Are Hume's models applicable across cultures?

Understand how our data collection and model training approaches make our models more applicable across cultures.

Data Collection and Model Training

To capture variations in emotional expression across cultures, we collect experimentally controlled data from around the world.

We have run some of the largest-ever psychological studies on emotional expression to build datasets that reflect the diversity of human experience and the nuanced ways people express themselves. In our studies, participants from a variety of countries react to stimuli, have conversations, make facial or vocal expressions, and provide detailed ratings of what their own expressions mean to them. We also collect ratings from around the world on what emotions people perceive in their expressions. (Our experimental procedures are IRB-approved.)

When we train our models, we transform the ground truth ratings to make them independent of demographics and predictive of ratings in each culture.

Applicability

Based on the diversity and experimental rigor of our datasets, our models are trained to pick up on the underlying visual and auditory cues of human expression independent of age, gender, race, or country of origin. However, careful testing should still be conducted to ensure that any decisions driven by our model outputs are beneficial to users of all demographics and cultures.

Understanding the meanings of expressions and how these meanings can subtly differ between cultures is a complex area of ongoing research and debate. At Hume AI, we aim to provide the most unbiased models of emotional expression available while contributing to scientific knowledge about how people from around the world express their feelings. However, it's important to acknowledge that this knowledge is still developing. Read more here.


What’s Next