A malicious verification attempt: Criticizing the government-biased scenario of Japan's revival generated by DeepResearch

Since the previous DeepResearch results from Gemini felt off, I decided to perform a stress test as the title suggests 💦

In the output below the separator line, citation number 1 is the previous output I had it read as a Google Document.

For what it's worth, this time it only noted "Excerpt:" at the beginning of the excerpted parts, which was fine, but upon closer inspection, it turns out that the content of the excerpts themselves seems to contain its specialty: summarization. If it contains specific numerical values for a summary, it probably failed at numbering. It is not logically broken, and it is consistent. However, it seems it barely followed the instruction to extract directly from the source as an excerpt 💦

As expected, it seems it was difficult to criticize this AI-generated scenario for Japan's revival, as the citation numbers are off, or rather, they don't match the citation numbers of the preceding paragraph... Also, there are citation numbers included that are larger than those listed in the bibliography. This is because the execution of DeepResearch involves the verification of a vast number of citations through iterations, so while it is fine that it listed what it considered important at the end as a bibliography, it is thought that things that were not that important remained as citation numbers.

In order to maintain its value as a DeepResearch agent, it seems that prioritizing the character count is unavoidable. The reason there were few citations last time was that by frequently performing large excerpts, the character count increased, which had the positive side effect of suppressing the number of citations themselves. Anyway, if the content is literally deep, I think an implementation with fewer restrictions on character count or token count would be better, but it seems that's not the case...

Anyway, thank you for your hard work, Gemini 🙇 It would be nice if there were explicit configuration items, but as long as there aren't, the DeepResearch agent has to be treated as a black box. However, it is true that it is consistent in that it does not cause logical breakdowns.

However, what I expected was for it to actively cite more English literature to objectively correct the parts that were too government-official. In particular, looking at the example where lifetime employment in the civil service system was largely dismantled in the West, especially in New Zealand, due to a sense of national crisis, it seems that the West views Japan as lagging too far behind.

This part should definitely be a hindrance to AI-driven drug discovery. If there is regulation by human judgment domestically without competition, it will continue to lose in international competition. Therefore, AI should be introduced into the revision of regulatory standards, not into regulations that rely solely on human judgment and cause delays. Although it involves risks at first, this is the same as the J-curve of GPTs; my scenario is that after 20 years, it will have a competitive advantage in international competition 🙇 (Gemini fleshed out the article above 💦 I didn't know the details of that lol)

...What bothered me is that I feel the term GPTs (General Purpose Technologies) only started appearing after ChatGPT (Generative Pre-trained Transformer) emerged, making the context even more confusing... Well, compared to the fact that in Japan, we created Japanese translations for the titles of Duke and Marquis, which are homonyms from around the same time, and the latter came to be called 'soro-koshaku', and since the former has a higher rank, it feels like you shouldn't call it 'oyake-koshaku', I don't think there is any social impact...

Technical Critique Report on the Consistency between Structural Transformation of the Japanese Economy and Artificial Intelligence Strategy

Chapter 1: Uncertainty in the Structural Transformation of the Macroeconomy and the Diffusion Path of AI Adoption

The Japanese economy is defined as being at a historical turning point from a "cost-cutting economy," where deflation and low growth have been entrenched for decades, to a "new growth-oriented economy" where wage increases and investment are linked to drive growth [1]. The high level of wage increases of 5.08% in the 2024 spring labor-management negotiations and capital investment exceeding 100 trillion yen are symbolic indicators showing a tectonic shift in the macroeconomy [1]. However, careful technical and statistical verification is required as to whether this nominal numerical improvement will lead directly to substantial productivity gains [1].

Excerpt: The current Japanese economy has high levels of stock prices, wage increases, and capital investment. It is a chance to transition to a "new growth-oriented economy." Due to long-term deflation, a "cost-cutting economy" of low prices, low wages, and low growth has become conventional wisdom, hindering appropriate price pass-through. [1]

The government's "Basic Policies for Economic and Fiscal Management and Reform 2024" clearly states that it will pursue real economic growth of over 1% and nominal growth of around 3% toward the 2030s [1]. The foundation of this growth scenario is the "democratization of AI," an attempt to overcome structural issues such as population decline and labor shortages by permeating AI throughout all industries [1]. However, the fact that the real GDP growth rate forecast for fiscal years 2025 and 2026 remains at around +0.7% suggests that the economic effects of AI will be limited in the short to medium term [1].

