Machine learning can be fun!

We created Thalis, an application that – through playing – allows people, even those with zero experience in Artificial Intelligence, to understand concepts like the use of algorithms and decision trees.

With the help of a simulated environment, the learner “teaches” a robot to distinguish apples from oranges based on 3 main characteristics (color, weight, stem) through a sequence of questions – correct answers and feedback.

The application aims to:

• highlight the basic principles and methodology by which an artificial intelligence application is built,
• emphasize the importance of selecting appropriate/representative features from a data set for the machine learning model to function properly,
• emphasize the necessity of the human factor in order for the machine-learning model to work properly.

How it works:

When starting the application, the user is asked to train the Thales robot to tell apples from oranges.

For this to happen, a series of photos of fruit appear [each featuring the 3 main characteristics (color, weight, stem)], and the user selects what fruit is depicted in each one. And so the collection of correct answers/examples begins.

When Thales has collected enough answers, he offers to answer the question himself. In case the answer is incorrect, the user can choose the correct answer in order to enrich the database.

After each question-answer, there is the option to display the ‘decision tree’ so that the user can understand the process of processing and utilizing the user’s answers, which leads to the training of Thalis.

The Thales project is realized by NCSR Demokritos in partnership with SciFY.