sooo i totally feel u on this. honestly finding stuff that isnt just a basic game or way too hard is a total nightmare. in my experience as a cs educator i would suggest checking out Machine Learning for Kids by Dale Lane. basically its a free tool that lets kids train their own models using text or images and then plug them into Scratch 3.0 Desktop... its literally perfect for a 3rd grader who knows basic blocks already.

also if she wants to understand chatgpt you might want to consider Hello Ruby: Expedition to the Internet for the logic stuff... but seriously make sure to look at AI for Oceans by Code.org too. its a free interactive module that teaches how training data works by sorting trash and fish. reallyyy simple but powerful way to explain bias and ethics without it being a snooze fest lol. gl with it!!

For your situation, I've tried a few things with my own kids and it's honestly a mixed bag.

* MIT Media Lab Learning Creative Learning with Scratch vs ReadyAI AI Lab for Kids: The Scratch extensions are okay, but I found ReadyAI AI Lab for Kids much better for 3rd graders cuz it actually explains the "why" behind the blocks.
* Khan Academy Khanmigo: This is a great AI-powered tutor. It's safe and monitored, but honestly, it can feel a bit like extra schoolwork if they aren't in the mood.

I'd start with the ReadyAI AI Lab for Kids free trials first since it's more hands-on. Gl! hope she loves it!

Soooo I'm a bit late to the party but I totally get the struggle!! Honestly, I was super nervous when my 3rd grader started asking about this stuff because of safety and like... weird data privacy issues? Plus, I didn't wanna spend a fortune if she just got bored after two days.

- We actually tried this one platform that used block coding to teach how AI "thinks" about pictures. It was GREAT because it had a free trial so we didnt have to commit right away.
- My biggest tip is to look for tools that dont require a real email or personal info for the kid. I'm always super cautious about privacy so the one we found let us stay anonymous which was amazing!!
- We just set a timer for 15 mins after school. It felt like a game to her but she was actually learning about data sets without it being "extra homework."

Honestly, just go slow and watch out for those hidden subscriptions!! It's such a wild west out there right now so staying cautious is highkey the move. gl!

Honestly, if you want to go the DIY route instead of just paying for a subscription, you should definitely check out Teachable Machine. Its technically a web-based tool from Google and its TOTALLY free. You dont need to code anything complex, but it lets her actually build a model using her own webcam or sounds. She can see how training data works in real time which is basically the core logic shes asking about. Its super cool because she can train it to recognize her favorite toy vs a snack, and it shows the confidence levels for each prediction.

• Teachable Machine for the hands-on DIY training
• Code.org AI for Oceans for the ethics and bias side of things
• Quick, Draw! to show how neural networks learn patterns I think starting with those experiment style sites is better than a full curriculum because its more about the aha moment. Just make sure your explain that the AI is only as smart as the data she gives it tho. Sometimes it gets things wrong if the lighting changes or whatever, which is a great lesson in how these systems arent actually magic. Its a fun way to learn the technical side without it feeling like a boring lecture.

I totally agree with the idea of sticking to the free or DIY route initially. Honestly, after years of trying different things with my daughter, I found that the software-only stuff tends to lose its magic pretty fast. If you want something that stays interesting for more than a week, you should look into hardware.

• Just get any AI-compatible robotics kit from a brand like Makeblock
• Look for general AI vision sensor kits that plug into basic microcontrollers
• Check out any of the educational kits from LEGO that mention machine learning In the long run, having something physical that she can actually touch and see reacting to her makes the logic click way faster. Weve had some of these kits for years and she still pulls them out to try new ideas. Its way more cost-effective because the hardware grows with them as they learn more complex logic. Plus, it makes the whole ethical side of things feel more real when she sees how a sensor can totally misidentify something just because the lighting changed... so keep it simple and focus on those big reliable brands!

Ngl, I think you should just look at any of the beginner kits from Makeblock. You really cant go wrong with their stuff for that age group since it combines the coding she already knows with actual hardware. My only real concern is compatibility tho... honestly, some of these AI apps are super picky about which tablets they actually run on. I would definitely check their site to see if your daughters device is supported before buying anything. It would be a total bummer to get her all excited and then find out the app crashes every five minutes because the OS is too old or the Bluetooth wont sync. But yeah, basically any kit from them that mentions AI or vision sensors is gonna be a solid start for a 3rd grader. It makes the logic feel way more real when a robot actually reacts to her face or voice instead of just staying on the screen.

Yep been there done that. Can confirm everything said above is spot on.

Interested in this too

Bookmarked, thanks!

Exactly what I was thinking

Just caught up on this thread and it is awesome you are starting her so young. Honestly I am still kind of a beginner when it comes to the technical deep dives but I have spent a lot of time testing how these tools actually perform in real-time with kids. Some are super laggy or dont give very clear feedback when the AI makes a mistake which can be really frustrating for an 8 year old. One quick question though, is she more into the results like making something cool or does she actually like the trial and error part? Knowing if she enjoys the testing and debugging side would really help narrow it down because some tools are better for one or the other. If you want to focus on how AI actually performs and why it succeeds or fails here are a few things I have tried that worked well:

• AI for Oceans by Code.org

This is basically the best intro for showing how data quality affects performance. It shows in real-time how the bot gets confused if you dont train it right and it is totally free.

• Cognimates

It is an extension of the Scratch idea but it feels a bit more like a professional testing environment. It lets her see the confidence levels of the AI in its answers.

• Quick Draw by Google

It is a simple game but it is great for discussing how a model performs better as it gets more data from millions of people. Basically I found that focusing on the benchmarks like how many times the AI gets it right versus wrong is a great way to teach the logic without it feeling like a lecture.

This ^