative AI, which has the characteristic of "creating on its own," has made remarkable progress in the past 1-2 years. Many generative AIs are currently being developed and deployed primarily in the entertainment industry, such as generating beautiful images and illustrations

This trend is expected to continue in the entertainment industry, but it is also believed to expand to solve business challenges in the future.

Our strength, "action recognition AI," is one of the areas that can solve those issues. While envisioning such a future, at Asilla, we are already united as a team and proactively advancing our research and development towards that future.

‍

Profile of Masahiro Wakasa 

After completing his postgraduate studies at Tokyo Institute of Technology, he worked for JGC Corporation where he was involved in plant design IT work on overseas construction projects. He then joined Asilla Corporation, where he was in charge of proof-of-concept and product development projects related to behavior recognition AI, and became the company's executive officer and CTO in 2022. He continues to focus on product development using AI technology and research and development of new technologies.
‍‍

"Human x AI Innovation" initiative strives to attain the world's leading action recognition AI technology.

Asilla is dedicated to the philosophy of "Technology Driven Future" and works on research and development every day to realize a world where everyone can live safely and securely. Our team is particularly driven to become the global leader in behavior recognition AI, and we work together to constantly advance and evolve our technology.

Moving forward, we will continue to push for the advancement of our behavior recognition AI technology and create an environment that inspires engineers to pursue their own careers and passions. If you share our interest in AI related to human behavior, we welcome the opportunity to collaborate with you and create meaningful results.  


Queries Regarding HSAR: pr@asilla.jp Person in-charge: Wakasa


Current Page：【The Era of AI trained by Generated Data - The Potential for Research Hidden in Generative AI 】
pr.asilla.jp/technology/hsar-vol5-eng

＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿

Information about HSAR can be found at the following link: 

‍●Human Science AI: Aiming to Develop AI that Takes on World’s Challenges
pr.asilla.jp/technology/hsar-vol1-eng
‍
●AI Field and its Different Specializations: Aiming to Develop AI that Takes on World’s Challenges 
pr.asilla.jp/technology/hsar-vol2-eng

●Why Human Science AI Research Team is Required - What is needed for the future of AI? 
pr.asilla.jp/technology/hsar-vol3-eng

●Demonstrating towards World's No.1 Behavior Recognition AI Technology
pr.asilla.jp/technology/hsar-vol4-eng
‍
●The Era of AI trained by Generated Data - The Potential for Research Hidden in Generative AI 
pr.asilla.jp/technology/hsar-vol5-eng