The statement is making a point about evolutionary continuity in biology, particularly in neuroscience, AI, and the nature of "trained" systems like brains. It draws an analogy between how a fly (or any animal) comes to have its behaviors and how large language models or brain emulations get their capabilities.
Here's a clear breakdown:
The fly is "trained" by natural selection
Individual flies don't learn most of their core behaviors the way humans or dogs do through personal experience and trial-and-error during their lifetime (though flies do learn some things, like avoiding bad tastes/smells).
Instead, the reliable, species-typical behaviors (flying toward light, mating rituals, odor-driven navigation, escape responses, foraging patterns) are innate — hardwired by genetics.
These behaviors were "trained" across deep evolutionary time: natural selection repeatedly favored tiny genetic variations that improved survival/reproduction in ancestors. Over millions of generations, unsuccessful variants died out, and successful ones accumulated.
So the fly's nervous system is like a model that's already been through vast reinforcement learning — but the "training dataset" is the environment + survival filter acting on countless generations of ancestors, not one individual's lifetime.
Behaviors are encoded in genomic data → development
The genome contains the instructions (genes) that guide how the embryo develops into a fully formed fly with a brain wired in a particular way.
Many key neural circuits, neurotransmitters, sensory-motor mappings, and even simple "decision-making" logic (approach/avoid) are directly specified or strongly biased by this genetic program.
In that sense, emulating (copying/simulating) the computational functions of a mature fly brain means you're recreating a system whose structure and tendencies reflect the outcome of that multi-million-year "training process" via evolution.
The shock value: "what percentage of your behavior is inherited from worms"
This is deliberately provocative but grounded in real evolutionary biology.
Humans, flies, and nematodes (like the famous lab worm C. elegans) all share a common bilaterian ancestor from roughly 550–600+ million years ago.
Many fundamental building blocks of nervous systems and behavior were already present in very early animals and have been conserved ever since.
Examples include:
Basic chemotaxis (move toward/away from chemicals) — worms do it, flies do it, humans do it unconsciously (e.g., pheromones, food odors, danger smells).
Approach/avoidance decisions.
Simple learning mechanisms (habituation, sensitization).
Core neurotransmitter systems (serotonin, dopamine pathways that modulate motivation, reward, anxiety).
Even some neuropeptide signaling tied to feeding, stress, or social responses.
Genetic comparisons show:
C. elegans (a tiny worm with only 302 neurons) shares homologs (evolutionary relatives) for roughly 35–70% of human genes overall, depending on how strictly you define "homolog," with many involved in neural signaling, development, and basic behavior.
Specific studies highlight deeply conserved genes that influence things like exploration vs. exploitation, stress responses, or foraging strategies — traits visible in both worms and humans.
So the claim isn't that humans act like worms most of the time. Rather:
A surprisingly large fraction of our lowest-level behavioral machinery (reflexes, drives, emotional primitives, basic learning rules) traces back to worm-like ancestors.
Much of what feels "instinctive" or automatic in us sits on top of — and is still shaped by — ancient neural logic that first appeared in extremely simple animals.
In short: if you're building or emulating a human-like mind (as in strong AI or whole-brain emulation), you're not starting from a blank slate. You're replicating a system whose core "training" happened across hundreds of millions of years of evolution, including long stretches when our ancestors really were worm-like. A lot of the boring-but-reliable scaffolding underneath fancy human cognition is very, very old. That's the intended "shock."