Excerpt: Real GDP growth rate is forecast to be +0.7% for FY2025 (forecast as of August 2025: +0.6%) and +0.7% for FY2026 (+0.7% for the same). [2]

It has been pointed out that there is a "J-curve effect" where the adoption share must exceed 50% before AI as a General Purpose Technology (GPTs) has a significant impact on the macroeconomy [1]. Looking back at the historical transition in the United States, a time lag of about 20 years was required from the advent of the PC to its contribution to productivity improvement [1]. For Japan to rapidly break through the current initial adoption stage and achieve its goals for the 2030s, the adaptation speed of the entire social system, beyond mere technology introduction, will be a decisive variable [1].

Excerpt: General Purpose Technologies (GPTs) have a cycle where the adoption share accelerates from around 10 years later and exceeds 50% around 20 years later. [2]

Uncertainty in the external environment is also a risk factor that cannot be ignored. External shocks such as U.S. tariff policies could exert downward pressure on Japan's exports and capital investment, potentially creating a negative feedback loop that suppresses AI investment in domestic industries [1]. To increase resilience against such external environments, it is necessary to accelerate the transition to a "growth-oriented economy" that balances the strengthening of supply capacity with the creation of new added value [1].

Excerpt: Looking at the external environment, it is pointed out that there is a risk that uncertainties such as U.S. tariff policies will exert downward pressure on Japan's exports and capital investment, but the continuation of AI investment that contributes to the productivity improvement of domestic industries is also considered important to increase resilience against such external shocks. [2]

Human factors for successfully implementing AI in society are also complex. While positive reactions to AI adoption reach 60% to 70% in the West and China, concerns about "increased work stress" due to an increase in dealing with complex problems are becoming apparent in the Japanese labor market [1]. This psychological barrier reflects not just a lack of skills, but a low level of acceptance of organizational culture and change, and could become a physical and psychological bottleneck that hinders productivity improvement [1].

Excerpt: In a survey conducted among workers in AI-integrated workplaces across six Western countries and China, approximately 60-70% responded that AI has a positive impact on the workplace. Regarding organizational impact, while AI-integrated workplaces show an increase in "job satisfaction," negative views such as "increased work stress" have also been noted. [2]

The government has set a large-scale investment target of 12 trillion yen from both the public and private sectors over the next 10 years, and is proceeding with the development of hardware infrastructure that supports AI computing resources, such as semiconductors and data centers [1]. The entry of TSMC into Kumamoto is estimated to bring an economic effect of 6.9 trillion yen by 2031, and there are high expectations that the development of domestic production bases for advanced technology will contribute to the revitalization of the regional economy [1]. However, if these large-scale hardware investments are not organically linked to value creation in the software and service layers, there is a risk of leading to a decline in investment efficiency [1].

Excerpt: Future investment amount (target for the next 10 years): 12 trillion yen combined from the public and private sectors. In relation to TSMC's first plant in Kumamoto, 1.2 trillion yen is expected to be invested by 2031, with an economic effect of 6.9 trillion yen expected to be created. [1]

Chapter 2: Multilateral Analysis of Factors Behind IT Adoption Delays During the "Lost 30 Years"

The reason Japan lagged behind in digital transformation (DX) during the "Lost 30 Years" is not a single technological delay, but a structural mismatch where employment practices, educational systems, and investment mindsets are intricately intertwined [1]. In particular, Japanese-style employment practices based on lifetime employment and seniority-based promotion have made drastic job re-engineering associated with AI adoption difficult, keeping efforts limited to partial efficiency improvements as an extension of existing operations [1].

Excerpt: Traditional Japanese employment practices, such as lifetime employment and seniority-based promotion, hinder flexible job placement and skill updating, becoming a factor that delays fundamental reforms of business processes associated with AI adoption. [2]

On-the-job training (OJT), which Japanese companies have traditionally valued, is suitable for continuous improvement (Kaizen) of existing processes, but it has exposed its limitations against disruptive technologies like AI [1]. Low expenditure on Off-JT using external professional institutions and a lack of opportunities for workers to autonomously acquire new digital skills have become barriers hindering the accumulation of human capital [1]. A lack of organizational learning ability is being exposed in the face of technological innovation that cannot be handled by on-site experience alone [1].

