In the world of technology, AGI (Artificial General Intelligence) is often called the “Holy Grail” of computer science. It sounds like science fiction — and that framing is the problem. Confusing AGI ambition with ANI capability leads organizations to misplace strategic bets.
What We Have Now: Narrow AI (ANI)
Most AI in production today — including the most impressive large language models — is Artificial Narrow Intelligence. ANI systems are specialists: high-powered pattern matchers designed for specific tasks such as generating text, recognising faces, or playing chess.
| Feature | Narrow AI (ANI) | General AI (AGI) |
|---|---|---|
| Scope | Limited to specific tasks | Broad — any intellectual task |
| Adaptability | Requires retraining for new tasks | Learns and adapts on the fly |
| Context | Struggles outside its training data | Understands nuance and abstraction |
| Examples | Siri, ChatGPT, Midjourney, AlphaGo | None |
What Would Define “General” Intelligence?
To qualify as general, an AI would need to demonstrate:
- Reasoning — logic under uncertainty, not pattern completion
- Planning — long-term goal-setting with self-directed steps
- Learning from experience — real-time adaptation from its own mistakes
- Common sense — the unwritten rules of the physical and social world
- Transfer learning — applying knowledge from one domain to a structurally different one
LLMs exhibit surface-level versions of some of these. They do not exhibit the underlying mechanisms.
Why AGI May Never Exist
The claim that AGI will never exist has serious philosophical and empirical support.
1. The Stochastic Parrot argument
Modern AI is sophisticated statistics — predicting the next most likely token based on massive datasets. Scaling data and compute does not transform a calculator into a conscious mind. There is no “progress toward AGI” on the current trajectory — only a better calculator.
2. Syntax versus semantics (the Chinese Room)
Philosopher John Searle argued that a machine can manipulate symbols according to rules (syntax) without ever understanding what those symbols mean (semantics). An AI that talks about “gravity” without experiencing it is not intelligent in any meaningful sense. If genuine understanding is required for AGI, digital code cannot produce it.
3. The embodiment problem
Researchers like Hubert Dreyfus argued that human intelligence is embodied — we learn by navigating physical and social reality. Without a body and a biological drive to survive, a machine lacks the contextual grounding that makes “general” intelligence possible.
4. The scaling wall
The internet has largely been consumed as training data. There is no clear next source of comparable scale. If progress requires exponentially more data that does not exist, the current path ends far short of AGI — not approaching it.
EA Does Not Participate — It Governs
Enterprise architecture does not evaluate AI by its marketing position on a notional AGI scale. EA evaluates capability against business need. The relevant question is never “is this AGI?” — it is “does this capability justify the architectural decision required to deploy it?”
ANI is real, deployable, and governable. AGI is a philosophical milestone whose reachability is genuinely in dispute. Treat them accordingly.
March 2026, Dusan Jovanovic