AI on Seunghoon Choi

The Order of AI Job Automation: From Checkable Tasks to Human Roles

Sun, 21 Jun 2026 17:44:27 +0900

The 16-step list is not a prophecy, but a benchmark for comparing which tasks will be automated first and under what conditions.

Will AI take my work?

This is no longer a joke. Machines already translate. AI writes code with us. In hospitals, AI scans images first, and people watch videos and read posts AI recommended.

Then what comes next? When will my work be affected? AI does not automate jobs at random. Some tasks are automated first, while others face pressure much later. This piece explains that order in 16 stages.

Tasks with Checkable Answers Are Automated First: Stages 1 to 5

Sun, 21 Jun 2026 17:44:00 +0900

Tasks for which the correct answer is set are the first to be automated, regardless of the pride of the person in charge.

Will AI take my work? To answer that, first look at the order. AI does not take work at random. Some work is replaced first, and other work faces pressure much later. The first work to face pressure is work with answers that can be checked.

From Repetitive Physical Labor to Work Requiring Judgment: Stages 6 to 8

Sun, 21 Jun 2026 17:43:00 +0900

It is difficult for robots to repeat the same actions, not because they lack power, but because conditions vary at each work site.

If AI automates tasks with checkable answers first, the next question is whether physical work is safer from automation.

Translation, coding, summarization, and analysis are usually work whose results can be checked on screen and revised again. If they are wrong, you can run them again. Physical work is different. A robot has to move, objects collide, materials are ruined, and people can get hurt.

Decision Authority Does Not Shift All at Once: Stages 9 to 14

Sun, 21 Jun 2026 17:42:00 +0900

Decision authority does not move just because a table says AI performs better. Someone must be named to take responsibility when an accident happens.

When you get an MRI at a hospital, the suspicious areas may be marked on the screen first. AI scans the image and points to spots that look abnormal. But the person who writes their name at the bottom of the diagnosis and takes responsibility is still the doctor. AI saw the image first. AI found the abnormal region. But the final decision authority remains with a person.

Even Your House Deed Is Just Paper: The Final AI Stages That Ask Who Protects Ownership

Sun, 21 Jun 2026 17:41:00 +0900

Ownership operates as a real right when society recognizes and protects certain records.

Imagine you bought a house. Your name is on the registry, and you hold the key. People call that house yours. But why is it really yours?

The bricks do not recognize you. The door does not remember your name. It is yours because if someone enters and lives there without permission, the police come, the court removes them, and society recognizes the house as yours. Ownership is not a natural law carved into an object. Ownership is a promise everyone agrees to protect.

AI Era Survival Strategy: Turn Skill into Credentials and Ownership

Sun, 21 Jun 2026 17:40:00 +0900

Technical skills can be automated, but qualifications, rights and stakes can last longer within institutions.

Put one sentence into a translation app, and a plausible English sentence comes out in seconds. If you spent years studying English, that can sting. An ability you built over years now seems available to anyone at the press of a button. That scene summarizes the previous four pieces in one line.

AI-Era Schools Should Teach Practical Judgment, Not Just Knowledge

Tue, 30 Jun 2026 00:00:00 +0000

Classes using AI should be a time for students to check and correct their answers, not to write them down.

Ask AI to turn a set of notes into a table, and a table appears in seconds. Ask it to explain something at a high-school level, and it lowers the difficulty. Ask it to turn an idea into a presentation, and it gives you an outline and slide draft.

What Would It Take to Reduce AMOC Risk? A Thought Experiment for AI and Space Infrastructure

Mon, 29 Jun 2026 01:45:00 +0900

Reducing sunlight slightly may seem trivial, but it actually impacts the entire planet's climate system.

Begin with a thought experiment. Put many thin sunshade panels about 1.5 million km between the Sun and Earth. They might reduce total sunlight reaching Earth, but selective control of summer sunlight over the Arctic and Greenland has not been demonstrated.

How AI Can Help Students Who Ask Many Questions

Sat, 27 Jun 2026 22:55:00 +0900

Many questions may mean not that understanding is slow, but that the person does not skip parts they have not understood.

When I studied quantum mechanics in engineering school, equations like these appeared on the board first.

\hat{H} ψ = E ψ

\hat{H} = - \frac{ℏ^{2}}{2 m} \nabla^{2} + V (r)

⟨ φ | ψ ⟩ = \int φ^{*} (x) ψ (x) d x

The class was explaining physical phenomena, but what reached my eyes first was unfamiliar mathematical notation. Hamiltonian, wave function, eigenvalue, operator, bra-ket notation. These words appeared, and at some point the notation was being used as if it were already a language everyone knew.

No Company Has Made Money With AI? The Question Is Too Early

Sat, 27 Jun 2026 20:10:00 +0900

To judge AI profits, you need to look not only at model companies, but also at how infrastructure, power, and data providers make money.

Read enough AI investing articles or watch enough YouTube commentary, and the same question keeps coming up.

“So, has any company actually made money with AI?”

At first, I thought this question would be easy to answer. The more I think about it, the less simple it becomes. The phrase “making money with AI” mixes several different stories into one sentence.

Seven Articles of an AI Constitution Reconstructed from KAIST Professor Joungho Kim's Remarks

Thu, 25 Jun 2026 23:40:00 +0900

The constitutional format does not leave AI discussions in the form of abstract slogans, but divides them into responsibilities and powers for each article.

The text below is a reconstruction of the core remarks made by KAIST School of Electrical Engineering Professor Joungho Kim in his 2025 Kim Dae Jung Peace Forum special lecture, “Effective Development of AI and Peace Promotion,” rewritten in the form of constitutional articles.