Excerpt: Traditionally, Japanese companies have emphasized OJT, but they are facing the limitation that on-site experience alone cannot handle disruptive technological innovations like AI. [2]

From an economic perspective, the prolonged deflation has caused companies to prioritize "certain cost reductions," and a mindset of suppressing investment against uncertain future returns has taken root [1]. This is a direct factor that led to Japan falling behind developed countries such as Singapore, the US, and Germany in the "AI Preparedness Index" [1]. The inferiority in multiple items such as digital infrastructure, human capital, and innovation stems from the country's overall investment decisions and priorities in resource allocation [1].

Excerpt: Compared to Western countries, corporate investment in software and IT infrastructure is relatively low, which acts as a physical barrier hindering productivity improvement. Japan ranks below these top countries in the overall score of the AI Preparedness Index. [2]

From a psychological perspective, there is a history where AI adoption has not been presented as a sufficient benefit to on-site workers, but rather perceived as increased management or increased workload [1]. The existence of a segment with a weak awareness of "lifelong learning" to continuously update one's skills is also a factor that significantly delays the penetration of new technology into the organization [1]. Motivation to increase receptivity to change is a core issue in future AI strategies [1].

Excerpt: Regarding organizational impact, while AI-integrated workplaces show an increase in "job satisfaction," negative views such as "increased work stress" have also been noted, which is a factor suggesting barriers to adoption and utilization, or the difficulty of adaptation. There is a weak awareness of "lifelong learning" among some workers to continuously update their skills, and motivation to increase receptivity to change is an issue. [2]

Chapter 3: Japan's "Counter-Offensive": Strategic Evaluation of the Artificial Intelligence Basic Plan 2025

The national strategy "Artificial Intelligence Basic Plan 2025," aimed at recovering from the delay in IT adoption and restarting the Japanese economy, identifies three fields—"Physical AI," "AI for Science," and "Drug Discovery AI"—as Japan's winning paths through selection and concentration [1]. This concept of a "counter-offensive" aims to secure international leadership by expanding Japan's potential, such as manufacturing site data and scientific knowledge, with AI [1].

Excerpt: To recover from this delay, Japan is focusing on a "counter-offensive" concept, with selection and concentration: concentrated investment in the three fields of Physical AI, AI for Science, and Drug Discovery AI. [4]

$$
\small \def\arraystretch{1.5}
\begin{array}{|l|l|l|} \hline
\textsf{\textbf{Winning Fields}} & \textsf{\textbf{Strategic Goals}} & \textsf{\textbf{Japan's Potential/Strengths}} \\ \hline
\textsf{\textbf{Physical AI}} & \textsf{AI autonomy in the physical world [1]} & \textsf{Precision hardware, manufacturing site data [1]} \\ \hline
\textsf{\textbf{AI for Science}} & \textsf{Acceleration of scientific discovery processes [1]} & \textsf{Material development, precision instruments, chemistry [1]} \\ \hline
\textsf{\textbf{Drug Discovery AI}} & \textsf{Fundamental efficiency of drug discovery processes [1]} & \textsf{Pharmaceutical industry foundation, medical data utilization [1]} \\ \hline
\end{array}
$$

"Physical AI" is a concept that realizes recognition, judgment, and action in the real world by fusing with robots and machines, unlike conventional AI that is completed in digital space [1]. This is considered to be the area where Japan can best utilize the precision hardware technology and high-quality "on-site data" it has cultivated for many years [1]. By applying the evolution of LLMs (Large Language Models) to robotics, autonomous work in ambiguous environments is becoming possible, but whether social implementation in this field succeeds depends not only on the speed of technological development but also on the reduction of implementation costs [1].

