Amid the hype, a natural question arises: How will we know when we’ve actually reached AGI? What counts as true artificial general intelligence—and who decides?

This debate is part of the backstory, because the lack of a clear yardstick only fuels more public curiosity (and confusion). It’s like a race with an ever-shifting finish line.

Defining AGI: a moving target

In theory, AGI refers to an AI that can understand or learn any intellectual task a human can. But in practice, experts disagree on what that threshold should be.

Some insist that AGI must match human intelligence in all respects—demonstrating not only reasoning and learning, but also common sense and flexibility.

Others offer more pragmatic benchmarks. For example, Open AI defines AGI as achieved when AI can perform most economically valuable jobs as well as humans. 

Demis Hassabis sets the bar even higher. The co-founder and CEO of Google DeepMind, has argued that an AGI should be capable of making groundbreaking scientific discoveries—like independently arriving at the theory of general relativity.

In short, there’s no single scoreboard. Each lab, company, and researcher uses different criteria. This definitional fog actually drives interest in the topic. People frequently search “what is AGI” to try to make sense of it all.

Tests and milestones that signal progress

Over the years, researchers have proposed a range of AGI tests to help keep score. The most famous is the Turing Test, developed three quarters of a century ago by famed mathematician Alan Turing. The test’s benchmark is can a computer, or in this case an AI chat convincingly enough to fool a human into thinking it’s also human?

For decades, no AI came close. But by 2023–2025, large language models began creeping into that territory. In one 2025 study, GPT‑4.5 was judged more “human” in conversation than real people—scoring 73% compared to 67% for the human control group. If that result holds, it may represent the first time a model has passed a serious Turing-like benchmark.

That kind of breakthrough news spreads fast, and raises questions. Such as are we crossing a line?

When other such “breakthroughs” are reported, whether real or over-hyped, the public often ends up both intrigued and confused, which sends even more people searching for answers.

Who gets to declare AGI achieved?

This uncertainty has fueled a kind of race among companies, startups, and researchers to be the “first to AGI.”

Some organizations exist solely to pursue AGI, and occasionally announce that they are “80% of the way there” or post internal scoring updates. On AI forums and forecasting sites like Metaculus, thought leaders regularly publish AGI probability timelines. In late 2024, one forecasting community estimated a 25% chance of AGI by 2027—and 50% by 2035.

Predictions like these often get picked up by the media and passed around in tech circles. For the public, they act as a kind of countdown, even though no single group has the authority to declare “AGI Day.”

In truth, it’s likely we won’t experience AGI as a single moment. Instead, we’ll look back and realize that we crossed into the AGI era without noticing exactly when.

A shrinking list of human-only tasks

While there’s no agreed-upon finish line, many of the benchmarks once considered sci-fi have quietly become reality.

A decade ago, no AI could reliably drive a car, speak fluently and win complex games. Now those accomplishments are old news. AI has mastered chess, Go, StarCraft, and can navigate city streets.

To the public, it feels like a scoreboard is shrinking. The list of tasks once thought to require human intelligence keeps getting shorter. Each time something moves from the “human-only” column to the “AI can do this” column, the perception grows: we’re inching closer to AGI.

In 2023, an AI-generated image even won a prestigious photography competition—until it was revealed that no human camera was involved. The controversy sparked public debate about AI creativity and sent many searching: “Can AI really think?”

The public steps in as the scorekeeper

In the end, the absence of a clear finish line has only intensified interest. Every new “AGI test” passed (or nearly passed) adds tension to the narrative—a kind of horse race without a flag.

People want to know: Are we already in the AGI era? Are today’s systems “emerging AGI,” as a 2023 DeepMind paper suggested? What tasks remain before the line is crossed?

By trying to answer these questions, the public is effectively keeping score themselves, if the surging volume of AGI-related searches, headlines, and debates is any indication. The scoreboard may be blurry, but the eyes watching it are sharper than ever.