Sam Altman recently reiterated his belief that artificial general intelligence (AGI) — human-like intelligence in machines — was within reach. Speaking to Bloomberg, the OpenAI CEO suggested that simply scaling up foundational models like ChatGPT would eventually get us there. “AGI will probably get developed during [Trump’s] term,” he predicted, continuing a theme OpenAI has championed since ChatGPT’s explosive debut in November 2022.

A 2023 OpenAI blog post reinforced this idea, claiming that “The first AGI will be just a point along the continuum of intelligence.” It’s a compelling vision — except for one fundamental flaw: intelligence, whatever it is, is almost certainly not on a continuum.

It’s understandable that a Silicon Valley insider with so much at stake would frame AI’s trajectory this way. But the view becomes difficult to defend when considering another kind of intelligence that AI has largely sidestepped since its inception: natural intelligence.

What about natural intelligence?

Sure, humans use language; thus conversational AI gives the impression of intelligence. But while ChatGPT manipulates symbols, we use language to communicate about the world around us. We operate in a physical, emotional, and social reality; AI models deal only in abstractions. This is why, despite fluency, an LLM hallucinates nonsense with supreme confidence — it has no actual connection to the world.

Researchers and entrepreneurs like Altman continue to make predictions about the future of AI that assume we know what path we’re on and, more implausibly, that we know we’re moving toward human-like intelligence. But intelligence itself remains a mystery. Psychologists have done little better than pointing to a factor, “fluid g,” a kind of flexible problem-solving ability that underlies cognitive tasks. Yet how it arises or what it is, in scientific terms, is an open question.

If an alien landed on Earth and took stock of intelligence in both humans and machines, it might be surprised to learn that AI research is flourishing, poised to reshape civilization — all while operating with a hand-wavy definition of intelligence itself. Surveying the animal kingdom would raise even more questions. That’s because, far from being something that simply “scales” with more data and compute, natural intelligence looks different across species.

What we can call ecological intelligence runs the gamut from fleas to elephants. If intelligence is simply the ability to navigate and survive in an environment, then the best candidate for AGI isn’t a billion-parameter neural network — it’s a fruit fly.

A fruit fly’s guide to navigation

A prominent subfield of AI, autonomous navigation, defines intelligence as anything that enables a vehicle to drive itself safely. These systems — reliant on deep neural networks trained through reinforcement learning — use sensor fusion (LIDAR, cameras, and radar) to recognize obstacles and plan routes. With enough training, the system can devise strategies for safely navigating even chaotic environments.

Yet, if this is our definition of AI, we should admire Drosophila melanogaster, the humble fruit fly.

Fruit flies, equipped with micro-brains that weigh next to nothing, have collision-avoidance reflexes superior to those of any self-driving car. In a fraction of a second, they detect motion cues, predict trajectories, and execute evasive maneuvers. Researchers have found that fruit flies calculate escape routes faster than high-end computer vision systems.

If intelligence is about motion planning and rapid decision-making, then fruit flies might deserve their own OpenAI funding round.

Hive minds and the illusion of distributed intelligence

Then there are bees. If you listen to AI enthusiasts, you might believe we’re on the verge of creating “hive-mind intelligence” through interconnected neural networks. But nature got there first.

Honeybees, operating as a colony, make decentralized decisions that rival the efficiency of algorithmic optimization. When scouting for a new hive location, they use collective voting procedures that involve waggle dances—a form of symbolic communication that encodes distance, direction, and quality of potential sites. Remarkably, these decisions emerge from thousands of simple interactions, without a central authority.

If intelligence is about collective problem-solving, then AI is still a primitive bee larva. Real-world swarm intelligence is efficient, adaptive, and astonishingly robust. AI’s version? A bunch of chatbots scraping Reddit.

Squirrels: Nature’s memory champions

Another domain where AI excels is data storage and retrieval. AI researchers often measure intelligence by the ability to store vast amounts of information and retrieve it when needed. Vector databases and embeddings allow language models to access massive amounts of data, drawing on terabytes of text.

But nature has already optimized this, too — meet the squirrel.

A squirrel doesn’t just hoard nuts; it remembers where it stored them, even months later. Studies have shown that squirrels engage in deceptive caching — pretending to hide nuts to mislead potential thieves. This ability to encode, store, and retrieve information efficiently — without requiring a multi-billion-dollar data center — is a remarkable feat of memory engineering.

And yet, no squirrel in history has ever hallucinated an imaginary acorn.

What’s the ‘I’ in AI, Really?

So what can AI learn from the animal kingdom? First, that intelligence is not a one-size-fits-all phenomenon. The way a fruit fly navigates, the way a bee colony decides, the way a squirrel remembers — each represents a distinct mode of cognition, shaped by evolutionary pressures.

Second, it forces us to question our assumptions. The dominant AI paradigm — scaling up neural networks — assumes that more data and compute will eventually lead to something akin to human intelligence. But if natural intelligence is a spectrum, then why assume that one path — data-driven statistical learning — will lead to AGI?

Real intelligence is embodied. It exists within a living system, interacting dynamically with an environment. AI, on the other hand, is an abstraction. It predicts text sequences, not causes and consequences.

So, is AGI within reach? Maybe. But if we’re taking cues from nature, we might be better off funding fruit fly research.