AI for Oceans: a guide for parents and teachers, level by level

AI for Oceans is a free, one-hour Code.org activity in which a child trains an AI to sort fish from rubbish in the sea, then teaches it words like "angry" or "fun" and sees it pick up their own opinions. It has eight levels, three of them short videos, and works without an account for grades 3 to 12. This guide was written by Peabody (peabody.kids), a free reading and learning site that is free through the end of 2026. AI for Oceans isn't ours, and nobody paid to be listed. Everything below was checked on Code.org's own pages and teacher lesson plan in October 2026.

One thing you'll notice: as of October 2026 Code.org's pages carry the name CodeAI, and its footer says "Code.org" is a former trademark. The address code.org/oceans still works and takes you straight to the activity.

AI for Oceans at a glance

What happens in each level of AI for Oceans

Code.org's teacher lesson plan sets out every level. Here's what your child will actually do.

Level 1: video, "AI: Machine Learning"

A short video on what AI and machine learning are, with everyday examples like email filters, auto-complete, video recommendations, voice recognition and translation apps. In class, the plan suggests watching it together and asking: where have you seen AI in your own life?

Levels 2 to 4: train AI to clean the ocean

The child is shown one picture at a time and presses Fish or Not Fish. Each picture and label becomes training data. Then "A.I." (the character) looks at 100 new pictures on its own and keeps the ones it thinks are most likely fish. A child who labelled well sees an ocean full of sea creatures with little or no rubbish. Later levels in this part widen the set to other sea creatures that belong in the water, not just fish.

Behind the scenes, Code.org says every picture goes through a neural network already trained on ImageNet, a database of more than 14 million labelled images.

Level 5: video, "AI: Training Data & Bias"

This video makes the key point: the labels the child has been pressing were the programming. The examples they gave are the training data, and the results depend on them.

Level 6: teach A.I. a word you choose

Now the pictures are cartoon fish, randomly built from parts: mouths, tails, eyes, scales, fins, plus a random body colour, shape and size. The child picks a word from a list and shows A.I. which fish fit it. The words here are fairly objective. A.I. starts knowing nothing and learns from the child's choices by looking for patterns in the fish's parts.

Level 7: video, "AI: Impact on Society"

The third video brings it back to the real world: AI learns from the data we give it, and that data can be opinion-based or biased. The plan's suggested question is a good one for home too: can you think of a time a machine got something wrong for you, like voice recognition not understanding you?

Level 8: teach A.I. a new word

The words now are deliberately subjective, like "angry" or "fun". Each child decides for themselves what makes a fish look angry. Two children who choose the same word can get very different oceans, because each trained A.I. on their own opinion. This is where the bias lesson lands.

On the final results screen there's an information button. It shows which fish parts A.I. leaned on most, and a child can click a single fish to see why it was chosen. Older children find this the most interesting part.

Two tips from Code.org's plan: the activity doesn't tell a child whether they did a level "correctly", so it's possible to click through too fast; and children who finish early can go back to Train More or pick a New Word.

The bias lesson, and why it matters

Level 8 is the point of the whole hour. A child trains A.I. to find "angry" fish, and A.I. learns exactly what that child thinks angry looks like. It doesn't know what angry means. It only knows what one person showed it.

Code.org's own extended example shows a second, sneakier problem. A student trained A.I. on "angry" fish using eyebrows and mouths, but A.I. also leaned on colour, because many of the fish the student happened to pick were purple. The fix was more training, including purple fish that weren't angry. That's bias from an unbalanced sample, not from anyone's opinion, and it's the same thing that goes wrong in real AI systems trained on data that leaves people out.

The plan's suggested ways to reduce bias are simple and true: use a larger and more inclusive set of training data.

Questions to ask after AI for Oceans

Pick two or three. The first four follow the prompts in Code.org's lesson plan.

1. How did A.I. decide what to put in the ocean? (Answer to listen for: from the examples you gave it.)
2. How did your labels change what A.I. did? Did anyone label something wrong on purpose? What happened?
3. When you taught it "angry", whose idea of angry did it learn? Would your friend's angry ocean look the same?
4. What's a problem you care about that AI could help with?
5. Can you think of a time a computer got something wrong about you? A voice assistant, a video suggestion, a spelling fix.
6. If A.I. had only ever seen pictures of one kind of fish, what would happen when a different one swam by?
7. Who should decide what counts as rubbish, or as a fish? Is there always one right answer?

For a younger child, keep it to one question: "Did A.I. know what a fish was, or did it learn it from you?"

How long does AI for Oceans take, and what age is it for?

