Main image of Accelerating the Evolution of AI Technologies and the Social Implementation of Applications With Open Innovation Based on Collaboration Between Different Fields

Open Innovation × PKSHA

Accelerating the Evolution of AI Technologies and the Social Implementation of Applications With Open Innovation Based on Collaboration Between Different Fields

The application of artificial intelligence (AI) has steadily spread.
AI is secretly being used in unexpected services of various fields such as retail, advertising, medicine, education, and finance.
For the very reason that these fields were previously disconnected from digital technologies, it appears that the utilization of AI is spreading. PKSHA Technology is a company that has greatly contributed to the expansion of AI application in Japan. The company has accelerated the expansion of effective AI application through the interlinking of AI algorithm research and development and the social implementation of solutions and products that apply those algorithms. Currently, the company is working on open innovation through links with different fields and tackling the challenges of AI technology evolution and the expansion of social implementation.
We spoke with the company's Representative Director Katsuya Uenoyama about AI utilization and the front lines of its application expansion. Part 2 of our series introduces the aims of the open innovation that PKSHA is pursuing and the AI evolution that can be expected as a result.

In Artificial Intelligence (AI) based on machine learning and deep learning, the inference accuracy is increased by training on a massive volume of data. Moreover, the processing results that can be obtained change significantly based on the quality of the data used for inference processing even for trained AI systems. In other words, the data handled by AI systems can be described as the source of the value that is produced.

The technologies needed to realize superior AI application systems are not only advanced neural networks and high-performance computers. Technologies for collecting the data handled by AI in an effective, efficient, and high-volume manner are also essential.

The Sensors That Are Essential for Data Collection Were Not Developed With AI Application in Mind

If AI is applied to search sites, e-commerce sites, and social media, then a large volume of data that serves as the AI training material and the data used for inference processing can be obtained as digital data.
However, in order to obtain data for manufacturing industry production sites, product flows at retail stores, human biometric, and other data, it is essential to utilize sensors with features that capture real-world information as digital data.

Nevertheless, many of the sensors that are currently commercially available were not developed with the expectation that they would be used to collect data handled by AI.
Most of them were developed with the idea that they would be applied to electrical and electronic equipment control systems. Specifically, the technologies were developed in the direction of improving the sensitivity, precision (resolution), and accuracy for application to more delicate and advanced control as well as improving the miniaturization, weight reduction, low power consumption, and environmental resistance to collect information from a greater number of locations.

So how did the developers of AI algorithms think about the fact that the features and performance of current sensors are not necessarily optimized for application to AI?
PKSHA Technology Representative Director Katsuya Uenoyama states, "We verify both the AI algorithm that is trained on the data and the value that is obtained in the social implementation destination and quickly run feedback loops to focus on exploring ways to achieve greater effectiveness. The question of how well we can utilize the data on hand is important, and it is not the case that we are considering any physical environment in the real world that data was collected in" (Figure 1).

Image of Uenoyama explains what PKSHA focuses on in AI algorithm development
Figure 1 Uenoyama explains what PKSHA focuses on in AI algorithm development

However, to look at it another way, I feel that if we improve such a situation, it has the potential to create future growth.
Software companies that research and develop technologies that handle data and companies that develop and provide hardware that collects data can in fact be said to be in a contiguous relationship in which they are positioned as the starting point and ending point of the data flow within AI application systems. At first glance, it may seem as if both types of companies are in different industries and have little relevance to each other, but perhaps they are essentially in a highly compatible relationship. If they were to collaborate and combine data collection technologies and utilization technologies to optimize the entire data flow, I have a feeling that innovation would be created from there.


[PKSHA Technology Inc.]

Under the mission of "Shaping Future Software," PKSHA Technology provides AI SaaS and develops AI solutions using machine learning and deep learning algorithms developed in-house to create future relationships between companies and customers. In addition to providing solutions according to customer issues based on a wide range of technologies including automatic answering using natural language processing technologies, image/video recognition, and prediction models, the company supports the promotion of Japanese DX in a multifaceted manner through the deployment of AI SaaS that solves common issues with the goal of realizing a prosperous society in which people and software evolve together.

Innovation Through Sensors Based on AI Data Utilization

Uenoyama explains, "What kind of sensors can we use to obtain data to use as training material and processing objects to achieve AI with more valuable results? It is important to aim to further advance AI and maximize the impact in application areas and start such discussions now."

For example, even if it is data that was obtained using conventional sensors, if it is converted to different information using an AI algorithm, new sensors may be created with unprecedented value. In fact, in the image sensor field, it has become possible by using AI to extract people from captured video, infer attributes such as age and gender, and search for suspicious persons based on movement and behavior. In other words, image sensors have evolved into age and gender sensors. If the same approach is applied to various sensors such as acceleration sensors, temperature sensors, and magnetic sensors, there is a possibility of obtaining more valuable information than what could previously be detected.

