2404 김병조

[2404 김병조]

Choosing the Question, Choosing a Position

Every project begins with a choice that nobody sees: the choice of the question. As the leader of a five-member team in our creative convergence inquiry, I had to make that choice before we wrote a single line of the report. Our topic was how artificial intelligence decides in life-or-death situations, and the question we finally chose was “Who Does AI Choose at the Crossroads of Life and Death?” I chose not to ask which AI performs best but to ask whom an AI should save when lives are at risk. A performance ranking would have been easier to produce and easier to defend, but it would not have told us anything about responsibility.

This choice came directly from my career goal. I want to become an AI-integrated aviation specialist who develops autonomous flight systems for disaster rescue and infrastructure inspection. A rescue drone with limited battery and limited time may have to decide which area to search first, and that decision affects who is found in time. An engineer who ignores this question builds not only a faster machine but also a machine that decides without accountability. Since I intend to design such systems myself, I decided that I should study the ethics before the engineering, not after it.

Our research examined ethical decision-making algorithms in systems where several large language models work together, and it also examined strategies for human intervention. We compared the ethical decisions of four large language models and read regulatory and risk-management documents such as the EU AI Act of 2024 and the NIST AI Risk Management Framework of 2023. We also referred to two books on AI ethics and to several recent research papers on AI decision-making. Both the answers of the models and the reasons behind them were recorded, because a conclusion without its reasoning cannot be checked by anyone else. Whichever framework we applied, our final question stayed the same: when should a human step in? This question mattered to me because a system that makes a decision without a clear moment for human review leaves no one able to correct it, and in a rescue mission the cost of an uncorrected error is measured in lives.

As the leader, I planned the research, organized the team’s work, and wrote the report, and our work was supervised by our teacher. The results were presented as a formal report, a twelve-slide presentation, and posters. Writing for readers who had not been part of our discussions taught me that an ethical argument is only useful when its steps are visible. If a reader cannot see why we chose a question and what evidence supports an answer, the argument cannot be checked or corrected.

A second choice came from my Korean reading-and-writing assessment, where I had to write a critical essay about the arrival of the AGI era for a general audience. I took a supporting position. The choice was not easy to justify, because the same technology that could solve problems beyond human ability could also make decisions that no one can explain. I used Park Taewoong’s AI Lecture 2026 as evidence and organized my essay around the features of AGI, my own eval‎uation, and a standard of judgment. Whatever benefits AGI brings, I argue that they should come with conditions under which a human can step in. The support I chose was not unconditional approval but a position that others could examine.

Looking back, I learned that choosing a question is already an ethical act. The question decides what evidence we collect, which numbers we trust, and which people we remember to include. Neither speed nor accuracy alone is enough to justify a decision about human life, and I now eval‎uate autonomous systems with that standard in mind. I also learned that a position becomes stronger when its conditions are stated openly, because conditions show where an argument can be tested. In the same way, a drone’s flight log is valuable not because it is long but because it lets others reconstruct what happened.

I plan to apply these lessons in my future career. Before I design a flight controller or a path-planning algorithm, I will ask who is affected by its decisions and when a human must be able to override it. Whoever designs an autonomous system must decide in advance when a human can override it, and I want to be one of those designers. That is why I will keep studying drone control, computer vision, and AI ethics together instead of treating them as separate subjects.

My choices so far were small ones: a project question, a reading position, a way of recording evidence. Yet I believe that engineers who learn to make such choices carefully will make better ones when the stakes become real.