Black Box
Work out a hidden rule by making predictions and comparing them with what happens.
When AI supplies an explanation, we can move on without trying to work it out ourselves. Here you make a prediction and test it, so you have something of your own to reconsider when the result is different from what you expected.
Signal
Check the evidence behind a convincing conclusion and decide whether it supports the claim.
AI can hand you a confident conclusion before you have checked its evidence. Here you inspect the source, judge the claim, and name the detail that supports your answer.
Keep a first thought of your own.
We can understand an explanation when somebody gives it to us and still struggle to come up with one ourselves. This activity gives us a chance to try before we see the answer, including time to notice what we do not yet understand.
The cost
You ask AI why a project stalled, how a formula works, or what a set of results means. It gives you an explanation, and the work moves forward. What can disappear from that sequence is the part where you form a view, notice what you do not understand, and decide what would settle it.
The concern in The Quiet Cost is that repeated assistance can remove opportunities to build or maintain the judgment you still need when an answer is incomplete or wrong.
What you practice here
Black Box gives you evidence without a finished explanation. You choose a test, predict a result, and compare it with what happens. A wrong prediction gives you something specific to revise. There is no timer because the useful effort is the reasoning.
The research connection
In three randomized experiments, a 2026 preprint found that AI assistance improved performance while it was available, but subsequent unaided performance was worse. The experiments involved mathematics and reading comprehension. They did not test Black Box or establish that visual puzzles preserve professional skills.
Liu and colleagues, AI Assistance Reduces Persistence and Hurts Independent Performance (2026) ↗
Take it back to work
Before asking AI to explain something, write down what you think is happening and one thing that would change your mind. When you read the answer, compare it with your own account so you can see what it adds and what still needs checking.
This studio offers practice. Its score is not a test of intelligence, a diagnosis, or proof of protection from skill loss.
A fluent answer still needs a check.
When a summary does the connecting for you, it can become hard to see which parts came from the evidence and which parts were added along the way.
The cost
AI can assemble a report, compare options, or explain a chart in seconds. If you carry the conclusion forward without opening the source, you miss the chance to notice a changed denominator, an unsuitable comparison, or something the data never measured.
When you decide to rely on an answer, you should be able to explain what supports it and what you have not checked yet.
What you practice here
Signal asks you to compare a claim with the evidence provided. Some claims are supported, some conflict with the facts, and others need information you do not have. You will need to decide which applies and point to the detail that led you there.
The cases and numbers are fictional. Their purpose is to make the reasoning visible, with an explanation after every answer.
The research connection
In three experiments published in 2023, people followed biased AI recommendations in a fictional classification task. Some of that bias persisted when they later worked without assistance. The study shows that recommendations can shape subsequent judgments in that setting. It did not test Signal or a workplace training program.
Vicente and Matute, Humans inherit artificial intelligence biases (2023) ↗
Take it back to work
Choose one important claim in the next AI summary you use and open its source. Check who or what was compared, what the numbers count, and whether anything important was left out before you pass the claim on.
This is an opportunity to practice checking. Improving at this exercise has not been shown to improve professional judgment outside it.
Spell It Out
You know how. Can someone else follow?
Play Spell It Out →Keep thinking for yourself and making time for people.
AI can help us finish work and find answers, and there are plenty of reasons to use it. But some of that work was also how we learned, and some of those questions gave us a reason to call a friend. I created Practice to help us keep making room for both. You can try a game here, work through a problem before asking for help, or choose something to do with another person. The aim is to keep those parts of life familiar while we use the help AI offers.
Your favorites
What would you like to practice today?
Choose an ability to explore the games and studios that use it.
Explore the activities
Use the star on a game or studio to keep it in Your favorites.
No games in this category yet. Choose Studios or All formats to see the activities here.
The two-minute daily check
Answer eight short questions about your habits and get two suggested starting points. This is a way to choose an activity, not a test of your abilities.
Would it be worse than cheating?
