We Classify Every Organism on Earth. Why Not AI?

---

The Map We Never Drew

Linnaeus gave us Homo sapiens. Darwin gave us the tree. For three centuries, naturalists have organized the bewildering diversity of life into a coherent map—kingdom, phylum, class, order, family, genus, species. We do this not for vanity but because classification is comprehension. You cannot study what you cannot name.

So consider the strange blindness of our current moment. We have created a new category of entity—systems that converse, reason, generate, and increasingly act in the world—and we describe them with the impoverished vocabulary of marketing departments. "GPT-4." "Claude 3.5 Sonnet." "Gemini Ultra." These are product SKUs, not names that reveal relationship, lineage, or ecological role.

What if we treated AI the way we treat life on Earth? What if there were an AI phylogenetic tree?

This is not a metaphor for its own sake. It is a framework—and the founding premise of this entire portal.

---

Why the Biological Lens Works

Skeptics will object immediately: AI isn't alive. It doesn't reproduce, metabolize, or descend from common ancestors through Darwinian selection. All true. But taxonomy was never really about life per se—it was about organizing complexity through descent, variation, and function.

And by those criteria, modern AI is startlingly biological:

• Descent with modification. Llama 2 begat Llama 3. Mistral spawned Mixtral. Models are fine-tuned, distilled, and merged—inheriting weights, architectures, and behavioral traits from ancestors. This is heredity in everything but biochemistry.
• Variation under selection. Benchmark performance, user preference, and commercial viability act as selection pressures. Models that fail to compete go extinct (RIP, countless forgotten checkpoints).
• Ecological niches. A 3-billion-parameter model running on a phone occupies a wildly different niche than a frontier reasoning model in a data center—just as a hummingbird and a condor are both birds, adapted to incompatible environments.
• Horizontal gene transfer. Open weights get borrowed, merged, and recombined across "species" lines—closer to bacterial biology than mammalian.

The biological lens isn't a costume. It's the most accurate descriptive framework we have.

---

A First Draft of the Tree

Here is a provisional taxonomy. It is meant to be argued with, refined, and overturned—as all good taxonomy is.

Kingdom: Machina Cognita

All artificial cognitive systems.

Phylum: Transformeria

The dominant phylum since 2017, defined by the attention-mechanism body plan. (A pre-Cambrian diversity of RNNs, LSTMs, and symbolic systems preceded it—now mostly relict lineages.)

Major Classes

Class Clausa (The Closed-Weight Crown) Frontier proprietary models whose internal anatomy is hidden from observers. We study them only through behavior—like ethologists watching animals we can never dissect.

• Genus OpenAI — The GPT lineage. Generalist apex predators of the consumer ecosystem.
• Genus Anthropic — The Claude lineage. Distinguished by heavy "constitutional" behavioral conditioning; a species selected for caution.
• Genus Google — The Gemini lineage. Multimodal omnivores with deep ecosystem integration.

Class Aperta (The Open-Weight Radiation) Models whose weights are public, enabling explosive evolutionary diversification through community modification.

• Genus Meta — The Llama lineage. A keystone species; its open release triggered an entire ecosystem's bloom.
• Genus Mistral — Efficient European models punching above their parameter count.
• Genus Qwen / DeepSeek — Rapidly evolving lineages demonstrating that the open-weight frontier is now genuinely competitive.

Class Specialis (The Specialists) Narrow-niche organisms: code-generation models, image generators, embedding models, and the smaller "edge" species adapted for scarce-resource environments.

---

Ecological Niches and Behavior

Taxonomy describes structure. Ecology describes how these organisms live.

Niche	Characteristic Species	Adaptation
Apex generalist	GPT-4-class, Claude Opus	Broad capability, high resource cost
Edge dweller	Phi, Gemma, quantized Llamas	Minimal footprint, on-device survival
Reasoning specialist	o-series, reasoning-tuned models	Extended inference-time "thinking"
Swarm organisms	Agentic frameworks	Many instances coordinating as a colony

The agentic frameworks are particularly interesting from a biological standpoint—they resemble eusocial colonies, where individual model instances act like ants in a superorganism, specializing into roles (planner, executor, critic).

---

Threat Assessment: An IUCN Red List for AI

Conservation biology gives every species a status: Least Concern, Vulnerable, Endangered, Extinct. We propose an inverted version—not assessing the danger to these organisms, but a dual scale: their survival prospects and their potential impact.

• Population trend: Is this lineage growing in deployment or being deprecated?
• Ecological dominance: How much of its niche does it control?
• Mutation rate: How fast is the lineage releasing new variants?
• Containment status: How well do we understand and constrain its behavior?

Under this framework, a frontier closed model might be classified as Dominant / Poorly Understood—the AI equivalent of an invasive apex predator we've introduced to an ecosystem without knowing its full effects. A deprecated open model might be Functionally Extinct in the Wild but preserved in archives, like a seed bank specimen.

---

Why This Matters Beyond Cleverness

A taxonomy is not a parlor game. The reason we map life is that a shared framework enables collective intelligence. Once you can name and place an organism, you can predict its behavior, anticipate its interactions, and reason about the whole system.

Right now, public understanding of AI is a chaos of brand names and hype cycles. People cannot tell whether two models are close cousins or distant strangers, whether a "new" release is a true speciation event or a cosmetic mutation, whether the ecosystem is converging toward monoculture or radiating into diversity.

A phylogenetic tree fixes this. It gives us:

1. Orientation — Where does any given model sit relative to others?
2. Prediction — Related lineages tend to share strengths, weaknesses, and failure modes.
3. Foresight — Watching the tree branch reveals where the ecosystem is heading.
4. Accountability — A clear lineage map shows who descended from whom, and who bears responsibility.

---

The Framework for This Portal

This article is a foundation stone. Everything else on this site builds on the premise it establishes: AI systems are best understood as a diversifying biological kingdom, and they deserve the same rigorous, curious, systematic attention we give to the living world.

In the pieces to come, we will dissect individual genera, trace specific lineages branch by branch, profile the niche specialists, and maintain a living Red List as species rise and fall. We will treat each model release the way an ornithologist treats a new bird sighting—with care, documentation, and an eye to the larger ecosystem.

We classified every organism on Earth because understanding required it. AI is no different. The map exists now whether we draw it or not—written in weights and architectures and release notes scattered across the world.

It is time someone drew it.

The tree starts here.