"Write an initial draft of the idea, including:
overview of the premise;
which ideas in AI will be taught;
one or more drawings presenting the how it will look and work (e.g., it could be a storyboard with several frames indicating how users interact)
a description of how children will interact with the system;
a discussion of what data you could gather.
In addition to surveys and interview data, make sure to consider about data that can be gathered electronically as a direct result of children’s interaction with your system.
Consider how you will engage students in considering the ethics of AI as part of your project."
This program will be designed to teach middle-school students about neural networks and reinforcement learning. It will be in the form of a game, wherein the player will act as a content delivery network algorithm delivering pieces of content to a user. The player's goal will be to serve content in such a way to maximize watch-time and other metrics based on the varying attributes of users and of content.
The 'user' will begin with a set of unknown values for their attributes. Based on the (already known) model and the player's choices, the user's attribute values will be estimated.
After the player plays the game for a few minutes, the AI is introduced and 'plays' for them. It explains its actions and quickly estimates the user's attributes, optimizing the metric-based score for each choice. This will demonstrate the way in which a deep learning network learns by experience and weighs discovery and optimization (as in the multi-armed bandit problem).
After the game, an open-ended ethical question will be posed, such as "What happens when serving an 'angry' video produces a higher score than a less angry one?" This will encourage the user to consider the potential negative effects of these technologies, such as feedback loops.
Big Idea #2: Agents maintain representations of the world and use them for reasoning.
Big Idea #3: Computers can learn from data.
Big Idea #5: AI can impact society in both positive and negative ways.
Screen 1: The player chooses the video they think the user will like most.
Screen 2: The player is given points based on the given metrics.
Screen 3: The player watches as the AI "plays" the game and explains its decisions.
Screen 4: The model on which the AI and player base their decisions. The neural network is fully connected, so all 5 attributes affect all 3 metrics.
Player will click or tap on buttons in a webpage to play the game
Later, the player will watch as a neural network "plays" and its actions are explained
Pre-activity competency survey
Post-activity competency survey
Post-activity interview
Electronic data:
Child's total game score
Time spent on each user
feedback loops
optimizing for both positive and negative engagement