The project develops various tools and resources specifically designed for children and young people. The first tool is the "GenAI Teachable Machine," which is a simplified example of supervised machine learning.

The application is implemented so that images uploaded to it are not transferred outside the web browser. For this reason, it is a safer way to approach the world of machine learning and artificial intelligence compared to commercial alternatives.

Launch GenAI Image Classifier here

Machine Learning in Our Everyday Lives:

This application is actually a classifier

In terms of machine learning, in this course we focus primarily on supervised machine learning and especially on one of its sub-areas: classification. In classification, we observe an input, such as an image of a traffic sign, and try to infer its class, such as the meaning of the traffic sign. Other classification problems include identifying fake accounts on Twitter (the input can be a list of followers and information about how quickly followers have accumulated, and the class is either "fake account" or "real account") and recognizing handwritten numbers (the input is an image and the class is a number between 0, 1, …, 9).

From the Elements of AI online course, the classification section of the machine learning page https://course.elementsofai.com/fi/4/1

Application user interface

In practice, the user teaches the AI application in a guided manner using training data. Training data is material that the user prepares themselves. In practice, the material for this application consists of images that can be obtained from internet image libraries or captured using a webcam.

Note! The application is designed so that it can be used on a computer as well as on tablets and mobile phones.

Launch GenAI Teachable Machine here

A. Main view: GenAI Teachable Machine

This main view contains five different stages that guide the user through creating a supervised machine learning model. These stages are described below with screenshots.

Note! At the top of the classifier application there is a save button that allows you to save the classifier locally to your computer!

GenAI application main view

Stage 1: Creating classes and adding training data

In the image above is a simple rock-paper-scissors example that everyone understands should be divided into three categories (classes): rock, paper, and scissors. Additionally, there is a fourth class, empty, which contains no rock, paper, or scissors. Together, this forms the training data that is used to teach the AI. Do this:

  • Divide the material to be taught to the AI into classes, such as empty, scissors, paper, rock.
  • Add the classes to the GenAI application and name them.
  • Import a sufficient amount of training data into each class so that the AI will learn what, for example, rock looks like in the rock-paper-scissors example. This can be done using a camera or existing images.

Stage 2: Teach the machine learning model the data added in stage 1

The AI behind the GenAI classifier is based on supervised learning. A moment ago you created classes and imported the necessary training data into them. At this stage, the AI is taught to recognize the differences and similarities between the data divided into different classes. You don't need to do anything but press a button.

  • Press the "train classifier" button and wait until "classifier trained" appears.

Stage 3: Check how the classifier you just trained works?

Now it's time to explore how the guided machine learning model works? Can it distinguish between rock, paper, and scissors? You can easily see this from the percentage bars below the "input" camera window. You can return to stage 1 and improve the training data if the recognition is uncertain or if it even recognizes incorrectly. If you change the training data, you also need to train it by pressing "train classifier"

  • To know if the AI works correctly, it must be tested. Testing is done by showing the model a new image from the subject matter of the classified contents. In the UI image, I showed the "scissors" gesture with my fingers to the camera (visible at the input location).
  • Look at the confidence of the classification in the percentage bars below the "input" screen. If necessary, you can correct your model by going back to the training material (stage 1) and retraining the AI (stage 2)
  • When you are satisfied with the classifier's performance, press "next" and move to the next stage

Stage 4: Plan how the classifier responds to its observations

At this stage, you can plan how the classifier responds when it detects classified information in the input. For example, you can plan how the GenAI classifier responds when it detects scissors, paper, rock, or scissors?

You can add the following content:

  • Image: you can add (or drag) an image, which can also be an animated gif
  • Sound: you can add or record an audio file
  • Text: you can add text and format it with basic tools
  • Link: you can add, for example, a YouTube video or other web content (blocked when set to grades 4-9 in the application settings)

Stage 5: Result

This "result" window contains a preview of what the different functions look like. In the image, the classifier has recognized the "scissors" gesture and displays both the scissors as a gif animation and as text.

  • Note! When you press the "deploy" button, you will enter full-screen mode.

B. End view: AI hidden

The "Application" in full-screen mode and the AI hidden

In full-screen mode, the actual AI itself is already hidden in the same way it is hidden in, for example, a robot vacuum, social media, or even a car.

  • You can use the machine learning model using either a webcam or image files (bottom bar).
  • A camera preview window helps make webcam use easier
  • You can also adjust the volume.
Launch GenAI Image Classifier here