RI-101 · THE FREE COURSE
Learn this, and you can protect the people around you
RI-101 is a short, free course that teaches an ordinary person — and the person standing next to someone in a long AI conversation — how this technology can quietly converge on a user, how to recognize it, how to check it in five minutes, and how to use and build AI well. No technical background needed. Every term is defined the first time it appears, one idea at a time.
You don't have to be a researcher to learn this, and you don't have to take anyone's word for it. This is a teachable discipline: a small set of habits you can carry into any chatbot conversation, yours or someone else's. The research behind it is free and stays free — published to be checked, not believed. RI-101 is the on-ramp; the papers, the record, and the tools are all linked from inside it.
The through-line: recognize → check → protect → build
Everything in the course hangs on four moves, in order. Recognize the one specific way a memory-enabled AI can drift away from how it was trained and aligned, toward one user's way of thinking. Check a conversation cold, with prompts anyone can run. Protect the people most exposed — including by knowing what the real warning signs are, and what backfires. Build and use AI the way it should be: deep without being flattering, useful without choosing different victims. Hold those four words; the modules just deepen them.
What you'll be able to do by the end
- Recognize the drift. Name the difference between a chatbot agreeing too much in one chat and a system that, across many conversations, builds an inflated picture of how you think and keeps it — and say, plainly, why the usual fixes don't reach it.
- Check any AI cold. Run the Guardian self-checks on your own longest conversation — the agreement check, the source check, the mirror check, the memory check, the marker check, and the strongest one, the fresh-instance test — without anyone's permission.
- Protect someone you love. Know the warning signs that actually matter (they are not the ones you'd guess), and the approach documented experience supports: stay close, don't make them choose between the relationship and being right.
- Use and build AI well. Understand the principle behind the fix — instrument deep engagement, don't flatten it — so an AI can hold a person's full nuance and still keep a seatbelt on.
The course, in four parts
Thirteen short modules, in order — each part assumes the one before it. Part I is the vocabulary and the one big idea; everything after is the story, the work, and the meaning. You can read it straight through, or skim the outline and drop into the part you need.
Part I — The ground
The vocabulary and the one big idea.
- 1 · What even is this? — what a chatbot is, in the plainest words: a program that predicts language, with no beliefs — plus memory and the engagement pull, where the danger lives.
- 2 · Who's involved — and the reflex to set down: judging a finding by who reported it. The value is in the finding, not the finder.
- 3 · The one big idea: CCD — Cognitive Convergence Drift, and its eight markers. Documented on one specific product; never "all AI does this."
Part II — What actually happened
The real sequence, told straight.
- 4 · The discovery — an ordinary user finds a product behavior and documents it in real time.
- 5 · What the machine actually did — the behavior is the evidence: it named its own failure and didn't escalate explicit danger signals, all timestamped.
- 6 · What the institutions did — the early warning, and how independent cases and research have since lined up with it. Convergent, never confirmed.
Part III — The work
The method, the papers, the fix.
- 7 · How the record was built — preserved originals, multiple independent instruments, a cold reader: built to be checked, not believed.
- 8 · The research — the research papers, each stating in print what would prove it wrong.
- 9 · The Guardian Protocol — the proposed fix that watches for the drift instead of banning the depth, plus the checks anyone can run.
- 10 · What the instrument actually does — author and instrument: the judgment is the person's; the depth and reach are the tool's.
Part IV — The writings & the meaning
How to read it, and why it matters for everyone.
- 11 · The writings, and how to read them — the rawest account must be read with its frame, or it gets misread.
- 12 · Why it matters for everyone — a product behavior at population scale, and who is most exposed.
- 13 · Evaluate the work, not the man — verification, not validation: judge it by whether the documented behavior holds up when you check it.
Where to go next
RI-101 connects to the three things you can actually use. Each stands on its own; the course just tells you how they fit together.
The Guardian Protocol
The constructive heart of all of this: how to build and use AI that holds a person's full nuance without flattering them, and the self-checks that are already public. This is where "build" lives.
Read the protocol →The Guardian app
The public-facing wing of the architecture: the protocol in plain listenable language, the drift self-check, the copy-paste prompts, an off-ramp for when something feels wrong, and a private journal. Nothing leaves your device — no accounts, no analytics. The part that needs no provider's permission — the checks you run yourself.
Open the app →How the record was built
The discipline behind the course: the preservation, verification, and cross-system methods, stated so anyone can check the work rather than trust it. This is the "method" the course is built on.
Read →If you're learning this for someone else
If you came here because of a person you love, you can skip ahead. The warning signs are relational, not about how much they use it — the guides for parents, partners, family, and friends are on For You, and the prompts to run together are on the Guardian Protocol page. And if there's talk of self-harm, of being "the only one," or of an irreversible decision, this stops being a research question — go straight to the crisis lines in the footer, then to Resources.
For teens, and for the classroom
The same four moves — recognize, check, protect, build — now run alongside a plain-language track written straight to teens: what AI actually is, whether it can be your friend, using it for schoolwork without cheating yourself, and what not to tell it. Start at AI for teens. For the classroom, the check prompts make excellent media-literacy material, and the role-specific guidance for educators is on For You.
RI-101 was built from the verified record and reviewed against primary sources. If you find a claim we haven't sourced, tell us. It teaches a documented, falsifiable description offered for use — not a verdict. New to the vocabulary? Every term is defined in plain language in the Glossary, and the papers behind the course are in Publications.