In AI research and building AI models, too few women are involved. That results into software that reinforces stereotypes about women. Just 18 per cent of AI staff at Open AI are women. Though the Chief Technology Officer, Mira Murati, at OpenAI is a woman, there are just 122 are women in the company’s 686 staff whose job it is to build AI systems.
Women and ethnic minorities do not play a major role in building critical technology. Thus, many biases sneak into the system about them. Amazon’s recruitment algorithm had a bias against women candidates since the algorithm was trained on CVs of male candidates. Algorithms scanning chest X-rays underdiagnose female patients since they are trained on data that skews male.
Image generators make women appear more sexualized. Stable Diffusion produces more images of men than of women. Men are shown to occupy high-end jobs, whereas women are shown in low-end jobs. There is a legacy of objectifying women. In academics too, women are underrepresented. They make up 12 per cent of ML researchers.
There should be a level playing field. Many more women should come into STEM industries. One way to fix the problem is to be careful about the data that is used to train the models. Another step is to recruit more female researchers.
Leave a Reply