Moreover, there may be some problems that could be easily solved with AI algorithm support within the noise countermeasures that were a focus of previous sensor development. Furthermore, if we think about it based on the assumption of AI data utilization, then there may also be cases in which the high performance and high sensitivity of existing sensor products may not be necessary. In these cases, rather than devote development resources to hardware technologies in line with the trend of conventional center development, devoting them instead to another field such as improving the application of AI utilization may advance the shift to high performance and low cost for the entire system.

In fact, in the development of processors to execute AI-related processing at high speed, they have succeeded by daring to advance technology development that runs counter to the conventional technology development trends. In the development of processor chips expected to be equipped in computers to date, the number of bits of data that could be handled by the arithmetic unit increased from 32 bits to 64 bits to improve the operational precision. However, in processors for AI-related processing, they dared to run counter to the trend and promoted a small bit size by changing the specifications to 8 bits. By simplifying the configuration of individual arithmetic units and instead increasing the number of arithmetic units equipped on the chip, the chip parallelism was increased. They were improved so as to be able to process more data at high speed, and the system performance was significantly increased.

Combining Hardware and Software With the Goal of Achieving Valuable AI Systems

Uenoyama emphasized the importance and significance of the current advancement of ties between the dissimilar fields of software companies and hardware companies saying, "I think that the approach of combining architectures between sensors and AI algorithms is an untouched area, globally speaking."

In fact, in areas outside of the combination of sensors and AI algorithms, the development to combine hardware and software is gradually becoming more active. For example, in the humanoid robots announced in recent years, there are examples of development that combine software technologies for movement control with hardware technologies, such as semiconductor chips for control, driving actuators and mechanical components, batteries, etc. As a result, those robots are capable of smooth movements as well as delicate and varied operations that cannot be achieved with robots created through a combination of existing hardware components.

At PKSHA as well, they have already started open innovation with several hardware manufacturers. They are said to be starting to explore ways to realize more valuable systems and services by developing the specifications and structures of the hardware and software that make up AI systems in an integrated way.

For example, "Robot arms used to automate plants will move in the direction of evolving around AI-based control going forward. Major changes may occur in the structure and how hardware parts are made such as the advancement of actuator development optimized for algorithms control by AI. I believe that in previous system development, in many cases it was advanced in a 'hardware first' manner. If hardware is redeveloped in a 'software first' approach, I think that examining what kinds of changes and value would be created in the system development would be a significant thought experiment."

PKSHA Technology代表取締役 上野山 氏の写真

PKSHA Technology Representative Director Katsuya Uenoyama

After primarily focusing on work relating to the Internet industry/software industry as a new graduate at the Tokyo and Seoul offices of Boston Consulting Group, Uenoyama participated in the launch of the Silicon Valley, US office of GREE International where he was involved in large-scale log analysis work for Web products. After earning a PhD (machine learning) at the Matsuo Lab (The University of Tokyo), he was appointed as a Laboratory Assistant Professor. In 2012, he founded PKSHA Technology in parallel. He is a member of the Cabinet Secretariat Digital Market Competition Council Working Group, and a committee member of the METI Expert Group on How AI Principles Should be Implemented. In 2020, he was selected as a member of the "Young Global Leaders YGL2020" by the World Economic Forum (Davos Forum).

Cross-Industry Collaboration Is Not an Easy Job but a Challenge That Is Still Worth Tackling

However, it would seem that surpassing company boundaries and changing system development procedures to develop hardware and software in an integrated manner is not that simple. This is because dissimilarities emerge between new businesses that define and confront new frameworks and the way that existing businesses with a track record advance development.

"We want to quickly execute a PDCA (Plan, Do, Check, Action) cycle according to software development methods to advance research and social implementation. However, depending on the collaboration partner, there may be cases in which there are fixed rules regarding the procedure and method when running the PDCA cycle, and it may not be possible to keep pace as expected. Therefore, as one choice for preparing a place for open innovation that can uniquely define a PDCA cycle as a kind of isolated island, we are working to prepare a system that can carry out smooth and effective collaboration" (Uenoyama).

Open innovation through collaboration by different industries is also a kind of business reform that changes how previous work was carried out from the ground up.
It is not a simple task. Even so, it is an essential initiative for the continuous advancement of technology evolution and business growth in anticipation of the future.
The collaboration between the previously disconnected software and hardware companies will likely further expand and accelerate going forward. Finally, what kinds of innovation and new forms of value will be created through the practical application of open innovation? We cannot wait to see what happens.

Related articles