Excerpt: Physical AI refers to a mechanism or concept where AI is linked with robots or machines to perform recognition, judgment, and action in the physical world. New technologies related to AI, such as "Physical AI" that moves robots in the real world, are advancing. [6]

In "AI for Science," centered on material development (Materials Informatics), AI is analyzing vast amounts of experimental data and accelerating the discovery of new materials [1]. AI-driven research and development in the precision instrument and chemical fields, where Japan maintains global competitiveness, is a lifeline for securing future technological sovereignty [1]. Efforts are underway to create an environment that promotes the sharing and utilization of data across organizations and utilizes failure data (negative data) that has been overlooked until now [1].

Excerpt: The second is "AI for Science," which accelerates scientific research with AI. AI is expected to play a role in strengthening new scientific discoveries and human creativity, and the goal is to lead the world in "AI-driven research and development." [5]

"Drug Discovery AI" aims to improve the success rate of pharmaceutical development and fundamentally streamline processes that require enormous costs and time [1]. The Japanese government has positioned the pharmaceutical industry as a next-generation core industry and is supporting the use of AI in protein structure prediction and target compound exploration [1]. The development of legal data infrastructure, such as the utilization of pseudonymized medical information based on the Next Generation Medical Infrastructure Act, holds the key to fundamentally strengthening drug discovery capabilities [1].

Excerpt: The third is "Drug Discovery AI," which advances new drug development. We will position the pharmaceutical industry as one of our country's core industries and aim to strengthen our drug discovery capabilities. [5]

Chapter 4: A Critical Analysis of Structural Stagnation in Physical AI and Robotics

While Japan is positioning "Physical AI" as a winning strategy, there is a harsh reality in actual market trends and the competitive environment [2]. Japan has maintained approximately 70% of the global share in the traditional industrial robot market, but its dominance is currently on a downward trend [2]. Furthermore, in the service robot market, which is expected to see significant growth in the future, Japan is lagging behind competitors such as the United States, China, and Germany in terms of the number of manufacturers and the scale of investment [2].

Excerpt: Although Japan maintains a 70% share in the traditional industrial robot market, its dominance is currently on a downward trend. More importantly, in the service robot market, which is expected to see huge growth in the future, Japan lacks the number of manufacturers compared to the United States, China, and Germany. [2]

The gap in investment and fundraising with the US and China has reached a decisive level [2]. Giant companies in the US, such as Tesla and Nvidia, continue to invest hundreds of billions to trillions of yen annually in humanoid development and AI infrastructure [2]. In contrast, the market capitalization and fundraising amounts of Japanese robotics-related startups remain in the range of tens of billions of yen, and this difference in capital strength is manifesting as a difference in the speed of social implementation of technology [2].

Excerpt: In the United States, Tesla and Nvidia are investing hundreds of billions to trillions of yen annually in humanoid development (such as Optimus) and AI infrastructure (such as the Isaac platform). In contrast, Japanese robotics startups generally have market values and fundraising amounts limited to the range of tens of billions of yen. [2]

2025 will be recorded as a landmark year in which physical AI and general-purpose robots (such as humanoids) transitioned from the "Proof of Concept (PoC)" stage to the "commercialization" stage [3]. In this transition, the US is dramatically increasing autonomy using powerful "robot foundation models (VLA/VLM, etc.)" as weapons, while China is establishing mass production systems by utilizing its mature supply chain [3]. Even if Japan maintains its existing strengths in precision machinery and control technology, it faces a "structural delay" in an era where innovation in the AI-driven software layer defines the added value of hardware [3].

Excerpt: While Japan maintains its strengths in precision machinery, precision control, and component technology, the structure of falling behind the US and China, which continue to develop against a backdrop of abundant capital, is becoming clearer. The market is transitioning from "proof of concept only" to "commercialization." [3]

What determines the performance of physical AI is the learning infrastructure using large-scale real-world data and digital twins [3]. In the US and China, "data factories" that collect massive amounts of driving and work data are being built to accelerate the model improvement cycle [3]. While Japan has also presented a strategy to accelerate the growth of open data infrastructure through industry-academia-government collaboration, it remains to be verified whether an open strategy can demonstrate sufficient competitiveness against the moves of the US and China, which are accumulating data proprietarily and developing exclusive models [2].

