Wikimedia Diff

https://diff.wikimedia.org/2022/04/20/how-smart-is-the-smart-copyright-act/

> you can let Congress know that you oppose mandatory censorship filters. Fight for the Future is leading a petition opposing the harmful impacts of the legislation that will be delivered to Congress on 25th April, 2022. You can sign the petition at https://nocensorshipfilter.com and make sure Congress knows just how many people are concerned about the impacts this bill will have on free speech.

The "SMART Copyright Act" would:

- Create universal automated censorship chokepoints across the internet by requiring expensive content filters that we already know don’t work and regularly silence lawful speech

- Encourage surveillance backdoors in everything from Canva to Google Drive to internet service provider traffic, since it is written to apply at all layers of the internet

- Mandate filtering of all content online at all levels, from Big Tech platforms like Twitch and Instagram to ISPs like Verizon and Comcast—and for any other platform that lets people speak or share online

- Force these companies to compromise the safety and security of the internet with mandated use of software even if it has massive flaws, or risk being sued

https://onezero.medium.com/nonstandard-measures-cf47c67e8f05

> the labels, publishers and other rightsholder groups backing “stay down” filters will all tell you that Content ID routinely allows infringing materials through. That’s because it is such a blunt…

Thanks for the detailed answers! Again, I share at least some of your underlying concern, and don't want that to be overshadowed. That said, some responses:

This is the perfect opportunity to do so, and they aren't doing it. Instead, they are picking the smallest possible battle they can. It looks like they've written three articles "strongly" opposing the "tremendous threat" posed by this bill: two when it was being considered[1,2] and another after it passed[3]. Yes, these articles are focused on the impact of the bill on Wikimedia's projects, but I think that's clearly a rhetorical strategy to garner some credibility from the notoriously-stuffy UK legislature. "Foreign nonprofit thinks your bill is bad in general" isn't exactly a position of authority to speak from (if you're thinking like a politician).

More recently, they've proposed the "Wikipedia test" to the public and to lawmakers (such as at the 2024 UN General Assembly[6]) that pretty clearly implicates this bill. The test reads as such: Before passing regulations, legislators should ask themselves whether their proposed laws would make it easier or harder for people to read, contribute to, and/or trust a project like Wikipedia.

That's ridiculous. ID systems endanger everyone, particularly the adults who participate. I was more making a point about why social media companies aren't involved than justifying that choice for them on a moral level. I suspect you have stronger beliefs than I about the…

Interesting article! Shopify's approach is cool, it's interesting they're using Kafka to generate datasets. I wonder if the explicit human rankings will get stale (and also be hugely outweighted by implicit judgements in the training data). The real-time feedback aspect sounds cool, I wonder if it's just for metrics or also for re-training in real-time.

I worked on a Learning To Rank implementation a year or so ago. What struck me then (and now reading about Shopify's implementation) is that the approach is often very similar across sites, but the implementation is usually rather tailored. You see the same patterns: online/offline metrics; nDCG; click models and implicit/explicit relevance judgements; re-ranking top-k of results, and so on.

Unfortunately there doesn't seem to be a technology tying all of the components of an LtR system together. A managed service like Algolia could be an answer. I wonder if industry will eventually converge on a framework, such as an extension to Open Source Connection's Elasticsearch Learning to Rank plugin (https://diff.wikimedia.org/2017/10/17/elasticsearch-learning-to-rank-plugin/).

It's a really interesting area of theory and practice - I hope Shopify write more about their implementation!

I'd also recommend reading Airbnb's really excellent paper - https://arxiv.org/pdf/1810.09591.pdf.

>REAL-WORLD APPLICATION

>Translating Wikipedia for everyone

Hmmm.

While there is very definitely utility in doing things like this, I do kinda fear "poisoning the well"-like effects of feeding (even partially-) AI-generated-data into extremely common AI-data-sources.

There's some info on it in a blog post[1] and the MediaWiki "Content translation" page[2], but does anyone know of any studies on the quality of the translations produced? I can absolutely see it being a huge time-saver for people who are essentially fluent in both (there's a lot of semi-mechanical drudgery in translating stuff like this that could be mostly eliminated)... but people are pretty darn good at choosing the easy option of trusting whatever they're given rather than being as careful as they should be. It kinda feels like it runs the risk of passively encouraging people to trust the machine's choice over their own, as long as it isn't obviously nonsense, and the cumulative effect could be rather large after a while.

[1]: https://diff.wikimedia.org/2021/11/16/content-translation-tool-helps-create-one-million-wikipedia-articles/

[2]: https://www.mediawiki.org/wiki/Content_translation

Worth emphasizing that this is described as basically just adding new ways to consume Wikipedia at scale, not taking anything away (not that that's impossible in the future, but unlikely IMO). In theory, systems that use a lot of Wikipedia data (search engines, "AI" assistant/question answering systems, language models) can keep doing what they're doing with Wikipedia dumps, but new APIs could make these services better (e.g., faster reaction time to current events).

A big upside of testing this out is the opportunity to keep better track of what services are Wikipedia-dependent. While search results often send people to Wikipedia (see e.g. https://diff.wikimedia.org/2021/09/23/searching-for-wikipedia-duckduckgo-and-the-wikimedia-foundation-share-new-research-on-how-people-use-search-engines-to-get-to-wikipedia/), voice assistants or Q&A systems might be (i.e. likely are) surfacing Wikipedia content in more subtle fashion. Same with large language models. Under status quo, it will be increasingly hard to track these relationships, so exploring alternatives is well worth it.

I find amusement in how the three big "knowledge for all" organizations are perfectly distributed along the D&D alignment 'law' axis (taken literally):

- Wikimedia (“a world in which every single human being can freely share in the sum of all knowledge”): lawful – religious copyright law adherence, thorough fair use rationales; proactively removes Het Achterhuis from Wikisource over DMCA "actual knowledge" of it being copyrighted in the US through an obscure series of laws[1]

- Internet Archive (“Universal Access to All Knowledge”): neutral – gets archival exemption for bypassing copy protection added to the DMCA;[2][3] making archived items publicly available lies in a grey area

- Sci-Hub (“free and unrestricted access to all scientific knowledge”): chaotic – just pirate all the paywalled articles

[1] https://diff.wikimedia.org/2016/02/10/anne-frank-diary-removal

[2] https://archive.org/about/dmca.php

[3] https://www.federalregister.gov/documents/2010/07/27/2010-18339/exemption-to-prohibition-on-circumvention-of-copyright-protection-systems-for-access-control