Page MenuHomePhabricator

GSoC 2026: Improve Commons Android app using privacy-friendly edge AI
Closed, DeclinedPublic

Description

Project title: Improve Commons Android app using privacy-friendly edge AI
Brief summary:

AI could help make the app more user-friendly, as well as improve the quality of the metadata of the uploaded pictures.
Tasks:

  1. Detect meaningless captions such as DSCF1234 etc
  2. Detect joke/unwanted pictures
  3. Detect wrongly-set caption language
  4. Add illustrations to the warning pop-ups
  5. Use the caption that the user filled, to suggest depictions (classification)
  6. Show metadata translations when needed (for instance if a picture only has an Arabic caption and the user can only read Spanish)
  7. Brainstorm and find other areas where AI/LLMs can really improve the quality of pictures/metadata the user uploads to Commons, then implement them. If none, implement phases 1 and 2 of Nearby notable things from OpenStreetMap instead.

Expected outcomes: AI-enabled version available at least to a test group on Play Store. GitHub article detailing pros/cons, technical challenges and early feedback received.
Skills required/preferred: Android programming. If using vibe-coding, the ability to understand what the code is doing and critically evaluate changes is required. Experience with LLM/ML coding is not strictly necessary, but a strong willingness to experiment with these technologies is needed.
Possible mentors: @Nicolas_Raoul (past maintainer), @RitikaPahwa4444 (current maintainer)
Expected size of the project: 350 hours
Rating: Intermediate
Microtasks: Please read gsoc.md which guides you on how to choose an issue and perform a microtask.
Any other additional information for contributors: https://github.com/commons-app/apps-android-commons
Why are you proposing this project? What needs are you aiming to meet? Is it for your Wiki chapter, your community, etc? Improving both usability of the Android Commons app (for its users community) and the quality of its output (for the wikis using the pictures), which we have been maintaining for years.
What is the expected impact? What does success look like? How will this affect the needs you have identified?
20% less metadata quality issues in uploaded Wikimedia Commons files.

Event Timeline

Please make sure the Commons community was consulted about these proposed uses before implementing. Wikimedia communities tend to be extremely skeptical of proposed uses of AI.

LGoto subscribed.

Hi @Nicolas_Raoul thank you for your proposal. Unfortunately we are declining it, as the use of AI is controversial in the Wikimedia communities. We do not have the capacity to sufficiently consult the communities on this project and we cannot risk putting a student participant in front of a contentious discussion. Our priority is ensuring a productive and safe learning environment for the participant. Thank you for your understanding.

Hi @Pppery and @LGoto, I understand the community's concern about AI usage, but I’d like to clarify the specific nature of these tasks to see if it eases some of those worries. I respect your priority to ensure a safe and productive learning environment for the student; we certainly wouldn't want to place a participant in the middle of a contentious community debate.

These are primarily detection tasks and are not meant for generating content, images, etc. They'll help in preventing spam images on Commons or improve the search functionality on the app by giving suggestions. Moreover, given that Wikimedia is extremely cautious when it comes to privacy, we decided to go with on-device LLMs, which means that the data won't leave user's device.

I also wonder if the concern is mainly with the "AI" label itself, as the term has become quite a lightning rod recently. I want to assure you that these tasks were proposed with the community's best interests in mind, especially since Nicolas is a very seasoned member of the Wikimedia community and deeply understands our culture and values. We have had thorough discussions around privacy, licenses, etc. before this idea was proposed.

GSoC has been promoting AI projects since the last two years, and I believe we can support some projects which leverage AI without violating anyone's rights or community's core beliefs.