RAISE Playground: how to start MIT's free AI builder

To start RAISE Playground, go to playground.raise.mit.edu and click "Open playground", or go straight to the editor at playground.raise.mit.edu/main. It's free, it runs in a web browser, and there is no account or login: MIT's terms say the site collects no information about you and keeps no projects on its servers. Below is what it is, who makes it, what age it suits, and a first project a child can finish in 15 minutes.

This guide is from Peabody (peabody.kids), a free reading and learning site for kids, free through the end of 2026. RAISE Playground isn't ours, and nobody paid to be listed. We opened every page linked here, and built the project below in the editor, on 29 October 2026. The facts come from MIT's own pages: the home page, the About page and the terms. Our RAISE Playground page in the software directory has the facts table.

RAISE Playground at a glance

What is RAISE Playground?

MIT's home page calls it "a block-based programming platform" that lets anyone make projects with machine learning, robotics and AI engines: image and text classification, gesture recognition, music generation, reinforcement learning, social robotics and microcontroller robotics. The About page says the team built its blocks on a forked version of the open-source Scratch Blocks, so anyone who has used Scratch will know their way around in seconds: the same Motion, Looks, Events and Control blocks, the same stage, the same green flag.

It comes from two groups at MIT. RAISE (Responsible AI for Social Empowerment and Education) describes itself as "MIT's home for AI education"; it also runs Day of AI, MIT App Inventor and FutureMakers. The Personal Robots Group at the MIT Media Lab builds the robot and AI blocks. MIT asks anyone who uses the materials to credit both, and licenses them CC BY-NC (free for non-commercial use, such as a classroom).

Is RAISE Playground free, and does it need a login?

It's free and there's nothing to log into. The terms, last updated in September 2025, say: "We do not collect or save any information about you when you use the Site." They add that MIT doesn't track how you use it, doesn't log IP addresses and keeps no information about you or your projects on any MIT server. Your projects "belong to you."

Two catches, both from the same terms. First, nothing saves for you: download your project or save it to your own Google Drive, or it's gone when the tab closes. Saving to Drive puts the file under Google's rules, not MIT's. Second, it's a research tool. MIT says it will "strive to keep the Site accessible" but has no obligation to, so a teacher should have a plan B for the day it's down.

What age is RAISE Playground for?

MIT doesn't set an age. The About page says it is for "students and beginning programmers", and its three ready-made curricula are all for middle school:

• Dancing with AI: "a middle school project-based curriculum about making interactive, movement-focused AI systems", with example projects such as a jellyfish that reacts to emotions.
• How to Train Your Robot: "a 30-hour curriculum for middle school students", from the Personal Robots Group and i2 Learning, on machine learning and ethics with a small robot.
• RAICA: the home page's example project for it is a lamp, fan and radio controlled by blocks, opened in the /raica editor.

In practice, about 11 to 14 is the sweet spot. A confident 9- or 10-year-old who already knows Scratch can do the project below with a grown-up beside them. Younger children are better off with the gentler starts in Teachable Machine for kids or Scratch's own Face Sensing (Scratch AI extensions).

playground.raise.mit.edu/main, /create and the other addresses

People search for "raise playground main" and "playground.raise.mit.edu create" because the site has more than one editor address. We opened each on 29 October 2026:

• /main is the full editor that "Open playground" goes to. Its extension library (the button at the bottom left) has Text Classification, Teachable Machine, Tables, Selfie Detector, Hand Sensing, Face Sensing, Body Sensing, Object Detection, Music Creation, three robot extensions (PRG Microbit Robot, PRG Gizmo Robot, PRG Arduino Robot) and the usual Scratch ones. A few test extensions marked "(INTERNAL)" or "Example" also show up; ignore them.
• /create is the Dancing with AI editor, with Teachable Machine, Virtual Agent, and hand, face and body sensing. The home page uses a /create link to open its example jellyfish project.
• /raica and /httyr are the editors for those two curricula, with the robot blocks each one needs.

If a lesson gives you one of these links, use it. For anything else, start at /main.

