People sometimes joke about having “ gaydar,” or a special ability to identify someone’s sexuality. However, a new study indicates that while you likely don’t have the ability, Artificial Intelligence can detect homosexuality with a high rate of accuracy just from a person’s face. Researchers at Stanford University used AI to determine sexual orientation for more than 35,000 facial images and compared the results against humans.

Using images from an online dating site based in the United States, the team downloaded 35,326 photos representing 14,776 gay and straight men and women. In a series of studies, they found that facial features of gay men and women appear to differ slightly, and that computers do a better job of picking up on these cues than humans.

“We show that faces contain much more information about sexual orientation than can be perceived and interpreted by the human brain,” the authors write in the report.

Using an algorithm, computer software determined whether a male was straight or gay 81 percent of the time, while it fared slightly lower for women, only accurately predicting 71 percent of the cases. According to the researchers, certain parts of the face offered more insight into one’s sexuality. In men, the nose, eyes, eyebrows, cheeks, hairline and chin were the most informative spots, while the nose, corners of the mouth, hair and neckline offered cues in women.

As background, the prenatal hormone theory suggests that there is a link between sexual orientation and one’s physical facial structure, due to male babies being underexposed to certain hormones and females being overexposed in the womb.

Next, the team compared the computer’s score against humans. Compared to machine, man was only able to detect whether males were gay 61 percent of the time and 54 percent of the time for female photos. It shouldn’t be a surprise that computers are better at detecting sexuality than humans. As the researchers note in their paper, computers using deep nueral networks, DNN, are much better at performing visual tasks. Previous studies have shown that AI could be used to accurately predict skin cancer.

The researchers are quick to warn that perceiving more feminine or masculine features is not a good indicator of one’s sexuality.

“First, the fact that the faces of gay men and lesbians are, on average gender atypical, does not imply that all gay men are more feminine than all heterosexual men, or that there are no gay men with extremely masculine facial features (and vice versa in the case of lesbians),” they write. “The differences in femininity observed in this study were subtle, spread across many facial features, and apparent only when examining averaged images of many faces.”

Instead of attempting to decipher the gender someone is attracted to, they believe this research can be used to highlight the importance of enhancing privacy and protection laws amid rapidly advancing technology.

The researchers definitely don’t advise assuming a stranger’s sexuality based on their looks, especially since according to their findings, there’s a good chance you’ll be wrong.