Chapter 4: Generative AI Risk Prevention, AI Ethics and Governance29%（当サイト設定） · 9 topics

Q1. Which is the most appropriate basic stance when using generative AI at work?

1. When AI is wrong, the service provider bears all responsibility
2. AI is an assistant; people bear final responsibility for whether to adopt output and for the results
3. Make AI-assisted work efficient by skipping human checks
4. AI output is based on vast data, so it always takes priority over human judgment

Q2. Which best describes the difference between search engines and generative AI?

1. Both work the same way and differ only in display format
2. Generative AI always searches the internet, so its evidence is more reliable
3. Search engines point to existing sources while generative AI writes new text, so checking evidence matters more
4. Search results are accurate, so only generative AI output needs checking

Q1. Fearing leaks, a company banned generative AI company-wide. Which best describes the "risk of not using AI"?

1. Internal data becomes more likely to leak
2. Missing chances for efficiency and competitiveness, while shadow AI may grow
3. Security updates for existing systems stop
4. Not using AI is itself a legal violation

Q2. Which is the most appropriate countermeasure against hallucination?

1. Have it answer with Temperature at maximum
2. Ask the same question many times and decide the right answer by majority vote
3. Check the facts, figures and sources in the output against primary sources
4. Have it answer at length, since longer output is more accurate

Q1. Which best describes shadow AI?

1. AI running automatically at night, out of people's sight
2. Embedding invisible watermarks in AI output
3. Employees using AI tools the company has not approved for work
4. A way of hiding model weights from outsiders to prevent leaks

Q2. Which best describes why shadow AI arises?

1. Because once AI tools are approved, usage is recorded and they automatically become unusable
2. Because approved tools disable all generative AI functions for security reasons
3. Because the law requires using AI at work and employees must provide their own tools
4. Because with no approved tool, a hard-to-use one, or a heavy request process, employees use tools close at hand

Q1. Which is the most appropriate concrete example of shadow AI?

1. Summarizing internal emails with an AI summary browser extension the company has not allowed
2. Summarizing an already-published press release with a company-approved generative AI tool
3. Using an approved translation tool in line with internal guidelines
4. Trying a tool IT is evaluating in a test environment, following procedures

Q2. If false content made by generative AI that lowers a specific person's social standing is published, what is most likely to be the issue?

1. Unauthorized access
2. Defamation
3. Design right infringement
4. Patent infringement

Q1. Which of the following is "personal information" under Japan's Act on the Protection of Personal Information?

1. A record with only an employee number that identifies the person when matched with the staff list
2. Visitor counts processed into statistics that cannot identify anyone
3. A corporate client's name, address and main phone number
4. A deceased customer's name and purchase history

Q2. Which is the most appropriate principle regarding the purpose of use of personal information?

1. Specify the purpose of use as far as possible and handle it within that scope
2. Only the internal staff in charge need to know the purpose of use
3. Acquired personal information can be used freely for any purpose
4. The purpose limitation no longer applies when entering it into generative AI

Q1. Which combination of factors is most important in judging whether copyright infringement occurred?

1. Publicity and registrability
2. Novelty and inventive step
3. Similarity and reliance
4. Usefulness and non-public status

Q2. Which is the most appropriate precaution when entering someone else's work into generative AI?

1. A work entered into generative AI becomes free for the user to use, even if the output resembles it
2. A work entered into generative AI is treated as a published work under the service's terms
3. A work entered into generative AI is considered out of copyright once used for training
4. Even if the input itself is no problem, if the output resembles that work and you use it, it can be infringement

Q1. Which is the most appropriate item included in the common guiding principles of the AI Guidelines for Business?

1. Human-centricity, safety, fairness, privacy protection, ensuring security, transparency, accountability
2. Total ban on use, domestic storage of training data, public release of models, caps on usage fees
3. Development speed, high performance, low cost, market share growth, promoting overseas expansion
4. Licensing of development, notification of use, fines for violations, business registration, regular state audits

Q2. Which best describes ELSI?

1. An indicator for estimating the economic effects of new technology before introduction (Economic Impact Index)
2. A scheme for certifying AI training data quality from a legal perspective (Legal Data Certification)
3. The ethical, legal and social issues raised by new technology (Ethical, Legal and Social Issues)
4. A test procedure set by international bodies to evaluate AI safety (Evaluation and Safety Index)

Q1. Which best describes AI governance?

1. Banning AI use in the organization uniformly with no exceptions to avoid risk
2. Government directly running AI service companies and centrally controlling AI use
3. Rules only developers must follow on how to write programs and evaluate models
4. An organization setting up and operating policies, structures and rules to manage risks of AI use and gain its benefits

Q2. A company that made internal generative AI rules wants to adopt agile governance. Which approach fits best?

1. Leave rule details to teams and set no company-wide policy
2. Write very detailed rules and handle change through exception procedures
3. Repeat design, operation, evaluation and review of the rules as conditions change
4. Have legal revise all rules together once a year

Q1. Which is the most appropriate statement about Japan's AI Promotion Act?

1. A law banning business use of generative AI except uses the state permits
2. A law giving AI legal personhood, liable with its developers for damage
3. A law promoting AI R&D and use, setting roles of the state, businesses and citizens, an AI Basic Plan and an AI Strategy HQ
4. A law licensing AI development, with heavy fines for unlicensed work

Q2. Which is the most appropriate description of the EU AI Act's approach?

1. A risk-based approach that classifies AI systems by level of risk and imposes obligations according to risk
2. Regulation favoring the region, allowing only companies based in the EU to develop and provide AI
3. Regulation imposing identical obligations uniformly on all AI systems regardless of risk
4. A policy of no legal regulation, leaving things to voluntary efforts so as not to hinder AI progress