Researchers at the University of California, Berkeley have developed a method to remove gender bias from machines that understand and respond verbally or in text.
A recent study found that researchers have discovered a way to reduce gender-bias in natural language processing while still preserving important information about words’ meanings. This could be an important step towards addressing the problem of biases creeping up into artificial intelligence.
Although a computer is unbiased, much of the information and programming that goes through it is created by humans. It can be problematic when unconscious or conscious human biases are reflected in text samples AI models analyze to \”understand\” the language.
Lei Ding explains that computers are not immediately able understand text. She is a graduate student at the Department of Mathematical and Statistical Sciences. Word embedding is a process that converts words into numbers.