[The Editors' Verdict] The Legitimacy of Regulatory Logic for AI Algorithms

Seong-Yeop Lee, Professor at Korea University Graduate School of Technology Management and Director of the Technology Law Policy Center.

From some time ago, internet platforms have been recommending not only news and movies that we might like but also products to shop for, and the technology enabling this is artificial intelligence (AI) algorithms. These are structured computational procedures designed to produce requested calculation values; they are sets of procedures, methods, and instructions to solve a particular problem. Simply put, it is a flowchart of the process that finds the result the user wants. Ultimately, the essence of an algorithm is a decision-making mechanism for organizational judgment.

In principle, a company’s decisions regarding its products and services are subject to the company’s autonomy. However, recently, the political sphere and government have proposed legislation mandating the disclosure and verification of algorithms. Starting from political suspicion about news recommendation algorithms on portals, this has expanded to include the disclosure of search and exposure criteria for transaction brokerage services, submission of algorithms by information and communication service providers to the government, and the disclosure of basic policies and specific criteria for article arrangement by internet news service operators.

The reason such legislation is necessary is to ensure the fairness and transparency of algorithms. Fairness is based on suspicions that algorithms discriminate against certain groups or make biased decisions. Bias can be categorized into intentional bias?such as the designer’s or operator’s bias reflected in the algorithm, or the operator arbitrarily manipulating recommendation results due to external pressure?and unintentional bias, which reflects biased patterns contained in the training data as is.

For companies, algorithms are optimization tools to provide satisfactory services to users, so there is no incentive to design algorithms in a biased way that would reduce user satisfaction. If a special purpose is incorporated to correct unintentional bias, this could lead to another fairness controversy. Moreover, it is very difficult to present fairness standards that all members of society can agree upon, and human subjectivity inevitably intervenes in algorithm design, making it hard to harmonize with fairness and objectivity from the outset. The most feasible approach is for companies to sufficiently explain the operating principles of their algorithms to users or partners.

Next is the demand for transparency. Originally, transparency is strongly required of entities performing public duties such as governments and public institutions. Since the public sector exercises power delegated by the private sector, it is necessary to realize popular sovereignty and prevent administrative corruption through information disclosure and citizen participation in administration. Although platforms are private companies, their increasing influence has raised the need for user control similar to that of the public sector, leading to demands for transparency legislation.

However, most corporate algorithms qualify as trade secrets. Therefore, disclosing them conflicts with the trade secret protection system, which legally safeguards trade secrets to promote corporate research and development activities. Also, since algorithms are frequently changed, the effectiveness of disclosing or verifying an algorithm at a specific point in time is questionable.

In conclusion, it can be acknowledged that external control is necessary for AI algorithms of platforms with enormous influence. However, whether preemptive legal regulation by the government is necessary as a methodology, or whether post-regulatory legal measures such as ensuring explainability and raising objections are sufficient, or whether self-regulation is possible, should be carefully decided considering the feasibility of regulation and the competitiveness of domestic tech companies.

Seong-Yeop Lee, Professor at Korea University Graduate School of Technology Management and Director of the Technology Law Policy Center