Does the Internet Know You're Gay?

slfisher 0 334 Views

Two students from the Massachusetts Institute of Technology have learned that it is possible to predict which men in social networks are gay, even if they aren't out, based on who their friends are.

It's the theory behind traffic analysis, or the process of intercepting and examining messages to deduce information from patterns in communication, even without the content of the messages.

According to an article in the Boston Globe, the project, code-named Gaydar, analyzed the Facebook friend links of 1,544 men who said they were straight, 21 who said they were bisexual, and 33 who said they were gay. "Gay men had proportionally more gay friends than straight men, giving the computer program a way to infer a person’s sexuality based on their friends," the article said.

The students then asked the program to determine the orientation of men who did not declare it on Facebook, and ten men -- whom the students knew to be gay -- were also predicted to be gay by the program. (The program was not so successful at picking out gay women, or bisexual men and women.)

It's not that people need to be worried about being outed by a couple of MIT students. But a number of organizations, including the American Civil Liberties Union, have warned users how much personal information they are revealing in social networking sites without realizing it. What MIT students can do, other people -- including potential bosses -- could also figure out how to do, potentially leading to the notion that users might hesitate at publicly friending others based on what inferences -- correct or not -- might be drawn.

Dani AI

Generated

A compact expert note tying the technical points together for and .

The headline: network structure can leak sensitive attributes, but interpretation requires care. As hints, small or imbalanced samples and the choice of evaluation metric (accuracy vs precision/recall) can make an apparently “good” classifier misleading — especially when the group being detected is rare (the base‑rate problem). Always ask for false‑positive rate, precision, recall, and how the authors validated their model. (en.wikipedia.org)

Why network links matter: people tend to befriend people like themselves (homophily), so friend lists give statistical signals about a person’s traits. Algorithms can exploit structural patterns in an anonymized graph to infer or even re‑identify individuals; these vulnerabilities have been demonstrated in the literature on structural inference and de‑anonymization. Treat a publicly visible social graph as another privacy surface rather than harmless metadata. (annualreviews.org)

Broader evidence and practical consequences: several studies show that easily collected digital traces (likes, group membership, links across networks) can predict sensitive attributes with nontrivial accuracy, so the risk is real and not limited to one project. Practical mitigations are available and worth applying by default. (pmc.ncbi.nlm.nih.gov)

Immediate, practical steps (short checklist):

  • Audit and tighten profile and friend‑list visibility; make friend lists private where possible.
  • Disable or remove unnecessary third‑party apps and review app permissions.
  • Avoid publicly linking accounts or posting membership in small, identifying groups.
  • When privacy matters, consider separate accounts or minimal public profiles and keep sensitive connections off public lists.

These measures reduce leakage from the graph itself; they do not eliminate inference entirely, but they make automated, cross‑site profiling harder.

Member Avatar for happygeek
happygeek 2,411 Most Valuable Poster

Hmmm. Should they not have looked at 1500 straight guys and 1500 gay/bi guys in order to get more accurate results here - rather than base the 'gay men have proportionately more gay friends ' assumption on such a very small sample size of 54 against 1544 straight men? Please note that I am not a statistician (can't even say the word properly!) so this is posed as a question rather than a statement of fact.

Member Avatar for slfisher
slfisher 0 Posting Whiz

To be sure, there are all sorts of methodological issues with this study. I am not a trained statistician, but I have had coursework in it. My primary concern with this story was not necessarily criticizing the methodology of the study itself, but the fact that people were even thinking this way, a sort of guilt by association thing, what it could mean to people themselves based on who their friends were, and what that could mean in terms of people's willingness to publicly friend other people.

Be a part of the DaniWeb community

We're a friendly, industry-focused community of developers, IT pros, digital marketers, and technology enthusiasts meeting, networking, learning, and sharing knowledge.

Edit Preview