I asked more than sixty couples to imagine a partner using AI to steer an important disagreement. Try the scenario yourself, then compare your reaction with theirs.
When are you reaching for AI?
Think First
Try a problem yourself before asking AI for help. Five minutes is a starting point, not a rule about how long learning takes.
Why this matters →The Attempt Timer
Choose a problem you were about to give to AI and spend a few minutes working on it yourself. Write down your first idea, try two approaches, and say what you expect the answer to look like. Then you can compare the help you get with something you have already thought through.
Compare your attempt with the help you get
Use AI where it helps, then come back to compare its answer with your attempt. What did it explain that you had missed, and is there anything in its answer you still need to check?
Logged.
Threshold check-in
One question, once a week or so. When did you reach for AI this week?
Protect the beginner
If assistance handles every first attempt, a beginner can miss opportunities to learn how the work is done. Pick one part of a task you want to understand and try it yourself this week. Then seek feedback or guided help, including AI where appropriate, and check what you can do afterward.
Your decision log
This log keeps your first view, what AI added, and the answer you chose. It lets you look back at those decisions, but choosing an answer is different from knowing it was correct. Where you can, check what happened afterward too.
Find Your Words
Give yourself ten minutes to describe what is on your mind, before a tool supplies words for it. You can revise what you write or use it to begin a conversation.
Why this matters →The Blank Page
Nothing you write leaves this device. There is no cloud. There is no draft saved to anyone's server.
Read it back once.
Then: did anything get more specific while you wrote?
If you keep writing about the same problem without understanding it any better, consider talking it through with someone you trust. They may notice something you have missed or help you work out what to do next.
Three sentences.
Use these three sentences to work out what you would want another person to understand about the situation.
Use this to start a conversation
You can copy the sentences below if they help you explain what is going on. When you speak to the person, leave room for questions and for something they say to change how you understand it.
Test the word
A friend, therapist, or AI may suggest a word for what you are feeling. Before you accept it, think about which parts of your experience it explains and which parts it leaves out. These questions give you time to do that and decide whether another word would fit better.
Why these activities belong here
I wrote The Quiet Cost because I wanted us to take the everyday costs of AI seriously alongside the bigger dangers we hear about. Getting an answer can feel like the whole point of a task, but working toward it may also have taught us something or brought another person into our life. If AI keeps doing more of that work, we need to decide how we will keep those opportunities. The activities here give us things to try, both on this site and with people we know.
What practice here can do
The games give us something to remember, explain, notice, or work out. They have not been shown to prevent AI-related skill loss, and a better score does not tell us that we have become better at everything outside the game. Daniel Simons and his colleagues found much stronger evidence for improvement on practiced tasks than for broad improvements elsewhere. That is why this site also asks us to do the actual things we want to keep doing, including making decisions and spending time with people.
Read the brain-training review ↗ · Discussed in Chapter Thirteen.
Why Speak Unscripted
What I noticed while hiring in hospitality, and why it matters that we keep talking to people.
Why Ask a Person
The favor that led me to meet my wife, and what a person gets from being asked for help.
Why Stay Raw
How to prepare for a difficult conversation while staying open to what the other person actually says.
Why Think First
What changes when help arrives before an attempt, and what a school experiment found when AI was taken away.
Why Find Your Words
Why trying to describe a feeling matters, even when AI can suggest the words.
Why Stay Curious
What a parent and child can learn about each other while trying to answer a question together.
The costs that accumulate while a technology becomes a habit
These examples look at what people gained from a technology and what else changed as they kept using it. Some costs were hard to see, while others were recognized but received less urgency than the immediate benefit. That difference matters when we think about how to use AI.
When do you reach for help?
William told Michael that he had begun asking AI earlier in a task, sometimes at the first sign of difficulty. His work was still strong. What he noticed was a change in when he asked for help.
William’s story appears in Chapter Six and returns in Chapter Thirteen.
