"Should I quit? Can you find me a company to move to?" Humans hand over even thinking and judgment to AI [AI Sapiens: Questioning Intellectual Sovereignty 2]

[Financial News]#. A salaried worker in his 30s, identified as A, was preparing to change jobs and entered job postings and his own career history into artificial intelligence (AI). When he asked which company would suit him best, AI ranked the companies based on salary, commute time, and career development potential. It also generated self-introduction sentences and likely interview questions.

A used the AI's answers as a reference and wrote his application. But at some point, he realized that he was no longer organizing what kind of work he wanted on his own and was instead judging companies according to criteria created by AI.

In the past, A would open several job sites and compare working conditions and job descriptions directly. Now he copies job postings all at once into AI and asks, "Pick the one that suits me best." He said, "The time I spend searching for information has gone down, but so has the time I spend thinking about what I value most."

AI does not stop at listing search results. It filters information and orders it according to the user's question. It also writes sentences for users in writing and data organization tasks, and explains the reasons for choosing one opinion over others. Users may end up only revising the content after reading AI's response.

The way people search for information has already changed. In the past, users entered search terms, opened multiple sites, compared materials, and selected what they needed. With generative AI, they can enter a question once and receive an answer that summarizes multiple sources. In the search process, AI handles some of the selection and organization that users previously had to do themselves.

This shift can also lead to a question of who controls information discovery. A structure may emerge in which users accept AI-selected answers first rather than comparing multiple sources themselves.
The Korea Information Society Development Institute (KISDI) explained in "Changes in Regulatory Issues in the AI Era and the Direction of Platform Self-Regulation" that generative AI-based search and recommendation services are moving away from listing multiple documents and toward providing a single or small number of answers selected and synthesized by AI. The report said generative AI "implies the possibility that the initiative in information selection and interpretation shifts from users to AI."
AI-generated answers are fast and easy to read. They pull out the key points from long reports and organize conflicting claims into tables. As the burden of finding and comparing materials decreases, the reason to keep using AI grows stronger.
When A chose a company to move to, he used the items organized by AI as they were. At first, he asked only for a comparison of salary and distance, but AI added workload, growth potential, and organizational culture to the evaluation criteria. He later realized that criteria he had never entered were included. Because the comparison table made by AI was convenient, he did not go back and review the criteria themselves.

When users ask AI to organize materials, they read the finished summary first. This saves time because they can see the key points and conclusion before reading the original text, but it becomes harder to check what was left out. That is because what AI deems important may differ from what the user should consider important.

B, a salaried worker in his 40s who was preparing meeting materials, put documents received from several departments into AI and asked it to summarize the common claims and differences. AI neatly grouped each department's position, but it shortened the background behind a problem repeatedly raised by one department. B read the original text again and added the missing details. He admitted, "If I only looked at the summary, all the departments' opinions seemed similar, but the original text showed that they were shifting responsibility to one another."
AI builds its answers by figuring out what the user is looking for. But if the user enters a wrong premise from the start, AI may continue answering without correcting it. Even if the question is biased or missing key conditions, the sentence itself can still be completed smoothly.

For that reason, generative AI answers are convenient, but they also come with risks if accepted as they are. Beyond the speed of receiving an answer, users need to check the source and context again.

The KISDI report said users may leave information search, processing, and decision-making to generative AI and skip source verification. It added that while generative AI saves time in information search, it also creates the risk that users will accept AI-generated content as the final truth.

Research also suggests that the more users trust generative AI, the less time they may spend thinking for themselves. Microsoft Research (MSR) researchers presented "The Impact of Generative AI on Critical Thinking: Reduced Cognitive Effort and Trust Effects in a Survey of Knowledge Workers" at the 2025 CHI Conference on Human Factors in Computing Systems.

The researchers surveyed 936 cases of generative AI use in work among 319 knowledge workers. The results showed that the higher the trust in AI, the lower the degree of critical thinking performed. By contrast, people with greater confidence in their own work ability said they engaged in more critical thinking.
In this study, critical thinking does not simply mean checking whether an answer is correct. It includes analyzing a problem, synthesizing multiple sources of information, and evaluating the results. A's effort to set priorities when choosing a company and B's effort to verify the background of each department's document also fall into this category.
The researchers explained that the content of critical thinking also changed after AI was introduced. Before using AI, the focus was on analyzing problems and creating solutions directly. After using AI, however, the main activities became verifying information, integrating answers, and managing tasks.

Reading a lot of information is not the same as thinking for oneself. Even if AI summarizes many sources, users can miss important conditions or opposing views if they do not check the original text. And even if they revise AI-generated conclusions several times, it is hard to apply them to the next problem if they have not set their own standards.

A later rewrote his job application by first listing the three working conditions he wanted. He entered those into AI to refine the wording and order, but he wrote the company recommendations and reasons for applying himself. He said, "Rather than asking AI for the answer, it was better to ask it to review the standards I had set."

An expert explained that AI can be a tool that lightens the workload, but if users hand over the thinking process as well, they may end up only checking the results. In an abstract for a presentation at Seoul National University's Institute for Future Strategy's "AI Coexistence Society Forum II," Professor Young Hoan Cho of the Department of Education at Seoul National University analyzed that "as AI reduces cognitive burden and delivers achievements that are difficult to obtain alone, dependence on AI is increasing." He warned that "the performance of human-AI systems does not always lead to deep human learning" and that "interpassivity can occur, in which people assign tasks to AI without thinking for themselves and are satisfied with the results."

[email protected] Han Seung-gon Reporter