The Ministry of Science and ICT (“MSIT”) has recently released a series of public notices and guidelines, along with the draft Enforcement Decree and Enforcement Rule of the AI Basic Act. The guidelines provide guidance on determining high-impact AI in each sector and key considerations to be taken into account.
In particular, AI systems in the recruitment sector, which are used for the purpose of making determinations or evaluations about individuals, may have a significant impact on fundamental rights, such as the freedom of occupational choice, and therefore require special caution in the use of AI. In this regard, we will examine below the concept of high-impact AI in the recruitment sector, as well as the legal obligations that high-impact AI business operators in the recruitment sector must comply with, and practical considerations that they need to keep in mind.
I. CONCEPT AND CRITERIA FOR DETERMINING HIGH-IMPACT AI IN THE RECRUITMENT SECTOR
A. Concept of High-Impact AI in the Recruitment Sector
The AI Basic Act (the “Act”) defines high-impact AI as “an AI system that (i) is likely to have a significant impact on, or pose a risk to, human life, physical safety, or fundamental rights,” and (ii) is utilized in one of eleven specified sectors (Article 2, Subparagraph 4). One of these eleven sectors is “recruitment,” and the area of its use is described below.
|
Items of Subparagraph 4 of Article 2 of the Act
|
Category
|
Area of Use
|
|
Item (g)
|
Recruitment
|
Determinations or evaluations that have a significant impact on individuals’ rights and obligations, such as recruitment and loan screening;
|
Here, “recruitment” refers to all activities conducted to secure the human resources required by an employer, and encompasses both recruitment (a series of activities to find talent required by the employer) and selection (activities to select the personnel required by the employer from among the applicants, evaluating whether the applicants possess the capabilities required by the employer and are suitable for the job).
B. Determination of Whether AI Systems Used in Recruitment Constitute High-Impact AI and Key Considerations
If an AI system is classified as high-impact AI, a business operator that develops such a system or provides products and/or services using high-impact AI will be subject to various responsibilities as described in Section II below. Therefore, a prior review under Article 33 of the Act is required to determine whether the system falls under the category of high-impact AI. In order not to be subject to the aforementioned responsibilities, it is preferable for its operation, etc. that the system not be classified as high-impact AI.
An AI system utilized in the recruitment sector automatically meets the sectoral requirement out of the two requirements to qualify as high-impact AI. Therefore, it is important to design the system so that it does not have a significant impact on or pose a risk to human life, physical safety, or fundamental rights. In order to determine this, the intended purpose, functions, and context of use of the system should be comprehensively considered from various perspectives.
In the recruitment sector, there is a potential risk that labor rights and the right to equality, including the freedom of occupational choice of individuals seeking to enter into or maintain an employment contract, may be infringed. Therefore, it is crucial to ensure that the outcomes generated by AI systems used for the purpose of making determinations or evaluations about individuals are free from discrimination or bias, and that substantial human intervention is possible so that applicants are not unfairly deprived of employment opportunities.
Accordingly, when determining whether an AI system used in the recruitment sector is classified as high-impact AI, (i) whether the outcomes generated by AI have an impact on whether to hire an applicant and (ii) whether there is substantial and meaningful human intervention in the decision-making process are examined. If there is no substantial intervention by a person with legitimate authority in the recruitment process—i.e., even if there is human intervention, if only formal procedures are operated by making a decision that heavily relies on or uses the outcomes generated by the AI as a material basis—then the AI system will be classified as “high-impact AI.”
For example, AI systems that fit this description include those that pose a high risk of restricting an individual’s autonomy, freedom of occupational choice, or right to equality, potentially leading to discriminatory outcomes, by providing an individual’s information to a company based on the AI’s own analysis and decision-making regardless of the intention of a specific applicant, or making hiring decisions solely based on the AI without review by the company. The examples below are illustrative only and cannot be applied uniformly. Whether an AI system constitutes high-impact AI is determined on a case-by-case basis by comprehensively considering the overall procedure and nature of individual recruitment, its specific characteristics, and the surrounding circumstances of the recruiting company.
|
No.