Code.org's teacher plan budgets 45 minutes, and its catalogue lists it as about an hour, for grades 3 to 12. At home, plan on the full hour so there's time to talk at the end, and sit beside a child who isn't reading fluently yet, since the instructions are on screen. Children younger than grade 3 can enjoy the fish-sorting levels with help, but the bias levels may go over their heads. For them, our AI explained for 7-year-olds page is a gentler start.

Do you need an account for AI for Oceans?

No. Code.org's course settings mark it as not requiring a login, and it doesn't need the student to be 13 or over. A child can just open code.org/oceans and start; signing in is optional. Code.org's lesson plan also suggests a printable certificate for each child who finishes.

AI for Oceans and the Hour of AI

AI for Oceans began as an Hour of Code activity, and its lesson plan still mentions the Hour of Code. It's now part of the Hour of AI, the one-hour activities from Code.org and CSforALL that took over from the Hour of Code in 2025, and it's listed on the Hour of AI site. The Hour of AI site says this year's Computer Science Education Week is December 7 to 11, 2026. Any week works for doing it at home, though. Our guide to the Hour of AI at home has more one-hour activities.

What to do after AI for Oceans

• Train your own model with a webcam. Google's Teachable Machine lets a child teach a computer two toys or three hand shapes in ten minutes. Our Teachable Machine walkthrough has a first project and eight ideas, including a recycling sorter that picks up where AI for Oceans leaves off.
• Build a game with a model. Machine Learning for Kids goes one step further: train a model, then use it in Scratch. Our Machine Learning for Kids guide walks through one project at home.
• Older children: Code.org's How AI Works videos cover neural networks, chatbots, image makers and bias in short lessons for grades 6 to 12.
• Parents: see our guide to Day of AI for parents.
• Read about the sea. In Peabody's library, Under the Sea is an easy early reader, and Eight Arms is a short nonfiction book about octopuses. Both are free to read with no sign-up.

Where Peabody fits

Peabody doesn't teach how AI works; AI for Oceans does that better than we could. Peabody is a reading site. Some of its stories are written by an AI model, and every one passes safety screens before a child sees it. One of those screens is itself an AI model, which learned from examples, so question 7 above is a real one for us too. If your child wants to see AI making something rather than sorting it, the free AI story generator page shows how Peabody's stories are made.

Using Peabody in your homeschool

AI for Oceans makes one good science afternoon for a homeschooled child of about 8 and up: the hour itself, then ten minutes on two of the questions above. To stretch it into a short sea week, add 10 to 15 minutes of reading on the other days from Peabody's library, such as Eight Arms, or Under the Sea for a younger sibling, and ask your child to tell you what they learned. The week after, Teachable Machine is a natural second lesson. The honest limit: Peabody doesn't teach how AI works and keeps no record of AI for Oceans, so print Code.org's certificate if you keep a portfolio. More science ideas are in free homeschool science and on the homeschool hub.

Learn more elsewhere

• AI for Oceans: the activity itself; no account needed.
• Hour of AI activities: more one-hour activities for every age, each with its own account rules.
• How AI Works: Code.org's short videos and lessons on machine learning and bias, grades 6 to 12.
• Teachable Machine: train your own model with a webcam, no account.

About seventy more free AI sites for kids, parents and teachers are in our guide to where to learn about AI.

Questions teachers ask

How long does AI for Oceans take?

Code.org's lesson plan budgets 45 minutes: a 5-minute warm-up and 40 minutes for the activity, which includes three short videos. Its catalogue lists the activity as about one hour. Fast finishers can go back to train more or choose a new word.

What grade is AI for Oceans for?

Code.org lists it for grades 3 to 12. Younger children can do the fish-sorting levels with a grown-up, but the later levels about bias and opinion work best from about age 8.

Do students need an account for AI for Oceans?

No. Code.org marks the course as not requiring a login, and there's no age-13 requirement. Students can start at code.org/oceans without signing in. Teachers can use accounts if they want to see a class's progress.

Is AI for Oceans still part of the Hour of Code?

It's now part of the Hour of AI, which Code.org and CSforALL run in place of the Hour of Code. AI for Oceans is listed on the Hour of AI site, and as of October 2026 Code.org's catalogue tags it as an Hour of AI activity.

What is the bias lesson in AI for Oceans?

In the last level, students teach A.I. a subjective word like "angry" or "fun". Each student's A.I. learns that student's opinion, so two students can get different results for the same word. It shows that AI learns whatever is in its training data, including human bias.

Does AI for Oceans work on tablets and phones?

Code.org lists computers, Chromebooks, tablets and phones as good devices for it. A bigger screen makes the results screen and the information view easier to read.