Excerpt: In the US and China, moves to accumulate data proprietarily and develop foundation models are accelerating. In Japan, we will promote the development and social implementation of foundation models by accelerating the growth of open data infrastructure. [2]

Chapter 5: Specific Fields to Revitalize Japan's Unique Strengths and Implementation Challenges

5.1 Technical Bottlenecks in Marine Resource Development and Underwater Robotics

Marine robotics, with Japan's vast Exclusive Economic Zone (EEZ) as a backdrop, is a strategic field that leverages geographical conditions [1]. There are high expectations for underwater drones (ROVs) and autonomous underwater vehicles (AUVs) that realize real-time image analysis and autonomous navigation by AI in the deep sea where radio waves do not reach [1]. However, there are still high technical barriers in underwater wireless communication technology and the development of actuators capable of withstanding extreme deep-sea environments, and further technological innovation is required for commercial-based dissemination [1].

Excerpt: AI analyzes images captured by underwater cameras in real time. Technology adapted to Japan's unique geographical environment (complex coastlines, earthquake-prone areas, deep sea, etc.) is required. [10]

5.2 Social Implementation of Disaster Response and Aging Infrastructure Maintenance

In Japan, which is prone to natural disasters such as earthquakes and typhoons, infrastructure inspection and disaster recovery using AI-equipped robots are directly linked to national security [1]. AI systems that automatically detect cracks and corrosion contribute to reducing maintenance costs and preventing accidents [1]. However, there are aspects where social implementation in terms of systems, such as legal development for operating these technologies in the field and clarifying the location of responsibility for AI judgments, is not keeping pace with technological development [2].

Excerpt: In the construction and infrastructure field, utilization for the inspection and maintenance of aging infrastructure is expected. A system where robots take charge and AI automatically detects and reports cracks and corrosion. [6]

5.3 Digitization of "Takumi (Master) Skills" in Manufacturing and Logistics Sites

Efforts to digitize "Takumi skills," the source of Japan's manufacturing industry, using AI and pass them on to robots are a decisive means of preventing the loss of skills due to the declining birthrate and aging population [1]. Automation of assembly and inspection by autonomous robots contributes to shortening cycle times and reducing error rates [1]. However, the adaptive capacity of AI in high-mix, low-volume production sites remains limited, and further improvement in data density is essential for AI to completely reproduce and expand the intuition and experience of skilled workers [1].

Excerpt: AI-equipped robots have improved cycle times by 20-30% and reduced error rates by 25%. The goal is to digitize and automate tasks that have relied on human experience and intuition by leveraging AI's recognition and judgment capabilities alongside the mobility of robots. [6][10]

Chapter 6: Social Implementation and Friction of AI/DX in the Medical and Nursing Care Sectors

The declining birthrate and aging population facing Japan serve as a powerful driver for the social implementation of AI and robot technologies [1]. The use of AI in the medical and nursing care fields is a pillar supporting the extension of healthy life expectancy and the sustainability of the social security system [1]. The 'AI Hospital' and national medical information platform promoted by the government aim to improve the quality of medical care through the standardization of electronic medical record information.

Excerpt: We are proceeding with system construction to fully launch the electronic medical record information sharing service, which is the core of the national medical information platform that enables the sharing of medical and nursing care information, by the next fiscal year. [12]

The introduction of robotics in nursing care settings reduces the physical burden on staff and substitutes for monitoring tasks, thereby contributing to the prevention of turnover among nursing staff [1]. Care robots and rehabilitation support robots are expected to be areas with significant social impact [1]. On the other hand, psychological resistance to digital technology in the face-to-face-oriented fields of nursing and medicine, as well as the transparency and ethical issues of AI judgment, are factors delaying adoption [1].