Your first project: a kind-or-unkind message checker (15 minutes)

This uses the Text Classification extension, which needs no camera. Your child teaches the computer what kind and unkind messages look like, then tests it. We built it exactly like this on 29 October 2026; after six examples of each, it read "thanks, you are the best" as kind.

1. Open playground.raise.mit.edu/main in Chrome or Edge on a laptop or desktop.
2. Add the extension. Click the Add Extension button at the bottom left and choose Text Classification. Its blocks appear in the palette.
3. Make the first label. Click Edit Model, then Add a Label. Type kind in the box that says Class 1.
4. Give it examples. Type an example in the empty box and click Add Example. Add six: "you are great", "thank you for helping me", "I like your drawing", "you are a good friend", "well done", "that was so nice of you". Click Done Editing.
5. Make the second label. Click Add a Label again, call it unkind, and add six unkind examples your child thinks of (mild ones are plenty: "go away", "you are bad at this"). Click Done Editing, then Done.
6. Build the script. From Events, drag when green flag clicked. From Control, add forever. Inside it put ask (Type a message) and wait from Sensing, then say ( ) for (2) seconds from Looks. Into the say block, drop predict class for ( ) from Text Classification, and into that, drop the answer block from Sensing.
7. Click the green flag and type messages. Try new ones it hasn't seen.
8. Try to fool it. Type "you are not mean" or "great, another rainy day". It will probably get some wrong. Add those as examples under the right label and test again.
9. Save it. File, then Save to your computer. Write the examples down too, so you can rebuild the model if you need to.

What to talk about. "Who decided what counts as unkind?" Your child did, with twelve examples. Change the examples and the computer's idea of kindness changes with them. That's the big idea of machine learning in one sentence, and why people argue about what goes into the examples.

A privacy note from our test. When we trained the model, the browser downloaded Google's sentence-reading model (Universal Sentence Encoder Lite, from TensorFlow Hub and Kaggle) and none of the requests we saw carried the typed examples. The sorting happened on the computer. The extension's other blocks, such as speak, may use outside services; we didn't test those.

Using RAISE Playground in class

• No student accounts to set up. That makes it easy to approve under most school privacy rules, since MIT says it collects nothing. Check your district's list anyway.
• Use the curricula. Dancing with AI and How to Train Your Robot each link free lesson materials from their sites; How to Train Your Robot also has a student workbook and a robot setup guide, so check what hardware it needs before you start.
• Plan for saving. On shared computers, have students save to the computer or a class Drive folder at the end of every lesson.
• Pair students. One types examples, one tests. Swap halfway.
• Have a fallback. If the site is slow or down, Machine Learning for Kids or Teachable Machine teach the same idea.

What RAISE Playground can't do

• It doesn't keep your work. No cloud saves, no gallery and no sharing community like Scratch's.
• It isn't a chatbot. Children train small models and use them in projects; they don't chat with an AI.
• Some extensions need extra things. The robot blocks need the matching robot or board, and the Teachable Machine extension needs a model trained on Google's Teachable Machine site first.
• There's little help built in. MIT's pages offer no step-by-step start, which is why this guide exists. It works best with a teacher or a grown-up nearby.
• Devices aren't listed. MIT doesn't say which browsers or tablets it supports. We only tested a Chrome-based browser on a computer.

7 things to try after the first project

Each of these uses an extension we saw in the /main library on 29 October 2026. Only the text one was built and tested by us end to end; the rest are ideas from the extension names and MIT's pages, so try them yourself first and expect to experiment.

1. Make the checker smarter. Add a third label such as question and test it on messages that are neither kind nor unkind.
2. Swap the topic. Train "animal" against "food" words, or "happy" against "sad" sentences, and see which topic the model finds harder.
3. Use Hand Sensing or Face Sensing. These need a camera, so allow it only for the session and check the screen when you finish.
4. Try Body Sensing for a dance-style project, which is the idea behind MIT's Dancing with AI curriculum.
5. Bring in a Teachable Machine model. Train it on Teachable Machine, then paste its link into the Teachable Machine extension.
6. Try Music Creation if your child likes sound; the extension is in the library, and we have not tested what it sends out.
7. Break it on purpose. Give the model biased examples, such as only short unkind messages, and see what it gets wrong. Then talk about why.