Try it in daily life
Try a task you want to remain able to do yourself. Make an attempt, use help where it is useful, then check what you can do without it.
Take time to find words that fit
“How am I going to pay for it? I don’t know. Bottle caps. What does it matter, just come and fix it.”
Michael’s mother, trying to get a tire repaired. “Bottle caps” became a shared family memory. Chapter Eight.
Try it in daily life
Friends, therapists, books, and AI can give you useful words. Check them against your experience. Write for a while, then decide whether you have a clearer description or something to tell another person.
What changed as the benefits became familiar?
These ten examples follow what people gained from a technology, what else changed, and why some costs became difficult to deal with. Sometimes an alarming prediction was wrong. Sometimes a warning was right and people still had reasons to keep going. The history helps us ask what we are accepting with AI, including costs that can seem less urgent than the dangers getting most of our attention.
Would it be worse than cheating?
In the fall of 2024, I asked a couple in our kitchen a question about AI and betrayal. Over the next several months I put versions of it to more than sixty other couples. After the first awkward attempt, I learned to build a scale with familiar betrayals before asking about the AI scenario. You can try a version of it here.
Rate each of these, 0 to 10.
I originally asked people to use a scale from one to ten, though some answered zero. This version allows zero too. Use it if something does not feel like a betrayal to you, and ten for the most serious end of your scale. These are your judgments, not right or wrong answers.
Now imagine this.
Imagine that you and your partner disagree about a decision that matters to both of you. While you are at work, your partner uses an AI version of you to practice a conversation about it, trying different questions and responses until they have a plan for getting you to change your mind.
The AI produces a script that your partner follows during the conversation. You change your mind, believing you have worked through the disagreement together.
Years later, you find the script on their laptop and realize how the conversation was planned. Think about what that discovery would change for you, including the next time you and your partner disagreed.
Your scale, their answers.
Why I was more worried
What would the discovery change?
I kept thinking beyond the moment of finding the script. What would it change the next time we disagreed, or the next time my partner tried to apologize?
The concern about later conversations
Imagine an argument months later. Your partner apologizes, but now you wonder whether they mean it or are following another plan. That was part of why I rated the scenario so highly. I wanted to understand what happens to a relationship when one person works out how to get agreement without giving the other an equal chance to change the conversation.
Preparation can help too
Thinking about what to say, asking a friend, or using AI to find less hurtful wording can help a conversation. The concern in this scenario is using a model of a partner to plan a route to agreement. Are you prepared to hear something that changes your mind?
What the answers can tell us
Daniel Gilbert and Timothy Wilson have studied how we predict our future feelings. Their work made me question my own number, which was ten, as well as the couples’ lower ones. I could explain why the scenario bothered me without claiming to know exactly how any of us would react. These were informal conversations, not a controlled study of AI’s effects on relationships.
Make time for a conversation
Try an exercise that leaves room for another person’s ideas and response.
Speak Unscripted
We get used to talking with people by having conversations in which we cannot know every question or response in advance. These exercises give you small occasions to do that, with people who are willing to take part.
Why this matters →Pick one to try
Start with one that makes you a little nervous but won't cost much if it goes badly. Choose another when you feel ready, or stay with the same one.
You expected: /10. Come back when it's done.
Your track record
How bad you expected each one to be, next to how it actually went. Use the comparison to notice what happened, without expecting the gap to shrink every time.
Ask a Person
Think about the last time somebody asked for your advice on something you knew about. You had a chance to help, but you also found out what was happening in their life. This studio gives us reasons to keep having those exchanges even when AI could supply an answer, so that both people still get something from taking part.
Why this matters →The Redirect
What were you about to ask AI? Is there someone you could ask who knows about it or would enjoy working it out with you?
Let people know what you can help with
People may not know what you would be happy to help with. Fill this in and send it to someone if you would like them to ask.
"I know a lot about ___, ask me before you Google it."
Someone asked you?
