AI Governance
AI governance refers to the “design and operation of technological, organizational, and social systems by stakeholders for the purpose of managing risks posed by the use of AI at levels acceptable to stakeholders and maximizing their positive impact”*1
*1: Explanation from the Ministry of Economy, Trade and Industry. The Ministry of Internal Affairs and Communications defines AI governance as “formulating frameworks, creating mechanisms, and executing them for appropriately managing the risks, etc. related to AI while adhering to legal frameworks social rules when planning, developing, installing, and operating systems that utilize AI.”
Current state of AI
Striving to accelerate digital transformation (DX), the Toshiba group has been employing AI in infrastructure and other applications and is working on solving various social problems. Presently, DX has become a global trend and the importance of developing and utilizing AI technology has been increasing with digitalization. The Toshiba Group is working on applying AI to solve a variety of social problems in order to contribute to realizing carbon neutrality and a circular economy through digitalization. Yet, while AI increases convenience, it also gives rise to issues that involve all of society, such as problems caused by malicious use and unintended operation of AI, and there is a growing call for ethics and governance related to AI (see figure below). In Japan, there has been a dedicated effort to examine principles of AI. For example, the Ministry of Internal Affairs and Communications (MIC) and the Ministry of Economy, Trade and Industry (METI) jointly released “AI Guidelines for Business,” which brings together “Guidelines for Utilizing AI” by MIC and “AI Governance and Guidelines for AI Principles in Practice” by METI based on the “Principles of Human-Centric AI Society” by the Cabinet Office. Outside of Japan, the trend toward legal frameworks is accelerating. This includes the EU AI Act and AI-related laws established in China and South Korea. Furthermore, under the Hiroshima AI Process that was established at the G7 Hiroshima Summit in 2023, support for AI governance—for example, for ensuring the safety and reliability of AI, particularly generative AI—was made an essential requirement for companies that actively utilize AI as a shared problem worldwide.
Toshiba AI Governance Statement
The Toshiba Group has published an AI governance statement of its basic philosophy concerning AI and has declared its position on AI governance to ensure that the AI provided by Toshiba can be safely and securely used by customers. The preamble of the AI governance statement explains the significance of measures in the AI governance statement in a format that breaks down the “Toshiba Group Corporate Philosophy” and “The Significance of Our Work.” This paper classifies the company’s philosophy regarding AI into four items of “our values” arranged according to seven perspectives.
- Respect for human dignity …AI that respects human dignity
- Ensuring safety and security …AI that considers privacy and security, and that maintains and improves quality
- Commitment to compliance …AI that complies with laws and social norms
- Developing AI and cultivating talent …AI that can evolve and personnel who can deeply understand and utilize AI
- Realizing a sustainable society …AI that supports a sustainable society
- Emphasis on fairness …AI that creates value that is fair and highly diverse
- Emphasis on transparency and accountability …AI where the internals are visible
The Essence of Toshiba and the Toshiba AI Governance Statement
Toshiba’s AI Governance
The Toshiba Group is working on various initiatives to construct AI governance aimed at developing, providing, and operating reliable AI systems based on the concepts laid out in the AI governance statement.
AI Governance
AI Governance
*2 MLOps (Machine Learning Operations): A mechanism for automating continuous delivery of AI and machine learning.
AI Talent Development
In 2019, Toshiba Group began an initiative for nurturing personnel to support businesses that have adopted AI. The goal was to increase the number of AI engineers from 750 in 2019 to 2,000 in 2022, and progress has been made in fostering and developing AI engineers.*3 The goal was reached, and the number exceeded 2,300 by 2024.
*3: Press release: https://www.global.toshiba/ww/news/corporate/2019/11/pr0701.html
To develop, provide and operate reliable AI systems, it is also important to have initiatives for not only technicians but for all staff related to the business to have a deeper understanding of AI. The Toshiba Group is therefore striving to educate all its employees in Japan and is working to improve AI literacy. To promote understanding of the importance of using AI in business, work has been proceeding since 2023 on increasing awareness at the management level. Progress is also being made with a culture change toward using AI through reinforcement and understanding of the importance and fundamentals of AI, with focus on managers.
Toshiba Group AI Engineers
The key to supporting the nurturing of AI personnel in the Toshiba Group is to clearly define the roles of AI personnel unique to Toshiba and an AI curriculum (AI education and training course framework). Personnel involved in AI are divided into four roles, and education paths are provided corresponding to each of them. Each role is provided with AI training courses for each level, and can receive training in around 40 courses, including 7 unique courses developed by Toshiba. The “Toshiba AI Engineer Training Program,” which was established in 2019 jointly with the Graduate School of Information Science and Technology at the University of Tokyo is held twice per year with a class size of around 50 people, and has trained around 500 high-level AI technicians so far. Going forward, Toshiba will hold basic literacy courses so that all employees wishing to work with AI can study the basics of AI as well as practice courses where those without technical expertise can try creating and using AI, and will work to provide education to suit everyone’s level.
