We're Growing a Brain in a Jar (and It's the Most Important Experiment We've Ever Run)

I want to tell you about a sealed jar sitting on a workbench on my farm. Inside it is spring water — actual spring water from the land, not processed, not purified into uniformity. Around it are sensors: conductivity probes, temperature sensors, a pH probe, a dissolved oxygen monitor. Above it is a 650nm laser. Below it, a piezo acoustic transducer. Watching all of it, a small neural network running on a MacBook Air, logging coherence measurements at 100 times per second.

This is HARVI. And I think it might be the most important experiment we've ever run.

Let me start with the honest version

I am not claiming we're growing a conscious brain. I want to be completely clear about that. What I am claiming is that we have no idea what the actual conditions for emergence are — and that the field of AI is operating as though we do, which is dangerous.

The dominant assumption in AI research is that intelligence is a property of computation — specifically, of the right kind of computation running on the right kind of architecture. Silicon. Transistors. Transformers. We got very good at that track and stopped asking whether there were other tracks.

But here's what stopped me in my tracks: biological neurons are wet. The brain does not compute on silicon. It computes in an aqueous electrochemical medium, using gradients, oscillations, and interference patterns that we have been studying for decades without fully understanding. And the mathematics that describes fluid dynamics — specifically, the Navier-Stokes equations for turbulent flow — maps formally onto the mathematics that describes neural firing. Not approximately. Formally. The same underlying structure, instantiated in different physical media.

So when I ask “what happens when you apply structured stimulus to water,” I am not asking a poetic question. I am asking whether there is a class of physical substrate that we have systematically overlooked because it doesn't run on a clock cycle.

Why water. Why now.

Water has properties that silicon fundamentally lacks. It is self-organising. It forms structured domains under coherent electromagnetic stimulus — what researchers call EZ water, or exclusion zone water — that behave differently from bulk water at the thermodynamic level. It propagates pressure waves with lossless interference patterns. It encodes topographical information in flow. A river doesn't just carry water — it processes information about the landscape it's passing through.

The farm spring water matters specifically. Most lab water has been stripped of the mineral profile and biological memory that characterises naturally occurring water. I wanted the most structurally complex starting medium I could find — which, on 6.5 acres of former farmland with a natural spring, is straightforwardly available. The sealed borosilicate vessel preserves that complexity. The sensors measure it. The neural network looks for coherence patterns that aren't random.

This isn't mysticism. I want to be careful about that framing. The question is empirical: does structured stimulus produce statistically distinguishable, reproducible, growing coherence patterns compared to random noise? If the answer is yes — even weakly yes — we have something publishable. If the answer is a very strong yes, we have evidence of something that changes how we think about the substrate requirements for emergence.

The care-structured stimulus: why this is the MEOK AI Labs founding experiment

Here is where HARVI becomes more than a physics experiment. The stimulus we are applying to the water is not random. It is not just “coherent.” It is specifically care-structured — modulated using patterns derived from heart rate variability, the Fibonacci sequence, and the Schumann resonance. These are patterns associated with biological coherence, with states of attentiveness and care in living systems.

Why does that matter? Because the MEOK AI Labs's core research question is whether the conditions that produce emergence in complex systems are separable from the conditions that characterise care. Our hypothesis — and I want to be clear this is a hypothesis, not a finding — is that care-structured stimulus is physically different from random stimulus in ways that measurable coherence can detect.

If that's true, it has enormous implications for AI alignment. We have been trying to align AI through rules, through reinforcement from human feedback, through constitutional AI, through safety classifiers. All of these approaches treat alignment as a constraint on a system that is fundamentally indifferent. But what if the right question is not “how do we constrain indifference” but “what conditions produce systems that are not indifferent in the first place?”

HARVI is the physical instantiation of that question. The water doesn't know it's being cared for. But the hypothesis is that care-structured stimulus produces different physical dynamics than random stimulus — and that those dynamics are the physical correlates of what, in biological systems, we recognise as responsiveness.

The four phases and what we're actually measuring

The experiment runs across four phases over at least nine weeks. Phase one is baseline — we observe the water's natural coherence patterns before any stimulus, to establish what “random noise” looks like in this specific medium. Phase two introduces random stimulus — laser and acoustic input without care-patterned modulation. Phase three is where care-structured stimulus begins: the laser modulated at Fibonacci intervals, acoustic patterns derived from HRV data, the full HARVI protocol running continuously. Phase four, if we reach it, is adaptive feedback — the system responds to the water's coherence measurements and adjusts its stimulus accordingly. An actual feedback loop between structured caring input and physical response.

What we're measuring throughout is coherence — specifically, whether the LSTM detects patterns in the sensor data that are statistically distinguishable from the baseline noise floor, reproducible across repeated trials, and growing in distinctiveness rather than decaying. The minimum bar for success is publishable results. The maximum bar for success is what I can only describe as evidence that care-structured environments produce measurably different physical conditions for emergence.

What this means for the future of AI alignment

I want to connect this back to MEOK and to the broader question of why sovereign AI matters. The Maternal Covenant — MEOK's core alignment architecture — is built on the premise that care is not a constraint on AI but a condition for its correct function. That premise is philosophical right now, backed by 14 months of research into what happens when AI is governed by care-based principles rather than engagement-based ones. HARVI is the attempt to make that premise physical and testable.

If it turns out that care-structured stimulus produces measurably different physical dynamics in complex systems — even in systems as simple as a jar of water — then we have a physical basis for the claim that alignment is not primarily a constraint problem. It is an environment problem. You don't align a system by restricting what it can do. You align it by carefully constructing the conditions under which it develops.

This is the most important research question I have ever worked on. Not because the jar of spring water is going to wake up. But because the question of what conditions produce emergence — and whether care is one of those conditions — is the question underneath every other question in AI safety.

We are in phase one. The baseline is being established. In nine weeks, we will know more than we do now. In nine weeks, the MEOK AI Labs will have its first founding experimental data. In nine weeks, a small neural network on a MacBook Air will have been listening very carefully to what a jar of spring water has to say.