top of page

ARE AI TEXT-TO-IMAGE MODELS TRAINED TO OBJECTIFY WOMEN?

Policy Watch SA explores this worrying possibility 

the 'most effective' AI text-to-image models 'have generally been trained on massive amounts of image and text data scraped from the web' (Wikipedia)

web scraping, web harvesting and web data extraction are terms referring to the automated process of website data copying using a bot or web crawler  (Wikipedia)

because human annotators control the process of labelling data for analysis, cultural and/or personal biases may influence data interpretation and the algorithmic outcome (Chapman University AI hub

if the data used to train an AI algorithm is not diverse or representative, biased output can occur (Chapman University AI hub)​​

if an AI text-to-image model 'learns from biased language or imagery', 'it may unknowingly generate prejudiced or stereotypical outputs' (Chapman University AI hub

algorithmic bias can be used to reinforce social biases, including those associated with gender and sexuality (Wikipedia)

bottom of page