Think about what you got from helping, as well as what the other person needed. You can record the occasion here.
Reach for a person
If you have already talked a problem through with AI, think about whether there is someone in your life you would still want to tell. They may have experience that changes how you see it, and hearing about it gives them a chance to show that they care. You do not have to share everything. Choose someone you trust, and ask whether they have time to talk.
Stay Raw
Preparing for a difficult conversation can help us avoid saying something hurtful or forgetting what we need to explain. The person we speak to will have things to say too, so these exercises ask us to prepare while leaving room for a response we did not expect.
Why this matters →The Unrehearsed Hour
For two people who know each other well. Draw a topic neither of you has worked out in advance, put the phones away, and see what you have to say about it. You may hear something you did not know about the other person, or find yourself changing your own answer as you talk.
The Stop-Line
Before a difficult conversation, write down what you are feeling and what you need to discuss. You can prepare further if it helps you speak clearly or avoid saying something hurtful. Leave room for what the other person might tell you, rather than planning every reply on the way to their agreement.
Stay Curious
When an answer is available immediately, it is easy to look it up before thinking about it ourselves. Try leaving a question open long enough to make a guess or talk it through with someone, then see what you learn when you check.
Why this matters →The Question Shelf
Save a question you were about to look up, as long as it is something that can safely wait. When you return, write down your best guess before checking the answer.
Sit in Boredom
The screen goes dark for a short pause. See what you notice without filling the time. You may have an idea, or you may just spend a few minutes without another input.
Parent mode
Try asking your child a question when you do not already know what they will say. Listen to how they explain their answer, and let that give you something else to ask about.
When your child asks a question that can wait, ask what they think before looking up the answer together. They get a chance to try an idea, and you get to hear how they are thinking about it.
Getting comfortable talking to people
“They talk about the uncertainty of it, and how that paralyzes them.”
A decade of hospitality hiring. Chapter 2 of The Quiet Cost.
Try it in daily life
Choose an ordinary interaction you feel able to try. Record how difficult you expect it to be, then how it actually felt. You may discover a gap, or that a particular situation really is difficult for you.
A small reason to make contact
I met my wife because my roommate needed a ride. I would never have known what I missed if he had found another way there. That is part of why I worry about making every small request unnecessary. We do not always know what else could come from helping someone.
The designated-driver night. Chapter Three of The Quiet Cost.
Try it in daily life
Ask something the person can freely decline. A request can give you an occasion to talk, but nobody owes you a favor or a conversation.
Prepare for a conversation that can change your mind
In the scenario that started the book, one partner uses an AI version of the other to rehearse a route to agreement. I wanted to know what that would change between them, including whether the person following the script would still be willing to hear something that changed their own mind.
The scenario begins in the Prologue and returns in the Epilogue.
Try it in daily life
Name what you feel and what you need to discuss. A friend, counselor, or AI may help you prepare. Leave room for something you did not expect when the other person speaks.
Give a question some time
The discussion of I Am the Cheese follows a child and a parent working through a question together. Part of what matters is the chance for the child to offer an idea and see the parent respond.
Chapter Nine of The Quiet Cost.
Try it in daily life
Choose a question that can safely wait. Try an explanation of your own before looking it up. Do not delay information you need for a practical or safety decision.
The Reach
Use the definition to try to remember the word before revealing the answer. If you cannot find it, check whether it is a word you knew but could not bring to mind.
Why retrieval matters →Hold This Question
Make a guess before looking up the answer. Once you have written it down, you can leave the reveal sealed overnight and see whether another explanation occurs to you. Come back when you are ready, compare your thinking with the answer, and try another question.
Why the interval matters →The Panic
Try to identify the technology behind each historical warning or finding, then check the source. The examples give us reasons to examine our expectations about AI, without assuming that a prediction today is wrong because an earlier one was.
Walk the full timeline →The Route
Directions the way a neighbor might give them. Each leg of the directions shows once, then it is gone. Try to keep the whole route in your head, because the questions at the end will come out of order. This asks you to hold more of the route yourself than turn-by-turn directions do.