Excerpt: There is potential for progress in the use of care robots and rehabilitation support robots. In Japan, where the birthrate is declining and the population is aging, this is expected to have a major social impact on the urgent issue of reducing the burden on nursing care sites. [6]

$$
\small \def\arraystretch{1.5}
\begin{array}{|l|l|l|} \hline
\textsf{\textbf{Medical/Nursing AI Adoption Items}} & \textsf{\textbf{Expected Effects}} & \textsf{\textbf{Emerging Challenges}} \\ \hline
\textsf{\textbf{Electronic Medical Record Sharing}} & \textsf{Prevention of polypharmacy, optimization of medical expenses [1]} & \textsf{Security measures, information standardization [1]} \\ \hline
\textsf{\textbf{Monitoring Sensors}} & \textsf{Fall detection, labor saving for night patrols [1]} & \textsf{Concerns about privacy infringement [1]} \\ \hline
\textsf{\textbf{Care Robots}} & \textsf{Prevention of lower back pain, reduction of staff burden [1]} & \textsf{High adoption costs, complexity of operation [2]} \\ \hline
\end{array}
$$

Chapter 7: Education and Human Resource Development: The Truth and Fiction of Human Capital Strategy in the AI Era

It is 'people' who support the social penetration of AI, and the digitization of educational settings and large-scale reskilling are essential prerequisites [1]. The 'GIGA School Concept' promoted by the government aims for an environment where all students can use AI on a daily basis [1]. However, the reality of educational settings is complex, and there are scattered voices criticizing generative AI as a 'hindrance to learning' or a 'cheating tool' [4].

Excerpt: In the field, there are cases where generative AI is criticized as a 'tool used for cheating such as writing reports' or 'useless in the first place.' [5]

The biggest challenge in educational settings is the busyness and literacy gap of teachers, who are the ones using AI [5]. While the usage rate among teachers remains at 19.3%, it reaches 30.3% among students, showing a significant gap between instructors and learners [5]. The development of guidelines to overcome disadvantages such as dealing with hallucinations (AI lies) and the loss of autonomy is still underway, and it cannot be denied that the introduction of technology is causing confusion in the field [4].

Excerpt: Generally, younger generations tend to have higher usage rates of generative AI, but in the case of university faculty, they are beginning to incorporate AI not only into education but also into research activities. Disadvantages of introducing generative AI into schools/educational settings: loss of autonomy, hallucination problems, and unclear thought processes. [4]

Regarding reskilling support for existing workers, flaws in system design are causing serious problems [6]. In the 'reskilling subsidy' fraudulent receipt case uncovered between 2025 and 2026, a return order for 1.9 billion yen is expected to be issued to 178 companies nationwide, and organized fraud by specific consultants has been pointed out [6]. Such scandals not only undermine the credibility of human capital investment but also risk stalling support for workers who truly need to acquire skills.

Excerpt: January 18, 2026, [178 companies, total return amount of 1.9 billion yen] Fraudulent receipt of 'reskilling subsidies' involving specific consultants, heading toward nationwide return orders. [6]

Regarding the mindset of AI utilization in Japanese companies, a serious 'lack of sense of effectiveness' has been reported [3]. According to a PwC survey, only 13% of Japanese companies responded that the effect of AI was 'better than expected,' which is a desperate divergence from the 86% average of five countries [3]. This is because many Japanese companies view AI not as the 'core of business transformation' but merely as a 'tool for internal efficiency,' and this low level of purpose hinders AI investment from contributing to productivity improvement [3].

Excerpt: In Japan, 57% of companies view generative AI primarily as an internal tool for 'improving operational efficiency' or 'solving personal/peripheral problems.' Japanese companies need to redefine generative AI not just as a means of efficiency, but as the core of business transformation. [3]

Chapter 8: Ensuring Safety and Reliability: Current Status and Limitations of AI Governance

In promoting the social implementation of AI, ensuring safety and trustworthiness is not just an ethical requirement but an essential condition as infrastructure supporting economic activity [1]. Japan is taking technical and institutional measures against risks such as the spread of misinformation, copyright infringement, and cyberattacks [1]. In particular, the establishment of the 'AI Safety Institute (AISI)' plays an important role in formulating international safety standards.

Excerpt: The AI Safety Institute (AISI) is an organization for examining and promoting evaluation methods and standards for AI safety, aiming to realize safe, secure, and trustworthy AI. [13]

The 'Hiroshima AI Process' led by Japan demonstrated a certain level of leadership in forming international rules regarding generative AI [1]. However, while attempting to harmonize regulations across countries, it is unclear whether a guideline-based approach without legal binding force can sufficiently suppress the risks of rapidly evolving AI technology. Strengthening effectiveness in governance is a challenge to prevent technology monopoly and inappropriate use by specific countries [1].

