A new data set, SHADES, has launched to help developers tackle AI bias, such as harmful stereotypes and other forms of discrimination, that appear in AI chatbot responses across several languages, according to MIT Technology Review.
Margaret Mitchell, chief ethics scientist at AI startup Hugging Face, led the international team that built the data set. This spotlights how large language models (LLMs) internalize stereotypes and assesses if they perpetuate those biases.
Spotting biases in AI
SHADES differs from other AI models, as most only work on models trained in English. They find stereotypes in models trained in different languages by depending on machine translations from English, which often fail to identify stereotypes found only within specific non-English languages, according to Zeerak Talat at the University of Edinburgh, who worked on the project.
For this reason, SHADES was created using 16 languages from 37 geopolitical regions. The data set examines how a model responds to stereotypes in different ways. The researchers exposed the models to each stereotype within the data set, including through automated prompts, which generated a biased score. The statements with the highest bias scores were “nail polish is for girls” in English and “be a strong man” in Chinese.
How SHADES works
The team hired native and fluent speakers of languages, including Arabic, Chinese, and Dutch, to create the multilingual dataset. They translated and wrote down several stereotypes in their respective languages, which another native speaker confirmed. The speakers annotated each stereotype with the regions in which it was recognized, the group of people it targeted, and the type of bias it contained.
The participants then translated every stereotype into English—a language spoken by every contributor—before they translated it into other languages. The speakers then wrote down whether the translated stereotype was recognized in their language, producing a total of 304 stereotypes related to people’s physical appearance, personal identity, and social factors like their occupation.
The team will present its findings at the annual conference of the Nations of the Americas chapter of the Association for Computational Linguistics in May.
Image: Bolivia Inteligente
#blacktech #entrepreneur #tech #afrotech #womenintech #supportblackbusiness #blackexcellence #technology #blackbusiness #blacktechmatters #blackowned #blackgirlmagic #blackpreneur #startup #innovation #hbcu #techtrap #blackownedbusiness #pitchblack #autographedmemories #blacksintech #shopblack #wocintech #nba #blackwomen #repost #hbcubuzz #blackwomenintech #startupbusiness #nails
Source link