|
Details
|
|
1
|
Where an AI system profiles the information submitted by a job seeker, processes it by scoring or grading it, scores descriptive or essay-type answers during written evaluations, or utilizes biometric information in interviews and practical evaluations
|
|
2
|
Where an AI system finalizes and issues written evaluation questions without the job offerer’s review
|
|
3
|
Where candidate selections or hiring decisions are made solely based on the outcomes generated by an AI system throughout the entire recruitment process
|
II. RESPONSIBILITIES OF HIGH-IMPACT AI BUSINESS OPERATORS
A. Key Considerations for High-Impact AI Business Operators in the Recruitment Sector
If an AI system used in the recruitment sector is determined, as a result of a prior review under Article 33 of the Act, to constitute high-impact AI, a high-impact AI business operator will be subject to various responsibilities. For the avoidance of doubt, these responsibilities are borne by high-impact AI development business operators (e.g., companies that develop AI models for video and voice analysis) or high-impact AI utilization business operators (e.g., companies that provide AI interview systems). It should be noted, however, that (i) recruiting companies that select and evaluate applicants using the AI interview systems provided by AI utilization business operators or (ii) applicants participating in the recruitment process through such systems are the users of or parties affected by each AI service, and therefore are not subject to the aforementioned obligations.
1) Responsibilities of High-Impact AI Business Operators in the Recruitment Sector
When providing high-impact AI or products or services utilizing high-impact AI, measures including the following must be taken to ensure the safety of high-impact AI (Article 34 of the Act):
- Establishment and operation of risk management plan
- Establishment and implementation of explanation plan, to the extent technically feasible, regarding the final outcomes generated by AI, the key criteria applied in producing such outcomes, and an overview of the learning data used in developing and utilizing AI
- Establishment and operation of user protection measures
- Human oversight and supervision over high-impact AI
- Preparation and keeping of documents verifying the measures taken to ensure safety and reliability
- Other matters deliberated and resolved by the Committee to ensure the safety and reliability of high-impact AI
In addition, AI business operators are required to disclose on their websites, etc. the key details of their risk management plans, such as risk management policies and organizational systems, as well as their user protection measures (Article 26 of the Draft Enforcement Decree of the Act). (i) The risk management plan must include details on the establishment and implementation of risk management policies and organizational systems (Article 4 of the Draft Public Notice on Responsibilities of Business Operators (the “Public Notice”)). (ii) With respect to the explanation plan regarding the final outcomes generated by AI, the key criteria applied in producing such outcomes, transparency must be ensured to the extent possible, and such plan must be made available to users (Article 5 of the Public Notice). (iii) User protection measures during the development phase must include lawful and safe data collection and management, as well as the design of safe algorithms capable of defending against adversarial attacks, and user protection measures during the operational phase must include measures to guarantee users’ rights and compensate for damages, if any (Article 6 of the Public Notice).
2) AI Impact Assessment in the Recruitment Sector
When providing products or services using high-impact AI, efforts must be made to assess the impact of AI on individuals’ fundamental rights in advance (Article 35 of the Act). The AI impact assessment must include (i) the identification of relevant stakeholders and the types of fundamental rights, (ii) the social and economic impact of AI, (iii) the purpose and method of use, (iv) evaluation indicators and measurement methods, and (v) risk prevention and loss recovery measures, and improvement implementation plans.
When conducting an impact assessment, high-impact AI business operators in the recruitment sector need to carefully review the fairness of the criteria for evaluating applicants and whether the data is biased, given the particular nature of such systems.
Therefore, it is necessary to assume various scenarios, individually examine the level of individual impact (i.e., the details, scale, duration, and possibility of occurrence of relevant impacts, such as unfair employment results), identify management measures (such as allowing HR personnel to make a final confirmation and decision, or analyzing only information for which the applicant’s prior consent has been obtained or reasonably disclosed information), and supplement them if necessary.
3) Obligation to Ensure Transparency
Recruitment-related products and services utilizing high-impact AI must notify users in advance of the fact that the services, etc. are operated based on high-impact AI, unless it is evident from the product or service name, the user’s screen, or labeling on the product itself that the product or service operates based on high-impact AI (Article 31, Paragraphs 1 and 4 of the Act). In this case, such notice may be provided by directly indicating it on the product, including it in the relevant contract, or displaying it on the user’s screen or device, etc. (Article 22 of the Enforcement Decree of the Act). Accordingly, when offering recruitment-related products or services that utilize high-impact AI, it is necessary either to clearly indicate this in the service name or to notify users in advance that the service operates based on high-impact AI.
III. IMPLICATIONS
It is advisable for business operators providing AI in the recruitment sector to continuously monitor the potential for bias in AI-generated outcomes and to establish procedures for human intervention or verification processes in advance to minimize the direct impact of AI decision-making on personnel decisions.
[Korean version]