Excerpt: Leadership in governance: Leading the Hiroshima AI Process and exerting influence in the formation of international rules. Harmonizing AI regulations across countries. [4]

In terms of securing computational resources, the Ministry of Economy, Trade and Industry is leading the 'GENIAC' project, which provides support for GPU cloud and other services. While this supports foundation model development by startups and academia, the current situation of relying on specific US vendors for the majority of computational resources contains inherent vulnerabilities in economic security [1]. How Japan can sustainably maintain and expand its own computational infrastructure, including the development of the next-generation flagship system (the successor to 'Fugaku'), will determine Japan's AI sovereignty.

Excerpt: Overview of 'GENIAC'. Securing computational resources is a major challenge in developing foundation models. Development is supported across the first, second, and third phases. Commenced development and preparation of a new flagship system with superior AI performance that will be the next generation of 'Fugaku'. [9][16]

Chapter 9: Economic Effects of AI Adoption and the Reality of Return on Investment (ROI) in Enterprises

AI adoption has entered a stage where concrete return on investment (ROI) is required in corporate management [1]. The government's roadmap recommends adoption for specific tasks such as AI-OCR and chatbots, and these are expected to see a return on investment in a relatively short period (12–18 months) [1]. However, the reality is that the introduction of these tools is merely 'point' efficiency and has not led to the 'surface' effect of improving overall corporate productivity [3].

Excerpt: Costs for introducing AI to specific tasks: AI-OCR (2 million to 5 million yen), investment recovery period (12–18 months), annual cost reduction effect (2 million to 5 million yen). [4]

Standard steps for companies to succeed in AI adoption are presented, including current status analysis, PoC implementation, system construction, and effect measurement [1]. However, as pointed out by PwC, in many Japanese companies, the focus of AI is on substituting existing tasks (cost reduction), and its contribution to creating new value (revenue expansion) is insufficient [3]. The lack of a perspective to place AI at the 'core of business transformation' and reconstruct the business model itself is leading to a low sense of effectiveness at 13% [3].

Excerpt: Roadmap for companies to advance generative AI adoption. Adoption Step 1: Current status analysis and goal setting (2–4 weeks). Due to a lack of awareness among Japanese companies, AI is treated merely as a tool, and it has not led to a fundamental transformation of business models. [4][3]

$$
\small \def\arraystretch{1.5}
\begin{array}{|l|l|l|} \hline
\textsf{\textbf{AI Adoption Steps}} & \textsf{\textbf{Recommended Period}} & \textsf{\textbf{Main Activities}} \\ \hline
\textsf{\textbf{Current Status Analysis/Goal Setting}} & \textsf{2–4 weeks} & \textsf{Issue identification, KPI formulation [1]} \\ \hline
\textsf{\textbf{PoC (Proof of Concept) Implementation}} & \textsf{3–6 weeks} & \textsf{Verification of technical feasibility [1]} \\ \hline
\textsf{\textbf{System Construction/Deployment}} & \textsf{6–12 weeks} & \textsf{Application from limited tasks, awareness [1]} \\ \hline
\textsf{\textbf{Continuous Improvement}} & \textsf{Continuous} & \textsf{Data tuning, scope expansion [1]} \\ \hline
\end{array}
$$

Although the 'democratization of AI' with small and medium-sized enterprises as the main players is being advocated, the lack of specialized knowledge and the burden of initial costs remain major barriers [1]. Although government measures such as dispatching experts, free consulting, and reskilling subsidies are available, the number of companies that can use these to achieve results is limited [7]. There are not a few cases where the introduction of technology itself becomes the goal, deviating from the resolution of original management issues [3].

Excerpt: The proposed 'winning strategy' is the democratization of AI with small and medium-sized enterprises as the main players. You can reduce consulting costs that are usually equivalent to 500,000 to 1 million yen. Reskilling subsidy: Up to 300,000 yen per employee. [5][7]

Chapter 10: Conclusion: Proposals for Realizing a 'Growth-Oriented Economy' in the 2030s and Summary of Technical Criticism

For the Japanese economy to escape the stagnation of the 'lost 30 years' and establish a sustainable 'new growth-oriented economy' in the 2030s, a fundamental review of AI strategy and the acceleration of the implementation process are necessary. Based on the analysis in this report, I point out the following three points as core issues.

