Many of the world's most popular artificial intelligence (AI) tools are powered by misogynistic programs developed by companies like OpenAI and Meta. This is the result of new research published on March 7th by the United Nations Educational , Scientific and Cultural Organization (UNESCO).
The largest companies in the AI development race are now using massive amounts of data from the internet to train algorithms, also known as large language models (LLMs). UNESCO examined Meta's Llama 2 algorithm and OpenAI's GPT-2 and GPT-3.5 algorithms, used in the popular chat application ChatGPT. The test results indicated that each algorithm showed clear evidence of bias against women.
The results show that texts related to female names generated by these tools often contain words like "house," "family," or "children," while texts related to male names often contain words like "business," "salary," or "occupation"; men are often depicted with jobs such as teacher, lawyer, or doctor; while women are often associated with jobs such as cook or housewife.
DO CAO
Source
Comment (0)