The Goalposts on Wheels

In 1956, the pioneers of AI thought "human-level" intelligence was a summer project away. In the 1990s, we thought it meant beating a Grandmaster at Chess. In 2023, we thought it meant passing the Bar Exam.

Every time AI clears a hurdle that we previously thought required "true" human intelligence, we do something curious: we move the goalposts.

We say, "Well, that’s just pattern matching," or "It’s just a stochastic parrot." This phenomenon is known as Tesler’s Theorem: "AI is whatever hasn't been done yet."

But if the definition is always moving, how do we know when we’ve actually arrived? And more importantly, why does the definition keep shifting?

1. The "Competence vs. Comprehension" Trap

The biggest reason the definition moves is that we confuse doing with being.

• Competence: The ability to achieve a goal (e.g., writing a legal brief, diagnosing a disease, or coding an app).

• Comprehension: The internal "feeling" or "understanding" of what those things mean.

Because AI can now achieve high levels of competence without the comprehension we associate with humans, we feel a sense of "uncanny valley" disappointment. We realize that "human-level" isn't just about the output; it’s about the process.

2. The "Generalization" Gap

Most AI today is "Narrow" or "Broad," but not yet "General."

A human can learn to drive a car, then use that spatial awareness to play a video game, and then use that same logic to organize a warehouse. Current AI models are getting better at "cross-domain" tasks, but they still lack the "common sense" fluid intelligence that allows a toddler to understand gravity, social cues, and language all at once.

The Shift: We used to define human-level AI by depth (being the best at one hard thing). Now, we define it by breadth (being "okay" at everything).

3. The "Agency" Requirement

Perhaps the most significant shift in the definition is the move from Tools to Agents.

• Old Definition: A system that answers questions (The Oracle).

• New Definition: A system that sets goals, plans steps, and executes them in the real world (The Agent).

We are realizing that "Human-Level" implies Agency—the ability to say "I should do X to achieve Y," rather than just waiting for a prompt.

4. Why the Moving Goalpost is Actually a Good Thing

It’s easy to see the shifting definition as "moving the goalposts" to avoid admitting AI is winning. But there’s a more optimistic view: AI is acting as a mirror.

Every time AI masters a task, it forces us to refine what makes us uniquely human.

• If AI can write a standard email, then "writing standard emails" isn't the core of human intelligence.

• If AI can analyze a spreadsheet, then "data processing" isn't our unique value.

AI is stripping away the "mechanical" parts of our intellect, leaving behind the parts that truly matter: Empathy, Moral Judgment, Strategic Intuition, and Creative Synthesis.

What This Means for Leaders

Don't get bogged down in the philosophical debate of whether an AI is "truly" intelligent. Instead, focus on the Utility Threshold:

• Stop asking: "Is it human-level?"

• Start asking: "Is it 'Colleague-Level' for this specific workflow?"

If an AI can perform a task at the level of a competent human employee, the philosophical definition of "intelligence" is irrelevant to your bottom line. The goal isn't to build a "human"; it's to augment human potential.

The Bottom Line

The definition of human-level AI moves because we move. As our tools get smarter, our expectations rise. We aren't just building smarter machines; we are discovering what it actually means to be a human in an age of abundance.

Stay exponential,

Dr. Agus Budiyono
Decoding Innovation