First is the dramatic strengthening of investment scale and speed. Compared to the multi-trillion yen gap with the US and China, Japan's investment target of 12 trillion yen over 10 years, while contributing to the maintenance of foundational technology, is insufficient to become a winner in the global market. Especially in technologies involving the physical world, such as 'Physical AI', it should be recognized that the depth of capital determines implementation accuracy and speed [2].

Excerpt: While Japan maintains its strengths in precision machinery, precision control, and component technology, the structure of falling behind the US and China, which continue to develop against a backdrop of abundant funds, is becoming clearer. Against the massive risk capital of the US and China, Japan's investment amount may be insufficient. [3]

Second is breaking the rigidity of systems and organizations. Japanese-style employment practices and the culture of emphasizing OJT are incompatible with the disruptive technical characteristics of AI. We must deeply reflect on the lack of governance seen in cases such as the reskilling subsidy fraud, and urgently design effective laws and systems that allow workers to transition skills without suffering disadvantages and promote labor mobility that puts the right person in the right place [1].

Excerpt: Traditional Japanese-style employment practices such as lifetime employment and seniority-based promotion are factors that hinder flexible job placement and skill updating, and delay the fundamental reform of business processes accompanying AI adoption. Subsidy fraud is a serious problem that undermines the credibility and effectiveness of the system. [2][6]

Third is a shift in mindset toward 'business transformation'. We need leadership that changes the current attitude of viewing AI merely as a tool for efficiency (13% sense of effectiveness) and redesigns business models, customer experiences, and industrial structures themselves with AI as a premise [3]. Technology is merely a means, and the lack of a 'question' regarding what value to provide to society using it is Japan's greatest weakness at present.
Excerpt: Japanese companies need to redefine generative AI not just as a means of efficiency, but as the core of business transformation. The lack of a sense of purpose is making the effects of AI investment limited. [3]

The 2030s are the 'last chance just before the young population drops sharply' [1]. The form of a 'growth-oriented economy' where AI and humans coexist and continue to create new added value does not exist on the extension of the present. Having the government and private sector share risks and having the courage to discard existing successful experiences to complete the strategy is the only way to make the 'counter-offensive' of the Japanese economy real.

Excerpt: The 2030s are positioned as the 'last chance just before the young population drops sharply', and we will overcome labor shortages and lead to growth through fundamental strengthening of measures against the declining birthrate, labor-saving investment, and utilization of foreign human resources. This is the path to maximizing the potential that Japan possesses. [1][5]

References

1. AI, Japanese Economy, and Technology Strategy Analysis

2. AI Robotics Study Group Reference Materials - Ministry of Economy, Trade and Industry, accessed February 22, 2026, https://www.meti.go.jp/shingikai/mono_info_service/ai_robotics/pdf/20251008_2.pdf

3. 2025, Deciphering the next steps from the essence of the breakthrough in physical AI and general-purpose robots..., accessed February 22, 2026, https://www.pwc.com/jp/ja/knowledge/column/physical-ai-robotics/general-robot-progress-future.html

4. Introducing the merits, demerits, and case studies of introducing generative AI into schools and educational settings | Think with Magazine, accessed February 22, 2026, https://www.kddimatomete.com/magazine/250829000020/

5. The reality of educational settings where there are still many generative AI 'naysayers'; can it solve the 'too busy' problem for teachers?, accessed February 22, 2026, https://www.sbbit.jp/article/cont1/154296

6. [January 13 Latest Information] Fraud case involving 1.9 billion yen in reskilling subsidies, is January 19 the final deadline to avoid 'company name disclosure'? | Press release from Monolith Law Office - PR TIMES, accessed February 22, 2026, https://prtimes.jp/main/html/rd/p/000000027.000053149.html

7. From 14th in the world to a counter-offensive | Japan's AI strategy and the 3 goals indicated by the 'Artificial Intelligence Basic Plan 2025', accessed February 22, 2026, https://note.com/kakeyang/n/ne62bc3592070