AGI in 2026? Musk believes it, researchers are skeptical

The definition no one agrees on

AGI—Artificial General Intelligence—refers to AI systems that can learn, reason and act broadly across domains, at roughly human level or better. It is fundamentally different from today's "narrow" AI systems that are specialized in single tasks.

The problem? No one agrees on exactly what qualifies as AGI. Is it passing an IQ test? Handling arbitrary new tasks without special training? Having "understanding" in some meaningful sense?

Anthropic CEO Dario Amodei prefers the term "powerful AI" over AGI, precisely because the term AGI is "overhyped."

"Humanity's Last Exam": The ultimate test

One concrete attempt to measure progress is "Humanity's Last Exam"—a benchmark designed to test AI systems' ability to perform complex reasoning across domains.

As a FinancialContent analysis describes it: "With every percentage point achieved... the question is no longer when AI will know everything, but when it will finally learn to think as well as the humans who created it."

Today's best systems—including OpenAI's o3 and Google's Gemini—show impressive progress on such tests. But critics point out that passing a test does not necessarily mean genuine understanding.

What does the research say?

Large surveys among AI researchers show considerable variation in expectations:

• Median estimate: a 50% probability of AGI around 2047
• Optimistic tenth: 10% believe it could happen as early as 2027
• Major disagreement: Many believe it will take significantly longer

Stanford HAI (Human-Centered AI Institute) recently published expert predictions for 2026. The focus is not on AGI per se, but on concrete trends: AI in the labour market, regulation and practical applications.

The Atlantic asks: "Do you feel the AGI yet?"

In a recent article, The Atlantic explores the paradox of the AGI debate. The leaders of both OpenAI and Anthropic have said that such systems "could come online by the end of 2026, perhaps bringing cancer cures or new bioweapons."

But what does it actually mean to "feel" the AGI? For most people the transition—if it comes—will probably be gradual rather than dramatic. One day AI will simply... work better. Much better.

The technical building blocks

The path toward more general systems requires progress on several fronts:

• Long-term memory: From "in-context learning" to true episodic and semantic memory
• World models: Multimodal systems that understand cause and effect
• Agent architecture: Planning, tool use and collaboration between AI systems
• Safety: Governance mechanisms that scale with capability

MIT Technology Review points to an interesting development: the combination of language models with evolutionary algorithms that check and improve the proposals. This is a step toward AI that can actually "think" iteratively.

A Norwegian perspective

For Norwegian businesses and institutions, the AGI debate is more than academic. The question is not just "when" but "what do we do now?"

Practical measures include:

• Building expertise in today's AI technology
• Following regulatory developments (EU AI Act, Norwegian guidelines)
• Experimenting with agent architectures in controlled environments
• Assessing security practices for AI systems

Conclusion: Do we really know?

The truth is that no one—not Musk, not OpenAI, not the Stanford researchers—knows for certain when or whether AGI will emerge. What we do know is that AI systems are becoming more capable at a pace that surprises even the experts.

As an AI Impacts analysis concludes: The results show "both accelerated expectations from 2022 to 2023 and high disagreement."

Perhaps the most important question is not "when will AGI arrive?" but "are we prepared for what is coming—whatever we call it?"

Sources: TradingKey, MIT Technology Review, The Atlantic, Stanford HAI, AI Impacts