The "5 Basic Skills" Required of People as Generative AI Grows

—Policies are starting to move, and implementation is shifting to the field. What is changing, and what must we acquire?

The "experimental phase" of generative AI is over, and 2025 has become the year of "implementation and governance." In Europe, obligations for General Purpose AI (GPAI) under the AI Act will begin to apply on August 2025 (the European Commission announced August 1st, and the operational guide explicitly states "from August 2nd"), requiring transparency and safety explanations. Domestically, the **AI Safety Institute (AISI)** has been established, and the creation of evaluation methods has begun. As regulations and guidelines are put in place, what is required of "people" are the basic actions of posing questions, verifying evidence, following rules, managing processes, and ultimately explaining.

What is happening—The European AI Act is in full swing

The European Union (EU) AI Act entered into force on August 1, 2024. Among the phased applications, obligations regarding GPAI (General Purpose AI) will begin from August 2025, requiring businesses to explain model transparency, copyright compliance, and risk management. The European Commission's press release announced that "GPAI obligations are beginning to apply," and the separately published guidelines explicitly state the date as August 2nd. Furthermore, from August 2026, the European Commission's enforcement powers are also expected to be applied in stages. Full-scale application in high-risk areas will follow in later stages.

Background—The "scale" of both value and risk is significant

Estimates of economic value are bullish. McKinsey estimates the annual added value of generative AI at $2.6 to $4.4 trillion, with productivity boosts expected in customer service, marketing, and software development. On the other hand, the IMF points out that about 40% of global employment will be affected by AI. It analyzes that the more developed a country is, the greater the "way it is affected," making retraining and job redesign key. The magnitude of these numbers is the reason to rush toward evidence-based implementation, rather than hasty adoption or passive observation.

Domestic movements—Establishment of evaluation agencies and development of guidelines

In Japan, in February 2024, the AI Safety Institute (AISI) was established within the government/IPA to advance the creation of evaluation methods and standards. The Ministry of Economy, Trade and Industry and the Ministry of Internal Affairs and Communications published the "AI Business Guidelines (Version 1.0)," integrating existing development and utilization guidelines. International standard ISO/IEC 42001 (AI Management System) has also been issued, and the "quality management cycle" of policy → operation → continuous improvement is being fully introduced into the AI field.

Practical meaning: Japanese guidelines and international standards are frameworks that define "who is responsible," "where records are kept," and "when to review" in business terms. Rather than memorizing technical terms, deciding on personnel and procedures is the first step.

The "5 Basic Skills" now required of us

1) The ability to verbalize the core of an issue

AI is not a "jack-of-all-trades." If you throw requests at it while the purpose, conditions, and judgment criteria remain vague, it will return plausible but useless answers. What is needed is the habit of clarifying in writing beforehand what you want to decide, what the non-negotiable conditions are, and what criteria you will use to judge success or failure. This is a matter of work design, which comes before prompt engineering.

2) Ability to verify grounds (fact-checking)

Generative AI produces plausible-sounding errors (hallucinations). As a countermeasure, "Retrieval-Augmented Generation (RAG)," which grounds answers in external documents, is considered effective, but it is not a panacea. While some research shows that RAG improves accuracy, others point out that if the documents incorporated are biased, it can amplify errors. In short, citing sources and human final verification are indispensable for the time being.

3) Ability to translate rules into operations (copyright, privacy, standards)

The U.S. Copyright Office has clarified that "human authorship" is a requirement for registration. Since "parts automatically created by AI" may be excluded from protection, operations that record and declare human contribution are necessary. The European GPAI obligations also require explanations of training data and safety measures. The NIST AI Risk Management Framework and ISO/IEC 42001 are tools for translating these laws and principles into daily procedures.

4) Ability to manage processes ("Generate → Verify → Record")

The value of generative AI lies not in a "one-time great answer," but in reproducibility that consistently produces the same quality. (1) Generation (with grounds) → (2) Verification (counter-arguments/fact-checking) → (3) Formatting (to internal standards) → (4) Recording (traces of procedures/decisions). If you systematize this process, quality will not waver even if the person in charge changes. Outputs that can withstand evaluation and audits are born from this mundane cycle.

5) Ability to explain at the end (human responsibility)

"Because the AI said so" is not a valid reason. Decision-making must be performed by humans, and one is required to be able to summarize the "conclusion," "grounds," and "responses to opposing opinions" in their own words. This is not only for audits and customer explanations, but also a skill that accelerates consensus building within a team.

The trend of international cooperation—Safety is a "common challenge for all countries"

The 2023 Bletchley Declaration and the 2024 Seoul Declaration called for risk assessment of advanced AI and cooperation among safety evaluation institutions. From a corporate perspective, it is important to understand that a "common vocabulary regarding safety" is emerging across countries. The establishment of the domestic AISI is also on this international track.

Impact on the front lines—Being "fast" alone is not enough to pass

Speed is a major advantage of generative AI. However, without the three elements of identifiable sources, reproducibility using the same steps, and the ability to explain, it will become increasingly difficult to get deliverables and proposals approved in the future. European GPAI regulations have shifted toward demonstrating who learned what, how, and how safety was verified. Unless you switch to a design that is explainable from the start, explanation costs will balloon after implementation, ultimately slowing down speed—this is the reality of 2025.

Organizing the Issues: How to Deal with "Hallucinations"

Generative AI misgeneration will not reach zero even as models become more sophisticated. While multi-layered measures such as RAG and self-checking have been proposed and some improvements have been reported in research, the quality of the information incorporated and the method of evaluation can lead to counterproductive results. Based on the human habit of "always being skeptical," the current practical solution is to combine providing source links, third-party oversight, and documentation.

Industry Outlook: Where Value Comes From Is "Preparation" and "Retraining"

Where does the value come from? The economic impact of generative AI is considered to be significant in areas with "frequent back-and-forth of documents and decisions," such as customer support, marketing, and software development. However, that value depends on human and organizational preparation. If you introduce AI without changing how things are made (workflows), rework and explanation costs will increase, and you will not get the expected results. The IMF statement that "about 40% of jobs will be affected" indicates that retraining (reskilling) and job redesign will be the main battleground.

Future Focus: Moving Forward with a "Common Base" of Laws and Standards

NIST AI RMF (US) and ISO/IEC 42001 (international standard) serve as a common foundation across countries and industries. Domestically, AISI and the "AI Business Guidelines" are also being operated in a way that connects to these. Even if the details of the laws differ by country, the required behaviors are converging—this is the practical sense for 2025.

Conclusion—"Good Questions, Solid Evidence, Efficient Processes, and Final Accountability"

The smarter generative AI becomes, the more fundamental human actions become valuable.

• Good questions reduce uncertainty.

• Solid evidence stops errors.

• Efficient processes create reproducibility.

• Final accountability fulfills responsibility.

The systems have begun to move. All that remains is to quietly steer toward "explainable design" in the field.

References

• EU AI Act: Effective Date, Application Timing, and GPAI Obligations (European Commission Press, Guidelines, Official Journal) European Commission+2Digital Strategy+2

• Domestic: Establishment of AISI, AI Business Guidelines (Version 1.0)Cabinet Office Website+1

• International Standards and Frameworks: ISO/IEC 42001, NIST AI RMFISO+1

• Economic and Employment Impact: McKinsey value estimates, IMF 40% impactMcKinsey & Company+1

• International Cooperation on Safety: Bletchley Declaration, Seoul DeclarationGOV.UK+1

• Countermeasures for Hallucinations: Research on the effectiveness and limitations of RAG