Creating Bias in AI
Students will train Google’s Teachable Machine using data sets and then revise the learning data to reduce bias and increase accuracy. Students will recognize that new AI inventions can help people but can also have unintended effects.
Overview
Students learn how machine learning systems can reproduce bias and how training data can be revised to reduce it. The lesson opens with a peanut butter and jelly activity in which students write algorithms for the best sandwich and see how personal preference shapes them. Students explore Google's Teachable Machine to learn about supervised classification and the difference between training and test data. In pairs, they build a cat-dog classifier from a dataset that favors cats, observe the uneven results, watch a Gender Shades facial detection video, and retrain with a revised dataset. They summarize bias in AI in a visual story map.
Lesson Objectives
- Train a type of AI, machine learning algorithm, using data sets then revise the learning data to reduce bias and increase accuracy. Students will recognize that new AI inventions can help people but can also have unintended effects.
Standards
NGSS · 2
- MS-ETS1-1Define the criteria and constraints of a design problem with sufficient precision to ensure a successful solution, taking into account relevant scientific principles and potential impacts on people and the natural environment that may limit possible solutions.
- MS-ETS1-2Evaluate competing design solutions using a systematic process to determine how well they meet the criteria and constraints of the problem