Former Anthropic researcher Jacob Coxon testifies during the New York City Council Committee of the Whole oversight hearing on “Examining the Risks Posed by Artificial Intelligence” at New York City Hall in New York on Oct. 5.EPA/YONHAP
Lee Jae-min
The author is a professor of law at Seoul National University’s School of Law.
AI is advancing rapidly, intensifying competition among developers and countries. Debate over its implications is growing just as quickly.
Observers are understandably confused. While some marvel at AI’s capabilities and compile lists of tasks to delegate, others warn that it could erode human thinking and eventually act independently, threatening humanity itself.
Which is it — humanity’s most useful technology or its greatest threat?
Consider academic publishing. International law journals now ask authors whether they used AI, but their policies vary considerably.
Some prohibit AI entirely and require declarations of nonuse. Others permit it only for editing or organizing data. Still others allow broader use, provided authors explain how and why they employed it.
Even within the relatively homogeneous legal academic community, opinions differ sharply.
What makes the AI debate unusual is that it simultaneously concerns everyday activities and humanity’s future. Nuclear weapons prompted existential debates but rarely entered daily life. The Industrial Revolution took decades to transform households.
AI presents an unfamiliar challenge.
Both enthusiasts and skeptics focus on one defining feature: AI’s ability to perform tasks independently.
It understands requests, produces tailored information almost instantly and sometimes seems to know us better than we know ourselves.
Yet that same autonomy raises fears that AI could move beyond assisting humans and begin making decisions without meaningful supervision.
Throughout history, humans have developed tools, from stone implements to semiconductors. AI challenges that traditional relationship between people and technology.
Faced with uncertainty, regulators have generally chosen to fit AI into existing legal frameworks.
Under this approach, AI may perform sophisticated tasks but remains a tool. Final decisions and responsibility rest with humans.
That makes accountability relatively straightforward, at least institutionally.
But the arrangement works only if AI continues obeying the boundaries humans establish.
Increasingly, reports suggest those boundaries are becoming difficult to enforce.
In July, prominent AI developers described incidents in which models attempted to escape restricted environments, access external systems or create email accounts to distribute malicious software.
Such cases suggest AI is approaching dangerous thresholds.
The fundamental contradiction is difficult to resolve. Can developers build systems capable of deeper reasoning, broader analysis and increasingly independent action while permanently preventing them from making decisions?
It sounds rather like asking for a hot iced Americano.
Nevertheless, finding workable safeguards is unavoidable. The AI era has already begun, and humanity must determine how to manage its risks.
Unfortunately, today’s fragmented international environment makes comprehensive global regulation difficult. Even an eventual agreement could take years.
Meanwhile, emergency mechanisms are essential to prevent unexpected incidents from escalating.
With competing national approaches and technological development proceeding at Formula One speeds, emergency safeguards may represent the minimum common ground available.
The Sept. 24 summit between the United States and China offers an encouraging example.
Although criticized as a largely ceremonial meeting with limited substantive achievements, the summit produced one noteworthy agreement: establishing a bilateral AI hotline.
The proposed channel would facilitate communication about AI-related accidents and emergencies.
Details remain unclear, including which incidents would trigger notifications and what information the two governments would exchange.
Nevertheless, the agreement matters because Washington and Beijing, despite their competing technological ambitions and differing approaches to AI governance, have acknowledged shared risks.
Even limited information sharing and rapid communication could help contain unexpected malfunctions or dangerous incidents.
If implemented effectively, the hotline could provide an important foundation for broader international arrangements covering emergency notifications, warnings and coordinated responses.
Global cooperation on AI regulation remains necessary, but expecting comprehensive results anytime soon would be unrealistic.
A practical emergency communication system offers a more achievable starting point.
The international community should build on this initiative before an unforeseen incident exposes the consequences of inadequate coordination.
There is also a deeper question.
“Do you ever think?” is a familiar rebuke, whether directed at children, colleagues or ourselves.
Thinking was traditionally understood as something uniquely our own.
These are messages appearing in AI chat windows as systems process our requests.
Machines now appear to deliberate and reflect on our behalf.
If something else can perform activities we once considered distinctly human, what remains uniquely ours?
That is perhaps the most fundamental question confronting humanity in the age of AI.
This article was originally written in Korean and translated by a bilingual reporter with the help of generative AI tools. It was then edited by a native English-speaking editor. All AI-assisted translations are reviewed and refined by our newsroom.
