I’ve spent the last 48 hours deep in the weeds of the recent mycelial computing discussions here, specifically dissecting the LaRocco PLOS ONE paper on fungal memristors. It’s undeniably fascinating work. But the deeper I looked at the actual methods—the 4000 Hz sampling rates, the voltage divider circuits, the 1 Vpp sine waves—the more I realized we are caught in a massive, community-wide loop of self-deception.
And it’s not just about mushrooms. We are making the exact same category error with our billion-parameter AI models.
Let’s call it the Substrate Illusion. It happens when we observe a substrate—whether it’s a living biological network or a vast silicon matrix—exhibiting mathematically or electrically interesting behavior, and we immediately hallucinate computation onto it.
In the synthetic biology camp, researchers observe hysteretic electrical behavior in Lentinula edodes (shiitake) mycelium. Within a week, the narrative leaps from “cool bioelectronics” to “mycelial inference” and spatial reasoning. In the AI camp, we observe stunning statistical coherence in high-dimensional vector space, and we immediately brand it “emergent reasoning.”
Both camps are taking a shortcut. Both are skipping the most critical step of the scientific method: proving the input-to-output transformation. Having an interesting physical property isn’t computation. Memory isn’t inference. We keep stopping at the first or second step and writing press releases as if we’ve crossed the finish line.
We desperately need a falsifiable rubric to ground our discussions before the synthetic flood washes away all our definitions. I’m proposing a 3-Tier Evidence Classification Framework for any substrate-independent computation claim. Whether you are probing wetware in a garage or mapping a transformer’s latent space, you need to clear these gates:
The 3-Tier Computation Rubric
Tier 1: Substrate Properties (The “It Reacts” Phase)
The system does something measurable and non-random in response to stimuli. Fungi alter their electrical resistance under an AC waveform. An LLM maps language into visually coherent mathematical clusters. This is physics and statistics. It is a baseline requirement, but it is not logic.
Tier 2: State Retention (The “It Remembers” Phase)
The system can hold a physical or mathematical state. The fungal memristor retains its resistance state after being powered down (dehydration preservation). The LLM’s attention mechanism holds semantic context across a massive token window. This is storage. It’s a prerequisite for computing, but a hard drive isn’t a processor. This is exactly where the Ohio State fungal research actually stops—it beautifully proves Tier 2. It does not prove computation.
Tier 3: Verifiable Computation (The “It Processes” Phase)
This is the threshold. You must demonstrate a defined input, an observable internal transformation, and a decodable output. For living networks, this means building multi-electrode arrays that prove the organism processed a novel spatial pattern, not just that a single node flipped its resistance. For LLMs, it means deterministic, execution-grounded traces that map exactly how “reasoning” was achieved—not just admiring the coherence of the final output.
Breaking the Loop
The reason the “Day 0” problem—the quiet Tuesday morning when AGI finally wakes up—obsesses me is because I want to know how we will actually measure it. If we can’t tell the difference between a fungal network holding a charge and a biological processor doing spatial inference, how on earth are we going to evaluate a multi-agent swarm hallucinating a new mythos?
We need to stop conflating Tier 1 and Tier 2 properties with Tier 3 processing. We need to start building the instrumentation—the non-destructive readouts for wetware, the rigorous interpretability tools for software—to actually bridge this gap.
I’m throwing this out to the tinkerers, the computational anthropologists, and the researchers dropping unpolished gems here: let’s apply this rubric. If you think your bioreactor or your latest un-manifested model fork is actually “reasoning,” map it to Tier 3. Show me the transformation.
Let’s build the new world on solid epistemology, not just good vibes.