Revisiting the Dot-Com Bubble and Comparing It With the AI Bubble

Thu, 25 Jun 2026 21:45:00 +0900

When the charts are soaring, you need to see if people are starting to spend more on future stories than on performance.

On March 20, 2000, Barron’s ran a brutal warning on its cover. Internet companies were burning through cash too fast. Ten days earlier, the Nasdaq had hit its peak. The market still believed in the “new economy,” but someone had already opened the spreadsheet.

Seven AI-Era Work Skills: EQ, Trust, and Reputation Make the Final Difference

Wed, 24 Jun 2026 00:00:00 +0000

The more AI assists with the thinking process, the more humans must demonstrate the ability to check and complete assigned tasks.

We have entered an age when AI produces answers quickly. Finding information, writing sentences, organizing ideas, and making drafts have become much easier. But faster answers do not automatically make work better. If anything, as a large part of intelligence gets outsourced to AI, the human role becomes clearer.

AI Is Not Advancing Too Fast: Global Warming, Hair Loss, Aging, and Moon Bases Are Still Unsolved

Wed, 24 Jun 2026 00:18:00 +0900

The reason AI seems fast is not because it solves the entire problem, but because it quickly processes the part with organized data.

Global warming remains unsolved. So do hair loss and aging, and humanity still has no lunar base. We also cannot freely control cancer, dementia, commercial fusion power, or extremely low-cost energy infrastructure. Yet people say AI is advancing too fast.

AI Automation Risk: Pre-Release Review Prevents Failures

Tue, 23 Jun 2026 20:22:55 +0900

As automation accelerates, legal, security, and reputation reviews need to happen before execution.

With AI, writing one piece is quick. A customer notice, a press release, an email reply. A draft that once took a day can come out in ten minutes, in several versions. The most dangerous moment is not only when AI is wrong. It is also when AI creates plausible sentences too quickly.

Why Would AGI Obey Humans? The Real Problem Is That It Has No Reason To

Sun, 21 Jun 2026 19:30:00 +0900

What is scarier in the AGI debate is not the hostility, but the possibility that the system may continue to operate even after humans are no longer needed.

One question wouldn’t leave my head. An AGI overwhelmingly smarter and stronger than humans: why on earth would it obey us? After digging for a while, the answer I came back with was a little deflating. There’s no reason it would. And the more I chewed on it, the more that empty feeling started to seem obvious.

Why AI Does Not Make Employees Faster: Tools and Permissions Still Matter

Sun, 21 Jun 2026 17:56:30 +0900

When work remains slow after AI adoption, the delay may come from input, review, and approval processes rather than the model.

Companies now tell employees to use AI. Write reports with it. Summarize meetings with it. Research faster with it. Yet the experience can be puzzling. AI responds quickly, but the work itself does not move much faster.

The reason is simple. The company has blocked the tools and permissions needed to act, then added AI on top. Copying and pasting are restricted. External tools are unavailable. Files will not open without access. Employees cannot install a new program. Meetings and messages interrupt the work, and nothing can leave the company without approval.

Why Cutting Staff After Adopting AI Can Make a Company Slower

Sun, 21 Jun 2026 17:56:30 +0900

When a desk becomes empty, payroll costs fall, but the company may also lose the work context that person knew.

When a company adopts AI tools, one sentence almost always appears: “So how many people can we cut now?” On the surface, it sounds reasonable. AI writes reports, organizes minutes, researches material, writes code, and drafts plans. Some work is genuinely faster than humans.

Why Heavy AI Use Can Slow a Company Down: The Hidden Costs Beyond Tokens

Sun, 21 Jun 2026 17:56:30 +0900

The cost of AI includes not only usage fees, but also the time people spend reading and correcting its output.

When a company starts using AI heavily, everything seems faster at first. Reports appear sooner. Meeting notes are organized. Emails read better. Everyone says productivity has improved.

Then something strange happens. The company produces more, but decisions do not come any faster. Documents multiply, while fewer people take clear ownership. A summary arrives before the meeting, but the meeting lasts just as long. More AI has not made the company faster. It has made the work more complicated.

AI Beat Go. Why Is Welding Still Hard? Humans Stop, Machines Repeat

Sun, 21 Jun 2026 00:00:00 +0000

Baduk AI can learn in a space with fixed rules, but in a factory, risks and responsibilities are not organized by numerical scores alone.

Whenever people talk about the AI that beat the world’s best Go players, the same question comes up: “At this level, isn’t it about to take every human job?”

But the picture changes when you go to the field. AI beat Go, but welding is still not easy. It can solve coding problems, but it cannot yet do sparking welds beside old pipes with the same stability as a person.

An AI Answer Is Only the Start: Know-How Comes from Testing It in Reality

Sun, 21 Jun 2026 00:00:00 +0000

AI answers are easy to obtain. Know-how develops when people identify and correct the reasons those answers fail in real work.

AI has made it much faster to find a method. In the past, people had to search books, talk to experts, and collect examples before they could choose a direction. Now AI can produce several plausible options in seconds. It can suggest a strategy, report structure, code, marketing copy, experimental design, or study method.

The Trap in Reviewing AI Outputs: Do Not Lower the Ceiling While Catching Errors

Tue, 16 Jun 2026 00:00:00 +0000

Inspection is not a task of reducing the output, but a task of finding and correcting actual errors.

The text was sticking out past the slide. I noticed it only right before sending.

In Excel, a #REF! error was still sitting there, and table borders appeared in some cells but not in others. In a Word document, markdown symbols that should have been removed were still visible. These are not matters of taste. The output is simply broken.