Series Index
Part 1 | Attorney Hong Seung-kwon: “AI Arrives in the Legal Profession”
Part 2 | How Should Lawyers Use AI?
Part 3 | Dividing Roles Between AI and Lawyers
By Hong Seung-kwon
Principal Attorney, PAN Law Firm (skh@law-pan.com)
Many people assume that what lawyers do is “litigation.” That is not wrong, but it is not the whole story. Handling litigation well matters, but so does preventing disputes from arising in the first place. That is why, when a situation emerges that could carry legal risk, clients ask their lawyer questions—hoping that clear guidance will help them avoid future conflict.
So, what is the first thing a lawyer does after receiving a question? We look for the relevant statutes. Statutes are the strongest form of legal authority. Yet this is not as simple as it sounds. There are thousands of statutes currently in force, and each contains dozens—or even hundreds—of provisions. Once you add regulations and case law, the universe of materials to review expands exponentially.
In the past, lawyers would ask a senior colleague, comb through legal databases, or search for similar situations in annotated commentaries and textbooks. Naturally, this took a great deal of time. These days, however, I start by asking ChatGPT. If I type, “Here is the situation—what laws might apply?” it quickly offers a shortlist of statutes that could be relevant.
For example, if I ask, “If I try to collect a debt using a debtor’s personal information I happened to learn, what could become a legal issue?” it may flag potential problems under the Personal Information Protection Act and the Act on Fair Debt Collection Practices, among others. But the truly important work begins after that. I still have to verify, through the National Law Information Center, that the statutes ChatGPT lists are currently valid and in force.
Laws are amended frequently, and AI does not always reflect the latest revisions. In my experience, ChatGPT rarely cites a statute that does not exist at all. But when I follow the cited article number, I often find that the provision has been deleted or rewritten into something entirely different. That is why AI’s output should be treated as a clue—not a conclusion—and the final authority must be confirmed against the original text.
The next step is deciding whether the statute actually applies to the client’s situation. Legal provisions are written in abstract and inclusive terms, which makes it difficult to plug them directly into the facts of a specific case. For instance, if AI points to the “Personal Information Protection Act,” I must determine whether the information at issue qualifies as “personal information,” and whether the client’s conduct can be characterized as “leakage” or “disclosure” under the law. In the end, the key question is one of application: have the legal requirements been met in this case?
This kind of judgment needs clear support, which is why precedents matter. Precedents reflect how courts have applied the law, and if a dispute arises later, it is the court that will ultimately decide. I therefore examine whether there are decisions addressing similar fact patterns, reviewing even lower-court rulings carefully. Decisions that match the facts perfectly are rare. So I rely on the closest available precedents to identify what facts the court considered important and what reasoning led to its conclusion—then build a persuasive legal basis from there.
From this point onward, ChatGPT often becomes less helpful. When I check the cases or theories it offers as support, a nontrivial portion turns out not to exist. Of course, ChatGPT can produce its own conclusions and explanations, but compared with the work of constructing a genuinely persuasive argument grounded in verifiable authority, it still feels insufficient. Ultimately, accurate citation of case law and the design of legal reasoning remain the responsibility of humans.
After going through these steps, I can produce a position and supporting rationale refined into legal language in response to the client’s question. Sometimes it is enough to present that analysis as it is. At other times, however, I must offer a clearer recommendation—even if doing so involves assuming a degree of risk. In other words, rather than stopping at “The law says X,” I may have to say, “Given your situation, I recommend you do Y.” I believe this is the most important role a lawyer plays.
This is how I would summarize the respective roles of AI and lawyers based on my experience. AI’s role is undeniably important. Simply pointing to the starting line and narrowing down which statutes may apply can be tremendously helpful. Over time, AI may also become better at constructing persuasive legal reasoning. Even so, the task of making the final call—deciding what applies, how it applies, and taking responsibility for that judgment—will likely remain with lawyers. After all, a tool is only as responsible as the person who uses it.
