How did Neurable Validate Our AI’s Ability to Track Focus?

While building good focus habits and racking up Focus Points should feel easier—and even fun—over time, the scientific process behind validating how our electroencephalogram (EEG) sensors and AI capture, track, and measure focus was incredibly rigorous and serious. Here’s how we did it.

Proving EEG Performance: Enten vs. Quick20

‍In 2021, to confirm that our consumer-friendly EEGs were capturing accurate brain data on par with Quick20 EEG hardware used by neuroscientists and neurosurgeons in clinical settings and that our AI was accurately estimating focus, we ran a series of 45 experiments with 132 participants across 337 sessions with one of our early prototypes, the “Enten.”

First, we needed to prove our Enten headset could reliably capture neural activity as well as Quick20 headsets, so we ran a series of industry-standard EEG protocols called Eyes Open and Closed, Auditory Steady-State Response (ASSR), and P300. In each of these experiments, we proved our headset’s performance at picking up neural activity was on par—or even slightly better—than industry-standard hardware.

¹ Our Eyes Open and Closed experiment measured Alpha wave activity. Closed eyes = greater alpha oscillations. Participants were instructed to keep their eyes open for 30 seconds, then closed for 30 seconds. We ran the experiment twice, concluding our sensors were at least on-par, or perhaps slightly better, at picking up ear-centric Alpha activity than Quick20 headsets.

Our Auditory Steady-State Response (ASSR) experiment measured how quickly nerves respond to certain stimuli. We played participants a series of tones modulated at specific frequencies, expecting to see an increase in that frequency in their brainwaves—and we did—particularly from the electrodes along the auditory cortex. We were excited to discover Neurable’s sensors were more sensitive at picking up these signals than Quick20 sensors.Our P300 experiment used an unexpected visual prompt to stimulate a rapid change in neural activity. When participants saw something “odd,” we expected our headset to record a signal spike around 300 milliseconds after that stimulus was presented. In our experiment, participants were shown four words and were told to press a button when they saw the word “BLUE,” which appeared 25% of the time. We saw similar P300 amplitude in both the Enten and Quick20 EEGs.

Proving Our AI Could Predict Focus

After we validated our hardware, we turned our attention to validating our AI by asking participants to complete a series of experiments called Distraction Stroop Tasks and Interruption by Notification, which were designed to mimic how focus might be maintained or disrupted in a real-world office setting. We wanted to demonstrate these tasks would cause changes in focus levels and that our proprietary algorithm would be able detect the changes in attention consistently, across several days, regardless of participant. And we did! Our AI correctly captured 80% ± 4.1% of distractions across subjects, time points, and conditions.

² Distraction Stroop Tasks experiment: The Stroop effect (also known as cognitive interference) is a psychological phenomenon describing the difficulty people have naming a color when it's used to spell the name of a different color. During each trial of this experiment, we flashed the words “Red” or “Yellow” on a screen. Participants were asked to respond to the color of the words and ignore their meaning by pressing four keys on the keyboard –– “D”, “F”, “J”, and “K,” which were mapped to “Red,” “Green,” “Blue,” and “Yellow” colors, respectively. Trials in the Stroop task were categorized into congruent, when the text content matched the text color (e.g., Red), and incongruent, when the text content did not match the text color (e.g., Red). The incongruent case was counter-intuitive and more difficult. We expected to see lower accuracy, higher response times, and a drop in Alpha band power in incongruent trials. To mimic the chaotic distraction environment of in-person office life, we added an additional layer of complexity by floating the words on different visual backgrounds (a calm river, a roller coaster, a calm beach, and a busy marketplace). Both the behavioral and neural data we collected showed consistently different results in incongruent tasks, such as longer reaction times and lower Alpha waves, particularly when the words appeared on top of the marketplace background, the most distracting scene.‍

Interruption by Notification: It’s widely known that push notifications decrease focus level. In our three Interruption by Notification experiments, participants performed the Stroop tasks, above, with and without push notifications, which consisted of a sound played at random time followed by a prompt to complete an activity. Our behavioral analysis and focus metrics showed that, on average, participants presented slower reaction times and were less accurate during blocks of time with distractions compared to those without them.

This is a high-level overview of our results. For those who enjoy digging into the scientific details, read our full white paper here.

Never Stop Iterating, Experimenting, Improving‍

Since 2021’s Enten prototype, we’ve done two product-cycle iterations, arriving at MW75 Neuro, our most cutting-edge Neurable AI device to date. We continued to evolve our focus algorithm, too, so it performs well on anyone regardless of age, race, head shape, hairstyle, skin type, etc.

After nearly a decade of research with thousands of users, we’re proud to say our technology and proprietary AI still perform as well—or better—than industry-standard EEGs, a big leap toward bringing brain-computer interface neurotechnology out of the research lab and into daily life. You could say we’ve been laser-focused on creating products to help people banish distractions and build better focus habits. What’s your Focus Goal for today?