“Removing gender from the annual survey is an egregious erasure of the problems of the gender gap that pervade the tech industry. And worse, it removes important context for the data that is scraped and fed into large language models,” says Catherine Flick, a computing and social responsibility scholar at De Montfort University. “Without that context, the bias of the data set is unknown, and it’s well documented that gender bias is frequently built into technology, from variable names to form fields to assumptions about occupations, roles, and capabilities.”

More women than ever are taking, and gaining, degree-level qualifications in science, technology, engineering, and mathematics, according to the National Science Foundation—though the proportion of women getting undergraduate computer science degrees has dropped by nearly 20 percentage points in the past 40 years. (The share of master’s degrees in computer science going to women has increased slightly.) But even if the pipeline is being fixed, retaining women in the tech sector is tricky. Half of women who enter the industry drop out by age 35, according to data from Accenture.

The problem becomes more pressing because of tech’s ubiquity in our lives, and the way in which artificial intelligence in particular is set to be integrated into everything we do and interact with. The humans behind tech platforms make countless decisions—big and small—about their products and tools that can act to the detriment of people who are not like them.

“With non-AI code, you can debug it, get a second pair of eyes from a different demographic, and check it quite straightforwardly,” says Luccioni. “But if you have AI code, all these decisions that drove the data or the model architecture, they’re baked in.”

Take early versions of ChatGPT: The tool provided responses that suggested its belief system was hard-coded with the idea that good scientists are white men, and everyone else is not. That issue was fixed, and OpenAI CEO Sam Altman asked users to help train the model by flagging such responses in the future—marking them with a thumbs-down button—but the broader issue perpetuates.

“Part of the legacy of those who have developed and implemented AI in the last two decades is to be partially responsible for worrisome backward steps in gender equality,” says Carissa Véliz, associate professor at the Institute for Ethics in AI at the University of Oxford.

Véliz worries that the gender imbalances in designing and coding major platforms—from social media to the new generative AI tools we’re using now—are negatively affecting how women are treated by those platforms. “From the way social media hurts women to hiring algorithms offering more opportunities to men and discriminating against women, tech bros have brought back a toxic culture that is not only bad for women, but for society at large,” she says.

Flick worries that without clear data about who is coding the tools we’re likely to use every day, the bias that will probably be encoded into them is “doomed to be replicated within the results that the LLM [large language model] produces, further entrenching it.”

It’s imperative that that changes—fast, particularly when looking at AI. “Until that happens,” Veliz says, “there is little hope that we will have ethical AI.”