In Quests And Requests: projects that need incubating (#1) Scott Alexander requested a replication of a 2022 study which found that brain entrainment boosts learning rate on a certain hard perceptual task — distinguishing Glass pattern between each other.
https://www.astralcodexten.com/p/quests-and-requests
Our core team: Alexander Putilin (Software Engineer with 10+ years of experience) and Andrew X. Stewart (Postdoc EEG researcher, experienced Software Engineer) want to:
1. Find the effect and replicate the original study “Learning at your brain’s rhythm: individualized entrainment boosts learning for perceptual decisions” on consumer-grade hardware — primarily $3.6k OpenBCI Ultracortex "Mark IV" EEG Headset and $1k EMOTIV EPOC+ 14)
2. Create a flexible open-source toolkit allowing any hobbyist with a bit of programming knowledge and a consumer-grade EEG headset to run their own entrainment experiments on different tasks measuring learning rate and rate of improvement. The toolkit would allow a programmer with a $1k..$3.5k headset try their own tasks, try different delays on different stimuli, sync up stimuli up to peaks or troughs of the alpha rhythm, measure hardware delays and log progress across experiments
3. Find applications of the effect for the kind of things normal people would want to learn. This intentionally left pretty vague: the space of possible things to learn is huge — we’d like to find the effect first, get a good amount of first hand experience with it working on ourselves, and with a refined mental model see where we could apply it.
Project plan
Part 0 - Arrange hardware kit - $6000
Andrew has access to some clinical/research grade EEG equipment, including a 64-channel Biosemi ActiveTwo EEG system giving very good EEG data quality. To facilitate rapid environment development and data acquisition, we will also acquire two mid-grade OpenBCI EEG Cyton chips and electrode kits, for 8 or 16 channels of fairly good EEG data quality. Alexander has already raised ≈$1k for this purpose from crowdfunding: https://gofund.me/ca2cc315,but we need two consumer grade headsets
Part 1 - Build EEG Entrainment Environment and test 10 people (~2.5 months)
Person time - $21500 ($11500 for Alexander working full-time + $10000 for Andrew working part-time)
With securing major project time, Andrew and Alexander can focus on building the EEG Entrainment Environment, including the codebase for precise experiment trial presentation, online real-time processing of EEG data to assess ongoing alpha rhythm, and precisely present new ambiguous trials at either alpha-trough aligned times or non-alpha-locked times. We will also record this on more than 10 people, and report results.
Part 2 - Scale Up (or Write Up) - test 80 people (~2.5 months) if part 1 is successful
Person time - $21500 (same breakdown as part 1)
Travel and minor participant incentives - $900
If initial results are not compatible with the existence of this alpha-entrainment learning effect, we will write up our negative result, and close the project.
If we have at least provisional success at finding a significant learning rate improvement, we can switch efforts to data gathering on more people. While the original study had n=100, we will design our study to be first a close replication, as well as extending the experimental environment to gather more data per session, and allow repeated sessions.
While recording dozens of more subjects on the initial session, we will also perform “Deep Data” extended sessions on our first few subjects, exploring the longitudinal nature of this learning effect, exploring how additional sessions do or do not add to increased learning rate with personalized alpha-stim lock.
Part 3 - Test Generalizability of Learning Improvement, Write Up, Scale Out(2.5 months)
Person time - $21500 (same breakdown)
We are building the first steps of this project in a way that allows easy extensions. If we find learning rate improvement comparable to the original paper, then there is a natural next step of exploring the nature of what domains this learning effect can apply to.
The original paper uses Glass images, which have a noisy pattern of dots, where the task is to determine if each image is either more radial or more concentric. If we do find success in replicating this, our codebase will allow rapidly iterating on different experimental designs. For example, we can change the stimulus set and generate other visual perceptive choice tasks, while keeping the alpha-lock of entrainment on the trough of the individualized 8-12 Hz alpha rhythm, presenting different images at the precise millisecond to optimize the learning effect.
Are other major cognitive abilities susceptible to improvement in this alpha entrainment protocol? This Part is where this question will be addressed. We have implementations of Raven’s Matrices, sequence completion and moral choice questions that can be dropped in.
Alexander is an experienced Software Engineer with 10+ years of experience who worked at big companies like Yandex and Meta (WhatsApp) and startups: https://www.linkedin.com/in/alexander-putilin-6803346b/
Andrew is a postdoc researcher at UC Davis. He is lead programmer and developer of the ERPLAB Toolbox for loading, processing, and analyzing EEG ERP brain data, used by thousands of researchers. https://mindbrain.ucdavis.edu/people/andrew-stewart
Our core team is two Software Engineers, one of whom is a researcher with an expert-level domain knowledge.
Alexander links
LinkedIn: https://www.linkedin.com/in/alexander-putilin-6803346b/
Substack: https://psychotechnology.substack.com/
Twitter: https://twitter.com/42irrationalist
Andrew X Stewart links
UC Davis: https://mindbrain.ucdavis.edu/people/andrew-stewart
Twitter / X:
https://twitter.com/andrewxstewart
Google Scholar with papers and citation counts:
https://scholar.google.com/citations?user=vuFq-BCD1ogC
LinkedIn:
https://www.linkedin.com/in/andrewxstewart/
$27500 for the hardware and the first stage of the project to find signs of the effect on small number of people. +$22400 for the second stage to replicate the paper. +$21500 for the third stage to explore applications of the effect. Total: $27500 — $71400
No response.
Success for this project is increased knowledge of the nature of the true effect size of boosting learning rate through EEG alpha entrainment.
Major milestones of success, with (P-success) are:
1. Successful re-creation of the key aspects of the environment from the original paper, including personalized alpha profiling, and presentation of Glass images precisely locked to this alpha rhythm. (85% with Part 0 and Part 1 funded).
2. Python + PsychoPy codebase to support the above, including scalable difficulty and phase intervention parameters
(85% with Part 0 and Part 1 funded).
3. Alpha-entrainment learning effect observed on most of first 10 subjects, with at least half the effect size reported in Michael 2023 (so around +0.02 in T-Match condition)
(55% with Part 0 and Part 1 funded)
4. Alpha-entrainment learning effect observed on most of first 80 subjects, with at least half the effect size reported in Michael 2023 (so around +0.02 in T-Match condition)
(65% with Part 2 funded, conditional on proceeding)
5. Maybe 25% chance of finding applications for the kind of things normal people would want to learn
App status across various funders