AGI · Artificial General Intelligence
Artificial General Intelligence
What is AGI?
- AI that solves a wide range of problems
- Beyond today’s AI that does only specific tasks well
- Still a future concept
AGI is not a product you can buy today. It is a concept that unsettles the baseline for investment decisions. The moment someone in a management meeting says, 'They say AGI will do it all in two or three years, so why spend hundreds of millions of won on this system now?', the word stops being a philosophical term and becomes a budget item. So in practice, the first job is to shift the question from when AGI arrives to what counts as AGI.
The industry usually distinguishes narrow AI, AGI, and ASI (superintelligence), which surpasses humans across the board. In 2023, Google DeepMind researchers proposed viewing AGI not as an on/off switch but as a ladder of performance and generality. Performance at the top 50% of skilled adults counts as “competent,” 90% as “expert,” and 99% as “virtuoso,” and this is then multiplied by how broad a range of tasks that level covers. Seen through this framework, today’s generative AI is expert-level in certain areas but is generally judged to be still at an early stage on the generality axis.
A neighboring concept that is easy to confuse with AGI is the AI agent. An agent is an evolution in how AI executes work, calling tools and taking multiple steps within a given goal. It is not a broadening of intelligence. Even when an agent that reads email, schedules meetings, and submits approvals runs well, it is following a business path that people designed in advance. The real question in the AGI debate is whether AI can cross on its own into areas it was never taught.
The most common mix-up in the field is the impression that “it talks like a person, so it’s practically AGI.” Fluent language and solving unfamiliar tasks on its own are different abilities, and confusing the two leads to results pasted into reports without review, and then to wrong numbers. The opposite misconception is just as risky. If you put off data cleanup and work standardization because “it will do everything for us soon, so let’s wait,” you will have no data to feed a more powerful tool when it arrives, and you will start further behind. Treat AGI as a forecast, and spend your budget and training on today’s narrow AI. That is the safer course.
John KimInstructor John Kim, author of B2B Marketing Design and CEO of Nanum Consulting
Seolyeong ParkGenerative AI and AX training specialist with 17 years of hands-on experience and 200+ lectures, helping non-technical staff put AI to work right awayThe term became popular in the 2000s and refers to the goal of 'human-level general intelligence.' Its roots reach back to Alan Turing's question 'Can machines think?' (1950).
Consider a hypothetical case. Company A, an auto parts manufacturer with 340 employees and KRW 190 billion (≈ US$142,151,700) in annual revenue, planned to invest KRW 800 million (≈ US$598,500) in automated quality inspection in the second half of 2024. But at a board meeting someone objected: “They say AGI will arrive in 2–3 years and do everything, so isn’t what we’re buying now a sunk cost?” The proposal was put on hold twice. The corporate planning team spent three months answering that question.
- Director Kim of the Business Planning team asked 12 executives one by one what they thought AGI meant. The answers ranged from self-driving robots to chatbots, so the discussion was going nowhere.
- We brought in an outside instructor for a 90-minute session and, borrowing DeepMind’s level framework, aligned the executives on a shared view: “What we’re buying is a Level 1 inspection tool out of five.”
- When Mr. Lee, the executive director of production, pushed back with “But if something better comes out in three years, what happens to the KRW 800 million (≈ US$598,500)?”, Manager Kim answered, “We’ll structure the contract so we can swap it out then.”
- The purchasing team renegotiated with the supplier and added two clauses: one allowing the inspection model to be retrained or replaced every two years, and one giving Company A ownership of the 340,000 inspection images that accumulate.
- Six months later, the proposal passed as originally submitted. The defect escape rate fell from 0.8% to 0.3%, and four AI projects that had been shelved with a 'let's wait and see' were restarted as well.