The navigation research →The Tell
Four pairs of practice messages. Choose the response that seems more grounded in the situation, then compare your reading with ours. The examples do not have verified authorship, and style cannot tell you whether a person or AI wrote them.
How to judge the message →What does the message give you to go on?
A personal detail can be invented. A polished sentence can be written by a person. Look at what the message says and what you can check before deciding how much to trust it.
Does it respond to this situation? Compare it with what was actually said. A warm reply may still miss the question.
Which details could you check? Specificity gives you something to investigate. It does not establish that a detail is true.
What does it assume? Watch for feelings, motives, or promises that the writer has no basis for supplying.
What would change your mind? An original source or clarification may settle something the wording alone cannot.
What remains unknown? Typos, grammar, warmth, and punctuation do not reliably identify who wrote a message. If authorship matters, you need evidence beyond style.
The Writing Room
Choose a prompt and try writing your response without autocomplete or suggested wording. You can take time to work out what you mean and change your mind as you write.
Why reaching for words matters → The cost no one can feel →Lend an Ear
You can care about someone and still struggle to find a useful response. Practice acknowledging what they said, asking a question, and deciding whether advice is wanted.
The study behind this →Practice saying what you heard
Researchers studied how people expressed empathy in conversations with AI role-play partners. Their work helped inspire this independent exercise. Our studio is not their platform or tested coaching program.
Try it in daily life
Ask what the person means, acknowledge their feeling, and check whether advice is wanted. Warmth With Backbone also asks you to respond without endorsing a belief you cannot verify.
The Unpolished Message
Write the part of a personal message that only you can supply. What happened between you and this person? What do you want them to know? Start here in your own words. You can edit it before sending.
Why your own first attempt matters →Keep your meaning in the message
A thank-you means something because you noticed what a person did and took time to tell them. AI can help organize the words. It can also supply feelings, memories, or promises you did not mean. Starting with your own account gives you something to check the draft against.
The personal-message practice in Chapter Thirteen of The Quiet Cost.
Try it in daily life
Use help with grammar, translation, or organization if you need it. Remove any detail, feeling, or promise you cannot stand behind. Roughness does not establish authorship or sincerity.
Companion Check-in
If you turn to AI for comfort, take a little time to consider what it is helping with and whether you are still involving the people you want in your life. This check-in asks about your own use, including what happens after the conversation is over.
The two things a companion can be →What happens outside the conversation?
AI conversations can offer comfort and a place to rehearse. They can also take time that might have gone to someone else. The research considers both possibilities, and this check-in asks what is happening in your own life.
Try it in daily life
Think about what the conversation helped you with and who else you might want to tell. You can feel better after talking to AI and still want a friend to know what happened or have a chance to share their own experience.
The Repair
After a disagreement, it can be hard to acknowledge your part and find a way to talk again. This studio lets you try that with a fictional situation. Use the preparation to listen more closely when you speak to a real person, without expecting them to accept your apology or respond on your timetable.
Why repair matters →What happens after a disagreement?
The book asks what happens if an agreeable machine starts to replace the people who can disagree with us. One concern is losing occasions to acknowledge hurt and work through differences with someone whose needs are different from ours.
Repair is discussed in Chapters Three, Four, and Thirteen and the Epilogue.
Try it in daily life
A draft can prepare you to acknowledge hurt. It cannot establish that the other person feels understood. Listen to their response and respect a request for time or distance.
Better individual work can still look more alike
If you only see your own finished story, you may not notice how much it resembles everybody else’s. That is a different question from whether you are pleased with the help.
Keep room for an alternative
Writing Room leaves the first draft to you. Three Before One asks you to try more than one idea before seeing a suggestion. These are practical ways to keep independent generation in the process, not tested remedies for a loss of creativity.
