Showing posts with label AI. Show all posts
Showing posts with label AI. Show all posts

Monday, November 02, 2026

Unlike the complete MediaWiki software stack, AI is not freely licensed

One of the cornerstones of our processes in the Wikimedia world are its licenses. Both he content and the software is freely licensed. It is why artificial intelligence companies are enabled to use all of the Wikimedia content for its training material.

The functionality of  AI products are not freely licensed, not for its software nor for its content. As relevant is that American models cannot be altered while some of the Chinese models can be altered. It follows that Chinese models are more open.

I have observed that so far American AI is used. However, the USA considers AI strategic and as a consequence AI functionality provided by USAmerican compnies is not globally available. This is in stark contrast with how the Wikimedia foundation has always maintained its ethics.

We should use for instance the Chinese DeepSeek functionality because we can adapt its functionality and we can make it available to all our communities.

Thanks,

        GerardM

Friday, October 30, 2026

Facebook meddles in science for its own interest

I read an article by Professor Danah Boyd about recently released discovery materials indicating that Meta directly targeted Jonathan Haidt by trying to stir up academics to oppose him. Read the article and honestly Meta deserves all the bad press that it royally deserves.

Professor Boyd was particularly upset for the targeting of Professor Candice L. Odgers so much so that she used "wtaf", an acronym I did not know.

So what to do. Given that Scholia is more or less functional, I did the Wikidata thing; I added a few awards, several recent publications and attributed known publications to the people involved. When you consider the sum of all knowledge, it is what AI systems including the one by Meta use for their training material, it introduces a dilemma. Will it include info in its systems critical of its operations and objectives? If not why trust its systems as a neutral source of intelligence?

Thanks,

       GerardM

Saturday, October 10, 2026

AI is a tool. The question is how to use it.

The sentiment among Wikimedians is clear; there is no room for artificial intelligence.

In my opinion that is stupid. You do not blame the hammer; the problem is where the hammer lands. The point of AI is in its analytics. It is not necessary to ask for new information. It can be asked to analyse existing texts, for instance all the Wikipedia articles on the same subject. The objective would be to find inconsistencies and with a corpus as large as the "sum of all knowledge" there will be plenty.

Once we are made aware of inconsistencies, the implicit challenge we always faced becomes explicit. Important is that all the knowledge necessary is in the "sum of all knowledge" itself meaning that the AI can be trained on all content from all Wikimedia projects. This would make it a multilingual and multicultural model,

What one Wikipedia knows and another does not is an inconsistency and at first the least interesting one. More relevant is to know what one Wikipedia knows different from another Wikipedia. Things like is he alive or is she the minister of defence for Whereeverland. The difference is in references. 

When an AI points to inconsistencies and to the latest references, how can any self respecting Wikimedian object to a tool like this?

Thanks,

        GerardM

Friday, July 24, 2026

Wikimania is on and Scholia is off

Wikimania the wonderful gathering of Wikimedians is on in Paris. I have been there before, I have presented at several Wikimanias. It is truly awesome. It is the one place where all aspects of what the Wikimedia Foundation is come together, mingle and have a conversation. There is one inescapable fact; it is mostly about Wikipedia and at that, if it does not fit the English Wikipedia community, conversations will not bring you much.

The two biggest revolutions that happened lead to two new projects and apart from their function they did not change English Wikipedia. It was too expensive to have the same picture on all projects, pictures are now stored at Commons. The English community objected to its policy of only freely licensed pictures so nothing changed as they may still serve licensed materials as an illustration. Articles on the same subject were linked through "interwiki links" the process to maintain them was too costly, a centralised database was created, the links are now stable at Wikidata. The English community was not bothered because it were nerds mostly from other projects who maintained those interwiki links.

The decisions to have both Commons and Wikidata were made by the Wikimedia Foundation.

The 25th birthday of Wikipedia is celebrated in Paris and it has its challenges and opportunities. The biggest challenges in my opinion are communities that are aging and decreasing in size and information that is outdated or incomplete. Opportunities can be found in the "sum of all the knowledge that we already have". One iterative question for an AI: "What are the inconsistencies within our Wikipedias". Another is to have Abstract Wikipedia provide information for all the misses when an article is locally missing.

Why not challenge our public to be part of the solution?

Thanks,

       GerardM

PS Great tools like Scholia should be adopted by the Foundation so that it is always available and we could open up our references in a more informed fashion.

Saturday, October 18, 2025

Using AI for both Wikidata/Wikipedia quality assurance

When people consider the relation between Wikipedia and Wikidata, it is typically seen from the perspective of creating new information either in a Wikipedia or in Wikidata. However what can we do for the quality of both Wikipedia and Wikidata when we consider the existing data in all Wikipedias and compare it to the Wikidata information.

All Wikipedia articles on the same subject are linked to only one Wikidata item. Articles linked from a Wikipedia article are consequently known to Wikidata. When Wikidata knows about a relation between these two articles, dependent on the relation they could feature in info boxes and/or categories in the article. At Wikidata we know about categories and what they should contain. Info boxes are known to Wikipedias for what they contain, relations are likely to be known both to Wikidata and Wikipedia

Issues identified in this way will substantially improve the integrity of the data in all our projects. We are expecting false friends and missing information in Wikidata and in all Wikipedias.

Using AI for identifying issues ensures that quality will be constantly part of the process. That basic facts are correct so that the information we provide to our audience will be as good as we have it.

Thanks,

       GerardM