Chapter 9 · Middle school
Design a brain-computer interface of your own: two of your states, a custom AI head that tells them apart, and a board that does something useful with it.
Choose a purpose: something a light or a signal that follows your mind could do. A sign that shows when you are concentrating. A signal that starts a small robot of your own.
Then choose two states, teach a custom AI head to tell them apart, test it fairly against the test you wrote first, and make the Raspberry Pi Pico 2 W act on it.
| Period | Milestone | What happens |
|---|---|---|
| 1 | Teach | Each wearer teaches a custom AI head with the group's two states. |
| 2 | Test | Every score next to guessing. Which wearer and states will you build on? |
| 3 | Build | The program on the board, paired with the group's IntoMind One. |
| 4 | Build | The board's light or pin doing the job you chose. |
| 5 | Test the whole | Ten switches: how often does the board follow? Then another group tries to fool it. |
| 6 | Improve | Change one thing, and test again the same way. |
| 7 | Improve | Keep the better version, and write down why. |
| 8 | Prepare | The presentation: purpose, states, test, result, and what you would change. |
| 9–11 | Present | Five minutes for each group, with the board working, or a recording of it. |
Record
| Try | Wearer | Score, of 88 | Clearly better than guessing? | Light followed, of 10 | What we changed |
|---|---|---|---|---|---|
| 1 | |||||
| 2 | |||||
| 3 | |||||
| 4 |
Work it outDid it meet the test we wrote first?
Wearing it is a choice. A group in which nobody wants to wear it teaches and tests custom AI heads on a simulated IntoMind One, by adding ?dev to the end of a demo's address. The board connects only to a real IntoMind One, so for the board's stages a volunteer, such as your teacher, teaches a custom AI head and wears it whenever the board needs someone to follow.
Projects on the board's pin stay with its own 3.3-volt signal. Nothing on it plugs into a wall.
Saved files come from someone's brain: keep them only with the wearer's permission, and delete them all together when the capstone is finished.
Chapter 10 · High school
Design a brain-computer interface for a person and a purpose. Does it pass the test you wrote first, and does it still work on another day?
Design for the person and the purpose in your brief from Designing a BCI, and keep to the test you wrote there.
Teach a custom AI head with your two states, at your site, on two different days. Compare every score with guessing, and find out whether the first day's head still works on the second. Then make the Raspberry Pi Pico 2 W act on it for your purpose, test the whole interface against your test, and present what happened.
| Periods | Milestone | What happens |
|---|---|---|
| 1–2 | Proposal | Your brief and your test, reviewed with your teacher. Change only what your teacher asks, before anyone teaches. |
| 3–5 | Teach, first day | Each wearer teaches a custom AI head with your states, at your site, and saves it and its training data, with the wearer's permission. Every score next to guessing. |
| 6–8 | Teach, second day | On another day, each wearer first installs the first day's head and counts how many of ten switches it follows. Then each teaches again. |
| 9–12 | Analyze | Both days side by side in a spreadsheet, with the skeptic's notes. Another group checks your analysis against your test. |
| 13–20 | Build | The program on the board, changed for your purpose, and your circuit on its pin, with the threshold you planned. |
| 21–23 | Test it | Your test as written: ten switches, five of them timed, five minutes of rest, and a timed setup. Then someone from another group tries to fool it. |
| 24–27 | Improve | One change at a time, each tested the same way. |
| 28–29 | Prepare | The presentation and the report. |
| 30–33 | Present | Eight minutes for each group, with questions. Then every saved file is deleted, together. |
Record
| Wearer | Day | Score, of 88 | Clearly better than guessing? | Skeptic's notes |
|---|---|---|---|---|
| 1 | First | |||
| 2 | First | |||
| 3 | First | |||
| 4 | First | |||
| 1 | Second | |||
| 2 | Second | |||
| 3 | Second | |||
| 4 | Second |
Record
| Wearer | The first day's head, on the second day: light followed, of 10 | The second day's head: light followed, of 10 |
|---|---|---|
| 1 | ||
| 2 | ||
| 3 | ||
| 4 |
Record
| Requirement | What we asked for | What we got | Met? |
|---|---|---|---|
| The score | |||
| Switches the light followed, of 10 | |||
| Median delay, in seconds | |||
| False positives in 5 minutes of rest | |||
| Setup time | |||
| Did blinks or a clenched jaw switch it? |
Work it outDid it meet the test we wrote first? Did the first day's head still work on the second day, and what might explain it?
One page, clear enough that another group could repeat what you did, in this order:
Wearing it is a choice. A group in which nobody wants to wear it teaches and tests custom AI heads on a simulated IntoMind One, by adding ?dev to the end of a demo's address. The board connects only to a real IntoMind One, so for the board's stages a volunteer, such as your teacher, teaches a custom AI head and wears it whenever the board needs someone to follow.
Projects on the board's pin stay with its own 3.3-volt signal. Nothing on it plugs into a wall.
Saved files come from someone's brain: keep them only with the wearer's permission, and delete them all together when the capstone is finished.
Chapter 12 · College
Carry a brain-computer interface from a user's need to a working board: requirements, a preregistered test, data collected under a data plan, a reproducible analysis, and the loop deployed with its latency measured. Does it meet what you said it would?