4 types of AI personnel roles
AI curriculum overview
Development and Utilization of AI Technology
The Toshiba Group is working on initiatives to construct an AI technology catalog to showcase its AI technical resources and to promote utilization of AI technical resources within the group. This makes it possible to reduce unnecessary development, to understand the strengths of AI technology and its suitable applications, and to utilize the AI technology appropriately in business applications. This initiative began in 2019, and there are currently over 300 AI technologies that have been registered (as of March 2025, updated periodically).
Around 100 candidate technologies from among these are published in the “Toshiba AI Technology Catalog” on the Toshiba website.
“Toshiba AI Technology” Leaflet
Maintaining and Improving AI System Quality
Toshiba is also working to create mechanisms for maintaining the quality of AI systems. In addition to formulating Toshiba’s own unique “AI Quality Assurance Guidelines,” the company is also undertaking initiatives to showcase quality assurance by using a “quality card” that is visible to the user. For example, by not only displaying the performance of an AI system using the recognition rate or correctness rate, but also showing the fields that the AI system is good at and the fields that the AI system is bad at based on the data trends used for training and evaluation, this enables customers to use the AI system while also understanding the characteristics of those systems. Furthermore, Toshiba is moving forward with the introduction of an MLOps Platform as a tool for continuously maintaining performance to ensure that there are no issues such as performance degradation due to environmental changes after starting AI system operation. Presently, the MLOps Platform is being employed on AI systems such as those at manufacturing sites inside and outside the company and social infrastructure, and there are plans to expand the range of applications in the future.
1. AI Quality Assurance
Because AI systems use an AI model that was built by training on data, the prediction results for unknown data cannot be defined and quality assurance is difficult. Based on characteristics of AI such as that its performance depends on the amount and quality of data used for model construction and the possibility that AI performance will vary depending on the usage environment and usage method, it is necessary to add new concepts and approaches to traditional quality assurance.
The Toshiba Group has summarized notable points in the quality assurance of AI systems in “AI Quality Assurance Guidelines,” based on external trends such as standardization and formulation of legal frameworks regarding AI. In addition, an “AI Quality Assurance Process” containing a quality checklist showing what specifically needs to be confirmed has been compiled as a reference. Using these guidelines and process, AI system quality assurance processes can be constructed or existing processes can be improved.
Given that the concepts of quality differ between AI systems and conventional systems, it is important to share an understanding of the quality of AI systems with customers. The Toshiba Group summarizes the results of the various evaluations and checks conducted during development into an AI Quality Card to make it easy for customers to understand and display to users.
*4: Quality checklist explained on previous page
*5: At the MLSE Summer Camp 2022, organized by the Japan Society for Software Science and Technology, this approach was highly praised by attendees and received the Best Presentation Award
2. MLOps Platform
MLOps*6 is a mechanism for detecting performance degradation and reconstructing AI models. Although AI models are developed based on available data (training data), if the usage environment changes after operation begins and new data is input that was not available during model training, it becomes difficult for the AI model to output correct responses, and the performance is degraded. AI systems where the usage environment changes need to constantly monitor for changes in the input data and for performance degradation. If a fault or performance degradation is detected, the operating model needs to be updated by retraining. The mechanism for automating this process is called MLOps.
*6: MLOps (Machine Learning Operations): A mechanism for automating continuous delivery of AI and machine learning.
Toshiba Group has developed the MLOps Platform for automating monitoring of data and AI model prediction precision, and processes for AI model management including data collection and AI model retraining with the aim of improving AI service development speed and quality. This makes it possible to execute the AI model update process to track changes in the data even for people who are not experts.
Toshiba MLOps Platform
*7: CI (continuous integration): A mechanism for automating the building and testing of software.
*8: CD (continuous delivery): A mechanism for automatically deploying developed software to the Operational environment.
Appropriate application of AI to Business
For Toshiba Group AI to become trusted and adopted by more businesses, it is important to understand and continuously improve the state of AI governance. The mechanisms for clarifying the impacts and risks brought about by separate AI systems to play a role in business decision-making and to maintain and improve AI governance is called AI risk management.
Although there is a tendency to look at the positive aspects (opportunities) brought by AI systems when using an AI system, unavoidable negative aspects (risks) brought by AI systems also need to be considered. AI risk management at the Toshiba Group aims to provide safe and high-value AI systems to society by minimizing risks (such as leakage of personal information and discrimination) in the AI systems provided by the Toshiba Group and maximizing opportunities (such as improved business efficiency and solving social problems). By performing AI risk management focused on the business units and companies that handle AI systems as products and services, customers can be provided with safe and secure high-value AI systems that reduce the likelihood of AI system faults and losses due to AI systems.
AI risk management overview: Risk is reduced through repeated cycles of AI risk assessment and AI risk response
The AI system developer fills in the required items in the “AI risk assessment sheet (survey sheet for AI systems).” This includes not only basic information about the AI system but also questions for measuring the degree of impact if that system causes an incident and the likelihood of an incident occurring. Based on these answer results, the degree of risk is evaluated, and feedback is given with the results and advice for mitigating risks. Based on the results, the AI system developer performs risk mitigation with the aim of reducing risk. By performing this iteratively, this AI risk management system serves as a mechanism for the Toshiba Group to reduce the risk inherent in AI systems.