A 2025 UNESCO study on generative AI, systems that produce text, images, and video on demand, found strong biases against women and girls, Black people, and LGBTQ+ communities. As these tools now flood newsrooms, streaming platforms, and classrooms, those blind spots aren’t theoretical. They’re actively shaping what we watch, read, and remember. Most viewers never notice whose stories quietly disappeared from the feed.
The bias finding comes from UNESCO’s research published via Fabricdata [Fabricdata], and it lands at a moment when the defaults are still being written.
Perspective A: AI as a Cultural Eraser
Here’s the core concern UNESCO keeps raising: AI doesn’t just reflect the world.
Photo by It edits it. When generative models, AI systems trained on massive internet datasets, learn from English-language, Western-centric content, the output tilts the same way. Indigenous languages, regional dialects, and local storytelling traditions barely register.
The problem runs deeper inside recommendation engines. As one UNESCO-linked analysis notes, incomplete metadata on newer or regional titles biases recommendations toward older, better-documented content [UNESDOC]. Sparse data produces sparse features, which produce shallow suggestions. The result is familiar to anyone who’s scrolled a streaming service in a smaller market:
-
Global blockbusters dominate the top rows
-
Local indie films get buried under “because you watched…” loops
-
Older Western catalog titles resurface endlessly
It’s not malice. It’s math trained on a lopsided library.
Perspective B: AI as a Cultural Bridge
There’s a counter-argument worth hearing, and it’s not just tech-industry spin.
Photo by Supporters point out that AI translation tools, subtitle generators, and discovery algorithms have genuinely opened doors. Korean dramas went global, Nigerian Afrobeats topped playlists, and Turkish series found audiences in Latin America. If you noticed how 『Squid Game』 or 『Money Heist』 broke through, algorithmic surfacing played a real role.
UNESCO itself isn’t anti-AI. In 2025, the organization released AI competency frameworks for students and teachers to promote safe, ethical, and responsible engagement with the technology [Screendaily]. The message is clear: the tool isn’t the villain. The training pipeline is.
“2025 UNESCO study on generative AI shows that there is a strong bias against women and girls, Black people and LGBTQ+ people.” — UNESCO, via Fabricdata [Fabricdata]
That framing matters. It reframes AI from inevitable threat to fixable system.
Synthesis: Two Truths, One Pipeline
Photo by Both perspectives hold up. AI genuinely amplifies some underrepresented voices while flattening others, often inside the same platform. A Senegalese filmmaker might gain global reach through algorithmic translation, then watch their work get under-recommended because the metadata tagging it is thinner than a Hollywood blockbuster’s [UNESDOC].
Bias isn’t one bug. It’s layered: skewed training data, sparse metadata on non-Western works, and feedback loops that reward what’s already popular. Fixing one layer without the others just shifts the problem.
AI isn’t destined to erase cultural pluralism, but it will by default unless the pipeline changes. UNESCO’s competency frameworks and bias research are pushing toward richer datasets, better metadata, and ethics-literate users. The stories AI learns to tell now will shape which cultures the next generation grows up recognizing. That’s worth paying attention to while the defaults are still being set.