Start from the user's need and the requirements you wrote in Designing a BCI. Preregister your hypothesis and analysis, with the amount of data justified by simulated power. Collect under the data plan your instructor approves, in at least two sessions per wearer on different days. Analyze with held-out rounds and a permutation test, test whether a head from one day works on the next, and reproduce your custom AI head from its training data byte for byte. Deploy it on the board, measure the loop's latency, and test the whole against your requirements.
| Session | Milestone | What happens |
|---|---|---|
| Meeting 1 | Design review | Each group presents its proposal: the user, requirements, the preregistration, the data plan and the simulated power, for critique and approval. |
| Lab 1 | Collect | First sessions for every wearer, under the approved plan, every head and its training data saved with permission. |
| Meeting 2 | Analysis clinic | Each pipeline checked against its preregistration: held-out rounds, the permutation test, no leakage. |
| Lab 2 | Collect again | Second sessions, on another day: the first day's head tested first, then a new one taught. |
| Meeting 3 | Analysis clinic | Transfer across days, the head reproduced byte for byte, and every deviation from the preregistration written down. |
| Lab 3 | Deploy and test | The board, the program, the latency measured, and every requirement tested as written. |
| Meeting 4 | Peer review | Drafts exchanged and reviewed against what makes a strong project. |
| Meeting 5 | Revise | Revisions, with the deviations stated. |
| Meetings 6 and 7 | Present | Twelve minutes for each group, with questions. |
| Lab 4 | Demonstrate | Working demonstrations. Reports due. Then every saved file is deleted, together. |
Record
| Requirement | As preregistered | Result | Met? |
|---|---|---|---|
| Effectiveness | |||
| Efficiency | |||
| False alarms | |||
| Setup | |||
| The first day's head on the second day | |||
| The loop's latency |
Wearing it is a choice. A group in which nobody wants to wear it teaches and tests custom AI heads on a simulated IntoMind One, by adding ?dev to a demo's address. The board connects only to a real IntoMind One, so for the board's stages a volunteer, such as your instructor, teaches a custom AI head and wears it whenever the board needs someone to follow.
Projects on the board's pin stay with its own 3.3-volt signal. Nothing on it plugs into a wall.
Saved files come from someone's brain: keep them only with the wearer's permission, under your data plan, and delete them all together when the capstone is finished.
Chapter 12 · Graduate
A question of your own, answered as a registered report: the question, the design and the analysis reviewed before any data, then the data, the planned tests, and an honest account of what they show.
Choose a question that this book leaves open and that your group can answer with what it has. Write it as an estimand, design the study, simulate its power with numbers from the labs, write the confirmatory analysis as code that runs on simulated data, and write the protocol: consent, data and deletion. That is the first stage, and it is reviewed before any data exist. After approval, collect, run the planned analysis unchanged, and label everything else exploratory. That is the second stage.
| A question the labs leave open | Where it comes from |
|---|---|
| Does the trainer's penalty cost accuracy on real sessions of 88 windows, and does a label-free heavier one recover it? | Models that learn, and the Mind-Controlled Light Switch |
| For which pairs of states does a head survive to another day, and does removing each session's mean rescue the rest? | The Mind-Controlled Light Switch |
| Is individual alpha frequency more stable across days than reactivity or alpha's power, at one site with dry electrodes? | Alpha Flower |
| Do the 10 and 15 hertz responses lateralize between O1 and O2 with the eyes on the center, and which way? | BCI Flight Simulator |
| Is the age head's estimate the same at two sites within a person, inside a margin set in advance? | Age Detector |
| How much of a head's separation of two states is left when the blinks and muscle that come with them are matched? | Signal and noise, and the Mind-Controlled Light Switch |
| Session | Milestone | What happens |
|---|---|---|
| Meeting 1 | First-stage review | Each group presents its question, estimand, hypotheses, design, simulated power, analysis code and protocol, for critique and approval. |
| Meeting 2 | Pilot and first sessions | The protocol run end to end once, then first sessions. Collection continues outside class. |
| Meeting 3 | Second sessions | On another day, for designs that need one. Every file saved and named under the protocol. |
| Meeting 4 | Analysis clinic | Each group runs the planned analysis, unchanged, and another group reruns its code to the same numbers. Deviations logged. |
| Meeting 5 | Presentations | Half the groups: fifteen minutes each, with questions. |
| Meeting 6 | Presentations | The rest. Second-stage reports are due a week later, then every saved file is deleted, together. |
Record
| Hypothesis | The planned test | Result | Supported? |
|---|---|---|---|
| 1 | |||
| 2 | |||
| 3 | |||
| Exploratory |
Wearing it is a choice. A group in which nobody wants to wear it can ask a methods question of simulated data, or of the data others agreed to share, and a demo on a simulated IntoMind One, with ?dev on the end of its address, can stand in for the device while a pipeline is built.
The board's pin stays with its own 3.3-volt signal. Nothing on it plugs into a wall.
Saved files come from someone's brain: keep them only with the wearer's permission, under your protocol, and delete them all together when the capstone is finished. A study meant for publication needs a review board's approval before its first recording.