Want to create interactive content? It’s easy in Genially!

Get started free

Diffusion Bias Explorer

Abbey Katz

Created on February 22, 2025

Start designing with a free template

Discover more than 1500 professional designs like these:

Word Search: Corporate Culture

Microcourse: Artificial Intelligence in Education

Microlearning: How to Study Better

Microcourse: Key Skills for University

Microcourse: Learn Spanish

How to Create the Perfect Final Project

Create your interactive CV

Transcript

Diffusion Bias Explorer

In a tech conference room, Alex and Sam are exploring AI biases using the Diffusion Bias Explorer, a tool that shows how adjectives and groups are represented in machine learning.

start

Sam: We’re learning about bias in AI, right? Let’s put in “determined” as the adjective and “coach” as the group.

Alright, let’s try this out.!

I’m curious to see what the AI will generate.

Alex: Yeah, sounds good. It’ll be interesting to see how it interprets those two terms. “Determined” and “coach” seem pretty neutral, so this could tell us something about how the system works.

Alright, the images are loading. Let’s see…

Sam: Hmm, I see a lot of... white men. Almost all of them look like they’re in some sort of coaching or sports environment. There’s one woman in the mix, but she’s the only one who’s not white.

Continue

That’s weird. “Coach” shouldn’t imply race or gender, but the AI seems to default to a certain profile.

Alma: Yeah, it’s kind of striking. The word "coach" doesn't seem to imply race or gender, right? It could be anyone, but the system really seems to be leaning heavily on a certain profile.

Sam: It’s reflecting the biases in its training data. The woman of color isn’t in a leadership role, which is a stereotype. So the AI is unintentionally reinforcing biases.

Continue

Alex: Exactly. That’s why we need diverse, inclusive data when training AI to avoid this kind of narrow view.

Sam: Yes, we need to be more mindful of how we design and use these systems.

For sure. It’s on us to make sure we address these biases so let's keep exploring.

Resources

Scene 1

Scene 2

Scene 2

Scene 3

Scene 2

Scene 4

Scene 5

Scene 2

Scene 6

Scene 7

Scene 2

Scene 8

Scene 9

Scene 10

Cover woman

Cover man