> Scaling the onboarding of new Wikipedia volunteers with guided mentorship.
Depending on what's meant by "mentorship": this is the bit I'm most keen to see. Much of the criticism lobbed at Wikipedia nowadays seems to come from how much of a pain it is to contribute (which perhaps leads to second-order criticisms about censorship, even if that's not the intention).
It's often been the case that the person most-worthy to speak on a subject is not the most worthy steward of a Wikipedia page on it. Any attempt to make those two people one-and-the-same seems welcome.
If this "mentorship" takes the form of UI hints, I think they'd go a long way. Having volunteers take an AI course, on the other hand, might be useful but might also be a complete waste of time.
I was afraid that mentorship was the point where human-to-human interaction is needed most.
I suspect what you're getting at is that "mentorship" is really code for using AI to step in when people are making the wrong kind of changes to a Wikipedia page. (IE, introducing bias, promoting products, edit wars, ect.)
I wonder too if it could be used to help the edit reviewing process, but I can imagine it runs a risk of becoming an accountability sink[^1] if reviewers can merely defer their judgement to what the bot says is OK to purge. It might have a chilling effect on edits if everyone, including procedure hawks, can rely on it in that way. I'm not enough of a contributor there to know.
Automation is not AI and pre-LLM-era AI is not LLM AI. The arguments for why LLM AI will broadly worsen things are well known. You may not agree with them, but saying things won't change much is pretty empty.
It would be fine as a translation option. For cases when the page is not found, or the local language is lacking so you want to translate from a different language.
It isn't a replacement for an expert writing the real page in the language though. But there are a lot of languages with very little content. German has less than half as many articles as English (Cebuano is number 2 by number of articles - but that seems fishy. German is number 3). At the bottom Cree has only 13 articles. There are also a small number of inactive wikipedias, some of which may want a translation option as better than nothing.
Agreed. The bullet point "Helping editors share local perspectives or context by automating the translation and adaptation of common topics" could be unfortunate if not carefully reviewed - and hopefully it will be.
Having recently spent some time doing translations with the current models, my experience is that their output quality can have a rather high variance. That, along with a tendency toward hallucination, often makes them a "fuzzy guess" at best. When I'm the user asking for that translation, I can factor that fuzziness into my understanding of the output. If its already translated for me, I'm depending on someone else to have done the work of making sure it is faithful to the original.
Not saying they should, but as a PSA, if you’re anglophone and looking something up about a place in the non-Anglo world, it’s always worth switching to that area’s language and auto-translating. Even the images won’t be shared. Structure might be all different. Perspective may not be the same.
I mean like using it for translating another existing article in another language sounds cool, but not that new articles in other languages are not getting written because it's easier to just translate the English version with AI. and what you said it's true, I think that Wikipedia should employ human translation as volunteers for those kind of articles
I actually like that some Wikipedia articles are smaller in my mother language than in English, it gets more straight to the point, so indeed replacing these with auto translations from English would be a loss
I appreciate that they emphasize the importance of their human volunteers. Too many bullish-on-AI folks have misanthropic incentives, and it's refreshing to not see more of that. (It probably helps that Wikimedia is a non-profit.)
Man these AI announcements really highlight CEO groupthink. Though I have to commend Wikimedia here, compared to the Shopify announcement, this one is much more sane and down to earth.
- Supporting Wikipedia’s moderators and patrollers with AI-assisted workflows that automate tedious tasks in support of knowledge integrity;
so, a chatbot on top of some tooling probably
- Giving Wikipedia’s editors time back by improving the discoverability of information on Wikipedia to leave more time for human deliberation, judgment, and consensus building;
extremely vague, but probably the "train a specialized ai on our own corpus to answer questions about it" style bot helper? these make stuff up left and right too
- Helping editors share local perspectives or context by automating the translation and adaptation of common topics;
automated translations would be a big step in the wrong direction
- Scaling the onboarding of new Wikipedia volunteers with guided mentorship.
you can't do "mentorship" with ai in any real way. all in all seems a box checking exercise for their ML officer.
>Scaling the onboarding of new Wikipedia volunteers
could be pretty helpful. I edit a bit and it's confusing in a number of ways, some I still haven't got the hang of. There's very little "thank you for your contribution but it needs a better source - why not this?" and usually just your work reverted without thanks or much explanation.
Could AI sift through removals and score as biased or vandalist?
And then what to do about "original research" that should've been moved to a different platform or better (also with community review) instead of being deleted?
I'm guessing any useful use of AI has already been adopted by some volunteers. Wikipedia might be able to build tools around the parts that work well, but the volunteers will end up footing the bill for the AI spend. Wikipedia will probably pivot to building an AI research product which they can sell to universities/ b2c.
