Why we use AI

We use language models to generate initial drafts of wiki articles. Each draft is then verified, corrected and supplemented with academic and institutional sources.

Each article contains on average 8-15 cited sources, with clickable inline references that link directly to the original source. The "AI-generated" badge is always visible in every article's header, because transparency comes first.

AI doesn't replace verification: it accelerates it. It allows us to cover more topics, in less time, maintaining the same rigor as a fully hand-written article.

Language models can generate incorrect information. This is a documented fact, and we take it seriously. For this reason, no content is published without human verification.

Every relevant claim in our articles is supported by an external source: peer-reviewed studies, guidelines from organizations like the WHO, WPATH or APA, and institutional data. Citations are numbered in the text and clickable, so anyone can verify independently.

The problem is not the tool, but how it's used. An AI-generated article verified with 12 academic sources is more reliable than a hand-written article with no citations.

Training large language models requires significant energy. However, it's important to put these figures in context.

According to the International Energy Agency (IEA), data centers represent about 1-1.5% of global electricity consumption, a share that includes all cloud services, video streaming, social media and AI. A single query to a language model consumes about 10 times the energy of a Google search, but remains in the order of a few watt-hours, comparable to keeping a light bulb on for a few minutes.

Our use of AI is limited to generating text drafts, a low-consumption operation compared to training new models. We don't train models: we use existing ones.

We believe the environmental cost of these operations is amply justified by the value of the result: accessible, verified and free information on topics often neglected or treated superficially.

TransWiki is an open source and non-profit project. We don't exist instead of a newsroom: we exist because no newsroom systematically covers these topics in Italian with the depth we offer.

AI allows us to create a resource that otherwise wouldn't exist. Without it, the time cost of producing over 20 in-depth articles, a glossary, interactive quizzes and an inline citation system would be prohibitive for a volunteer project.

We haven't replaced anyone. We've created something new.

Language models reflect the data they were trained on, which may contain cultural, gender or other biases. This is a real risk, and we address it with concrete measures.

• Every piece of content is reviewed by people before publication.
• Claims are verified with external sources independent of the model.
• The code and content are open source: anyone can report errors, biases or inaccuracies via GitHub.
• We regularly update articles to reflect the latest scientific evidence.

Bias is not a problem exclusive to AI. Human authors also have biases. The difference lies in the transparency of the process and the willingness to self-correct.