The Lost Qualifier
Compare a research finding with a summary that makes a bigger claim. Look for the word or detail that changed who it applies to, how certain it sounds, or what it says caused the result. Each set has three rounds, and you can try another set afterward.
Why this one is a game
A summary can drop the words that limited a finding to a particular group or situation. Compare it with the original before treating the broader version as established. The examples here are constructed to make those changes visible.
The research notes distinguish experiments, associations, reviews, and practice suggestions. Those labels describe the source, not proof that the activity itself works.
The Rebuild
Read an explanation for forty-five seconds, then put it into your own words. Compare your version with the original and decide what you want to revisit. Missing a detail can reflect recall, attention, or understanding. The game cannot tell you which.
Why the answer is not the knowing →Try explaining it yourself
A clear explanation can help you learn. Trying to explain it yourself gives you another way to check what you can recall and where you need more work.
Try it in daily life
Read the explanation, reconstruct what you can, and compare it with the original. A missing piece is somewhere to look again, not proof that you never understood it.
Other learning papers discussed on the site
- Faster Completion, Less Learning
A quasi-experimental analysis of learning records. AI exposure is inferred through the study design, not observed for every learner.
- Decomposing the Doer Effect
Associations between programming practice and learning, not an experiment establishing AI-caused loss.
- Students’ Practices and Skills in the LLM-Era
An analysis of Reddit posts, not a direct measurement of changes in skill.
- Is Solving Better Than Evaluating GenAI Solutions?
The study found no significant difference in exam performance or course grades between the relevant conditions. It does not prove that evaluating solutions produces worse learning.
These links are to preprint versions. None tested The Rebuild.
Three Before One
Try three possible answers to a question before seeing the research. For the second, change an assumption you made in the first. For the third, use the prompt you are given to consider another possibility. Then compare your thinking with the explanation.
The cost no one can feel →Reckonings
The restaurant memory game. Looking for the newsletter? Read Reckonings here.
In this restaurant game, you take an order without writing it down, then try to remember what each person ordered, deliver it to the right seat, and split the bill. A reckoning is an old word for the check, which you will be asked for at the end. There is a new table each day and a practice floor if you want to try more.
Why waiters became a memory science →Keeping track of who ordered what
Waiter-memory research helped inspire this game. In a small study in Buenos Aires, matching orders to customers became harder when the customers changed seats. The finding points to the role of location and person cues in that task. It does not demonstrate a general memory advantage or an effect of AI.
Bekinschtein, Cardozo & Manes (2008), Strategies of Buenos Aires waiters ↗
Try it in daily life
Connect the order with the person and their place at the table. This is one strategy to try within the game. It does not replace written records when accuracy matters.
By Heart
A contact list lets us call somebody without remembering their number, so we rarely have to practice doing it. Here you study a small phone book and then put it away. When a situation calls for one of those people, you try to recall the number yourself.
Why keep a few numbers in mind →Remembering a number when you need it
A contact list makes it easy to call someone without knowing their number. By Heart gives you a reason to try remembering numbers again. Improvement here tells you about this task, not the state of your memory as a whole.
Memory, offloading, and practical activities are discussed in Chapters Two and Thirteen of The Quiet Cost.
Try it in daily life
Try grouping the digits and connecting them with the person. Keep useful contacts stored too. Remembering information and having a reliable backup can both be helpful.
Where to go from here.
Here you can find the book, the research connected to the activities, and other people writing about these questions. There are also links to support services if you need help that a practice activity cannot provide.
The book
The book examines AI’s possible costs for learning, judgment, and relationships. Chapter Thirteen offers daily, recurring, and occasional ways to keep those parts of life active while using AI.
The newsletter
Michael’s free weekly look at AI stories and studies, what they could mean for everyday life, and a related Practice activity and reflection.
Read about the evidence and the limits of these games →
The evidence
These sources explain why we included the activities. Each entry distinguishes what the research found from what has not been tested here.
