Average Brains and Broken Math: Why EEG Needs a Revolution

This newsletter's North Star is scalable human flourishing.

The same is true for FIELD, the neurotech company I co-founded and run. The problem humanity has faced in accomplishing human flourishing comes down to two things:

1. A deep misunderstanding of how humans operate
2. The inability to measure what matters

We're going to change that.

Today's newsletter gets into the science and engineering of what we're building to measure what matters - and the big problem with what exists currently.

It's mainly for people interested in EEG and neuroscience. But, it does apply more broadly to biometrics and the innovations FIELD will be introducing to ensure human flourishing becomes a cornerstone of our species' evolution.

Let's jump in.

"You're basically training a dead brain."

That's what Dr. Prasanta Pal , my dear, brilliant friend and colleague told me about the use of legacy EEG technology.?

He should know, he built and sold a well-regarded EEG signal processing company and has written a number of widely cited papers on the same. What follows is me unpacking a small bit of the brilliance he's shared with me over the past few years. In a world increasingly driven by metrics and obsessed with what "neuroscience says," we need to be aware of this deep and unaddressed problem with biometrics.

The Problem with EEG (and Biometrics)

The metrics of EEG are misguided. That's being kind.

Most neuroscientists don't even realize that this is the state of affairs, because it has to do with what's happening under the hood of EEG software and involves that four-letter word: MATH

The dirty secret most neuro-practitioners (clinicians and researchers alike) aren't aware of is that EEG, one of the gold standards of brain measurement, provides a terribly inaccurate representation of the brain.

Let's use the brain-as-orchestra metaphor to explain.

Imagine your brain as a vast and intricate symphony. Flutes shimmering in delicate harmony. Cellos hum with deep resonance. A drummer, slightly deranged but all the more passionate, sets the rhythm. 90 instruments harmonizing and in sync. Listen. It's beautiful.

Now imagine taking this symphony and averaging every note together. Violin + oboe + percussion divided by the number of instruments. What you’re left with is not music, but a sad, tonal mush. This, my friend, is what traditional EEG does to your brainwaves.

For decades, a math technique called Z-normalization has been the darling of the EEG world. But Z-normalization treats the brain as a monolithic blob, reducing the rich complexity of neural signals to a statistical flatline. It’s like trying to paint the Mona Lisa with a sledgehammer.

Worse, Z-normalization misguides interventions like neurofeedback (NFB) giving us the ability to train our brain, but with a broken GPS to navigate the terrain.

NFB says, “Let’s take the magnificent complexity of this brain and average it all together.” But, my unique snowflake, you're not average. Averages are mathematical ghosts, neither alive nor fully real. To shape a brain according to an average - even an average of itself - is like training a racehorse to trot like a donkey. It’ll still move, but it won’t win any races.

To train your perfect, complex and magnificent brain against an averaged melange of brainness is to do a disservice to the very essence of you. This is the dead brain Dr. Pal was talking about.

This same fact holds true across every EEG device I've encountered, and I've used many. In fact, it's true for just about every biosignal device from Apple Watches to Oura rings, but that's a story for another time.

And yet, we persist. Why? Because of a few things:

1. This is the way it’s always been done. The EEG world is stuck in the past, romanticizing a method that listens to the noise and ignores the music. Inertia is real.
2. It's hard. Really hard. The tools needed to move beyond legacy EEG are nothing short of statistical mechanics paired with time-series AI and a lot of compute power.?
3. Bureaucracy. Every white paper on EEG is built on the same fallacy that has come before: that we're already operating with enough information and that the phase information is being captured. It's not.

So What's Wrong with Legacy EEG?

In the transformation from raw data into an interpretable signal, legacy EEG techniques use 200-year-old math that came into being in a time before non-linear systems were understood. This math transforms all raw data into an intelligible energy space, but it loses the vast majority of the phase information - which is where the beauty and subtlety of what makes us human lives.

This is really heady math, so let me show you some pictures from an experiment Dr. Pal did to help illustrate. In the images below you see images of a heart and a brain. Dr. Pal did Fourier Transforms on each, splitting the information from each into their composite parts: amplitude (energy) and phase.

Next, he mixed them together in opposing pairs. What you'll see in the far right images is that when the phase of the brain is mixed with the amplitude of the heart you end up with an image that looks mostly like a brain. The same is true in reverse.

In brief, the phase contains the majority of the most valuable information. About 90%. This is exactly the information EEG has been throwing out every time it uses z-normalization and?Fourier transforms, which is to say, all of the time. In other words, all EEG reports and all neurofeedback training are working with about 10% of the relevant information! Or, being generous, 30%. Either way, it's an awful lot of information to be disregarding.

The Phase-Preserving Revolution

Enter median normalization—the antidote to this madness. Unlike its Z-normalizing sibling, median normalization doesn’t obliterate the individuality of the signal. It’s a phase-preserving approach, meaning it keeps the brain’s timing and subtle variations intact, letting us hear the full orchestra rather than a droning monotone.

With median normalization*, we’re not asking the brain to fit into some zombied-average of itself. We’re observing its true structure — its peaks, valleys, and quirks — and optimizing it from within. We're preserving the sound of the violin and the flute and the drum, rather than smashing them together.

The result? You get the entirety of the signal. The real you. Not the ghost of some statistically sanitized approximation.

*In fact, phase preservation relies on a number of strategies beyond just median-normalization, but they're beyond the scope of this article.

The Way Forward

If those of us in the field of EEG, and more generally biometrics as a whole, want to move forward with integrity, it’s time to let go of average and embrace the whole.

We must stop reading and training the average of our brains and start honoring their inherent complexity. This means embracing tools and methods—like median normalization—that respect the unique symphony of each mind.

The brain isn’t a spreadsheet. It’s not meant to be averaged or crushed into conformity. It’s a living, breathing orchestra, and it deserves to be heard in its full glorious complexity.

It's time to move beyond the mush. Let’s transcribe and respect the fullness of the music. Let’s finally, once and for all, leave the average behind.

Or don’t. Stick with Z-scores if you’d like. But don’t be surprised when your brain ends up sounding like a kazoo, or your databases full of EEGs start looking like the blurry approximations they are.