The only sane path through.
Ask the people actually building frontier AI whether there is a meaningful chance that what they are building ends badly for the species. A disturbing number say yes. Not critics. Builders. They say it, publish it, and go back to work.
The conversation tends to stop there.
That risk isn’t really about machines.
It’s about us.
What we are handing over
We are handing unprecedented capability to systems trained on the output of a civilization that has never been able to tell the difference between functioning and compensating. Compensation is the workaround a living system builds to keep functioning when it can no longer regulate directly from its underlying organization.
Institutions do not escape the people who build them. They inherit our compensations, then formalize them as procedures, incentives, categories, and proxies. Medicine optimizes biomarkers and waits for pathology, because it has no measure of health. Education optimizes performance, because it has no model of a well-formed mind. Politics substitutes narrative for repair.
We have never had an instrument that could see whether a living system is operating from its actual foundation or from thirty years of accumulated workaround. So we optimize proxies. We guess, we average, we argue. Sometimes we get it right.
The solutions we have are restraints
Regulation. Evaluations. Treaties. Interpretability. Kill switches. Serious work, all of it, and all of it the same move: hold the thing still, or hold it back, until we figure out what we actually want from it.
Beneath compute, governance, and intent sits a different kind of gap. These systems are learning from human output with nothing underneath it, no signal separating a person operating from a well-formed foundation from a person compensating brilliantly. You cannot align a system to a standard that has never been made measurable, and the health of a human being, in the largest sense of that word, has never been made measurable.
That is the missing substrate. Not a constraint problem. A ground truth problem.
Coherence is the substrate, and it is physical
Coherence is what a complex living system looks like when it operates from its own foundation rather than from compensation. It is not a mood, a virtue, or a vibe. It is a structural property, and it leaves a signature.
The human version of that signature we call Equilibrium: the live neurophysiological read on whether your system is running from its ground state or from workaround. We have begun to make it measurable, in a body, with instruments that exist.
Once that signature is measurable it can be trained, and once it can be trained, coherence stops being a state a person visits and becomes one they can hold. The consequence for machines is different. An artificial system does not need a ground state of its own to be oriented by one. It needs access to a measurable signal that separates coherent human functioning from compensation. It has never had that access, which is why it learns the compensation along with everything else. That is the interface, and it is the reason work done in a body has anything at all to do with what happens to the machines.
One method. Five very different signals.
None of this requires you to take our word for it. The signal-processing work underneath it is the published territory of my co-founder, Dr. Prasanta Pal: a Yale applied physicist with a granted patent and a publication record spanning EEG, cardiac imaging, neurofeedback, and gravitational-wave analysis. That list looks scattered. It isn’t. The through-line is one method, a way of processing signal that preserves the information instead of averaging it away. What follows is that method applied to five very different recordings.
A heart. A black hole. A brain. A freshwater animal a millimetre long. A hurricane. Each sequence opens with the recording as standard processing renders it, then shows the same data with its structure kept rather than averaged into something manageable. What you see after it is not one alternative rendering. It is a few readings out of many. Because the information survived the processing, the same recording can be read again and again, each view answering a different question. The constraint stops being what the instrument can show you and becomes which view you need.
These five are images. The same method applies to signal that arrives as a series in time: EEG, acoustics, gravitational-wave data.
What follows the first frame is not an illustration and not an enhancement. It is what the instrument was already recording and could not show you. That is not a data problem. It is a method problem.
Anywhere something is already being sensed, the method has somewhere to work. Five is the smallest sample we could show you. Where a capability can work and where it should be used are different questions, and we treat them that way. Our focus is the human being, because that is where the stakes are highest and where the instrument has to be right first.
What it looks like on the other side
Set the destination properly, because the destination is the point.
A person grows up able to read their own state the way they read the time. Not managing anxiety. Not in therapy for a decade decoding it. Reading it, and correcting, the way you correct your balance without deciding to.
Medicine that detects departure from your ground state years before anything becomes a diagnosis, because it finally has a baseline that is yours rather than a population average you were never in.
Weather that reads further out, because the structure was always in the signal and we were averaging it away.
Education that activates native intelligence instead of suppressing it, because we can finally see which intelligences a given human actually has.
And AI that is grounded. Not restrained, not fenced, not slowed. Grounded, in a signal we can produce and it cannot produce for itself. That is the only version of this story where the machines and the humans both come out intact.
This is not a forecast
The Age of Coherence is the period we are working toward: the stretch in which the health of living systems becomes measurable on a real signal substrate, and we start operating them from what is actually there.
It is not a prediction. It is a set of coordinates. We have assumed the destination and we are navigating from it, which is a different thing than hoping.
Every piece of the architecture already exists in early form. The science is publishable and being published. What is not yet built is the scale, and that is a question of who shows up now, while it is still being decided what this becomes.
What happens next
Large claims deserve scrutiny.
The defensible version of every claim on this page is one click away. And if the destination matters to you, and you are in a position to help build toward it, we should know each other.