Ong, Goldenberg, Inzlicht & Perry. A careful map of where machine empathy helps and where it hits its limit.
Kumar et al. Coaching improved responses to AI role-play partners, as judged by independent human raters. Transfer to everyday conversations with real partners was not established. Our studio is an independent exercise, not the tested intervention.
Li, Folk, Singh, Ungar & Dunn. In this trial, a reduction in loneliness was observed only in the human-conversation condition.
Reporting on AI-assisted study and later recall. A reason to check learning after help is removed, with the limits of the study kept in view.
The wider conversation
Other writers and researchers considering related questions. Linking to their work does not mean agreeing with every conclusion.
Four decades studying what our machines do to our relationships, aimed straight at the companion moment.
Joseph Weizenbaum’s ELIZA showed how people could respond to a simple chatbot as though it understood them. His concerns are part of the longer history behind today’s discussion of AI companionship.
Kevin Kelly considers what people value when digital copies become easy to make and distribute.
For parents and teams
Evidence-based guidance for families, educators, and clinicians on kids and screens.
On designing technology that respects human attention and development, with a youth toolkit.
Jonathan Haidt on the growing-up years. The Quiet Cost’s education chapter adds a related question about AI: does the help leave a child more able to work independently afterward?
If you need a person
If you are struggling or need more support than these activities can offer, these services can help you find a person to talk to.
988 Suicide & Crisis Lifeline (US). Call or text 988. 988lifeline.org ↗
Crisis Text Line. Text HOME to 741741. crisistextline.org ↗
SAMHSA National Helpline. Call 1-800-662-4357, free and confidential, 24/7. samhsa.gov ↗
Outside the US: Find a Helpline ↗ lists free lines by country.
To find a therapist: Psychology Today directory ↗.
Use your practice record on another device
The site works without an account, and guest favorites stay in this browser. With an account, your account favorites and practice record can follow you between devices. The daily games also open their back catalog: any day you missed, back to ten days before you joined, is yours to play.
What an account stores, and what it never will
Stored: your chosen favorite activities and your practice record. Reps done, attempts logged, questions held, words reached, panics called, routes recalled, check-in results, and the notes you chose to keep in your decision log.
Never synced: your writing. The Blank Page, the Stop-Line, the Writing Room, Lend an Ear, the Unpolished Message, the Companion Check-in, the Repair, the Rebuild, and Three Before One all stay on your device and never reach our server. And never your behavior on this site. No analytics, no tracking, no engagement metrics. That part of the deal does not change.
Sign-in: a password you choose, or a one-time email link if you would rather not keep one. Passwords are stored hashed by the sign-in service, never in plain text and never by us.
Your practice record
Look back at what you tried, when you did it, and your results in the games. The record stays on your device and syncs when you are signed in, so you can return to it elsewhere.
About Practice
I created Practice for people who want to act on the concerns in The Quiet Cost. We can use AI and still do some of the thinking ourselves, work on something with our children, or ask a friend for advice. The activities here give us places to begin. Some are games, while others ask us to do something away from the screen. You do not need this site to keep those habits, but I hope it makes it easier to find something you want to try.
Our design rules
Practice outside the site too. A game can give us something to remember or work out, but it cannot take the place of the actual conversation, decision, or shared task we want to keep in our lives.
No chatbot. These activities leave the thinking and the conversations to you.
Nothing leaves your browser unless you say so. By default your logs live on your device and nowhere else. If you choose to make an account, your practice record syncs so it can follow you between devices. Your writing never syncs, and there are still no analytics. You can choose whether to create an account.
Evidence with its limits. The explanations distinguish experiments, correlations, and the book’s arguments. The games themselves have not been shown to prevent skill loss or improve abilities outside the practiced task. Read more about those limits.
The book
The Quiet Cost looks at what earlier technologies gave us, what people lost as they came to depend on them, and why I expect us to face similar choices with AI. It includes things we can do in everyday life to keep our skills and relationships active.