> Wikipedia will probably pivot to building an AI research product which they can sell to universities/ b2c.
Why would they do this? All of wikipedia is publicly available for any use. They literally do not have a competitive advantage (and don't seem interested in it, either).
Exactly. But using AI to summarize articles, stitch them together, etc. under the Wikipedia brand as a product is something they could easily sell. I can totally see a university buying WikiResearch™ for every student.
Readit News
Depending on what's meant by "mentorship": this is the bit I'm most keen to see. Much of the criticism lobbed at Wikipedia nowadays seems to come from how much of a pain it is to contribute (which perhaps leads to second-order criticisms about censorship, even if that's not the intention).
It's often been the case that the person most-worthy to speak on a subject is not the most worthy steward of a Wikipedia page on it. Any attempt to make those two people one-and-the-same seems welcome.
If this "mentorship" takes the form of UI hints, I think they'd go a long way. Having volunteers take an AI course, on the other hand, might be useful but might also be a complete waste of time.
I suspect what you're getting at is that "mentorship" is really code for using AI to step in when people are making the wrong kind of changes to a Wikipedia page. (IE, introducing bias, promoting products, edit wars, ect.)
I'm curious to see how this plays out.
I wonder too if it could be used to help the edit reviewing process, but I can imagine it runs a risk of becoming an accountability sink[^1] if reviewers can merely defer their judgement to what the bot says is OK to purge. It might have a chilling effect on edits if everyone, including procedure hawks, can rely on it in that way. I'm not enough of a contributor there to know.
[^1]: https://news.ycombinator.com/item?id=41891694
It isn't a replacement for an expert writing the real page in the language though. But there are a lot of languages with very little content. German has less than half as many articles as English (Cebuano is number 2 by number of articles - but that seems fishy. German is number 3). At the bottom Cree has only 13 articles. There are also a small number of inactive wikipedias, some of which may want a translation option as better than nothing.
Having recently spent some time doing translations with the current models, my experience is that their output quality can have a rather high variance. That, along with a tendency toward hallucination, often makes them a "fuzzy guess" at best. When I'm the user asking for that translation, I can factor that fuzziness into my understanding of the output. If its already translated for me, I'm depending on someone else to have done the work of making sure it is faithful to the original.
E.g. a city in France’s BRT system:
Contrast: https://en.m.wikipedia.org/wiki/Nantes_Busway
With: https://fr.m.wikipedia.org/wiki/Busway_de_Nantes
Some random examples where the Wikipedia page in the native language is much more detailed:
https://en.wikipedia.org/wiki/Trams_in_Florence vs https://it.wikipedia.org/wiki/Rete_tranviaria_di_Firenze (and an entire second article for the historic system!)
https://en.wikipedia.org/wiki/La_D%C3%A9fense vs https://fr.wikipedia.org/wiki/La_Défense
https://en.wikipedia.org/wiki/Stasi vs https://de.wikipedia.org/wiki/Ministerium_f%C3%BCr_Staatssic...
https://en.wikipedia.org/wiki/Muslim_conquest_of_the_Iberian... vs https://es.wikipedia.org/wiki/Conquista_omeya_de_Hispania ... but https://ar.wikipedia.org/wiki/%D8%A7%D9%84%D9%81%D8%AA%D8%AD... is better than both of them
Does anyone else have good examples? Or are there counter-examples, where enwiki has a better article than the native language wiki?
Dead Comment
Deleted Comment
so, a chatbot on top of some tooling probably
- Giving Wikipedia’s editors time back by improving the discoverability of information on Wikipedia to leave more time for human deliberation, judgment, and consensus building;
extremely vague, but probably the "train a specialized ai on our own corpus to answer questions about it" style bot helper? these make stuff up left and right too
- Helping editors share local perspectives or context by automating the translation and adaptation of common topics;
automated translations would be a big step in the wrong direction
- Scaling the onboarding of new Wikipedia volunteers with guided mentorship.
you can't do "mentorship" with ai in any real way. all in all seems a box checking exercise for their ML officer.
could be pretty helpful. I edit a bit and it's confusing in a number of ways, some I still haven't got the hang of. There's very little "thank you for your contribution but it needs a better source - why not this?" and usually just your work reverted without thanks or much explanation.
And then what to do about "original research" that should've been moved to a different platform or better (also with community review) instead of being deleted?
Wikipedia:No_original_research: https://en.wikipedia.org/wiki/Wikipedia:No_original_research #Using_sources
Why would they do this? All of wikipedia is publicly available for any use. They literally do not have a competitive advantage (and don't seem interested in it, either).