5 mistakes first-timers make

1. Closing the tab without saving. MIT keeps nothing, so save to the computer first.
2. Too few examples. Six each worked for our simple test, but more varied examples make a steadier model.
3. Examples that are all alike. The model learns the pattern you show it, not the idea you meant.
4. Starting on a tablet or phone. We tested only a Chrome-based browser on a computer.
5. Using the wrong address. Lessons that give you /create, /raica or /httyr need that exact editor; otherwise use /main.

Alternatives to RAISE Playground

More tools sorted by age are in machine learning for kids by age, and about seventy free sites are in where to learn about AI.

Where Peabody fits

Peabody isn't an AI builder and doesn't teach coding. It's a reading and learning site: stories at a child's reading level in the library, and learning games where the questions are made and marked by code, not AI. After the kind-or-unkind project, a nice next step is to make a story on the free AI story generator and ask the same question: who decided what the AI would write? For a grown-up, the quiz is your child ready for an AI chatbot? turns the safety side into five minutes of questions.

Learn more elsewhere

• RAISE Playground: the home page, with the three curricula and their example projects.
• RAISE Playground terms: what MIT collects (nothing) and where projects go.
• MIT RAISE: the initiative behind it, and Day of AI.
• Dancing with AI: the movement curriculum and its example projects.

Questions parents ask

How do I open RAISE Playground?

Go to playground.raise.mit.edu and click "Open playground", or type playground.raise.mit.edu/main to go straight to the editor. It opens in a web browser with no download and no sign-in. Click the button at the bottom left to add AI extensions such as Text Classification or Teachable Machine.

Does RAISE Playground need an account or login?

No. There is no sign-in. MIT's terms say the site collects no information about you, doesn't track use or log IP addresses, and keeps no projects on MIT servers. To keep a project, save it to your computer or your own Google Drive from the File menu.

Is RAISE Playground free?

Yes, as of October 2026. It's a free MIT research tool, and its teaching materials are licensed CC BY-NC, so classrooms and families can use them without paying. MIT asks for credit to the MIT RAISE Initiative and the Personal Robots Group at the MIT Media Lab.

What age is RAISE Playground for?

MIT sets no age, and its curricula are written for middle school, so about 11 to 14 suits it best. A 9- or 10-year-old who knows Scratch can manage a simple project like the message checker with a grown-up nearby. Younger children do better with Teachable Machine or Scratch's Face Sensing first.

What is playground.raise.mit.edu/create?

It's one of the site's editor addresses: the Dancing with AI version, with Teachable Machine, Virtual Agent and body, hand and face sensing blocks. The home page links to it to open an example project. For general use, start at playground.raise.mit.edu/main, the full editor.

Is RAISE Playground the same as Scratch?

No, but it looks and works like Scratch. MIT built its blocks on a fork of the open-source Scratch Blocks and added extensions for training and using AI, robots and music. It has no community, sharing or cloud saving the way Scratch does.

Is RAISE Playground safe for kids?

It has no ads, chat, accounts or sharing, and MIT says it collects no information about users. In our test, training a text model happened in the browser. The usual care applies: save files somewhere you control, and if a lesson uses the camera, the browser will ask first and you can say no.

How long does the first RAISE Playground project take?

About 15 minutes for the kind-or-unkind message checker in this guide. Most of the time goes on typing the twelve examples, so let two children share the typing.

Does RAISE Playground need a camera or microphone?

Not for the text project in this guide. Some other extensions, such as hand, face and body sensing, use the camera, and the browser will ask before it turns on. You can say no.

What are good alternatives to RAISE Playground?

Teachable Machine, Machine Learning for Kids, Scratch with Face Sensing, micro:bit CreateAI and Day of AI. The alternatives table above lists the best age for each.

Can I use RAISE Playground on a Chromebook or iPad?

MIT does not list supported devices. We tested only a Chrome-based browser on a computer, so try it on a Chromebook first and do not count on a tablet.