Tournesol for Art [platform-as-a-service as a potential business model]

[sergeivolodin]

Currently, Tournesol works quite well for science videos, but mostly lacks the other part of human culture -- art. Art here is understood broadly, from music videos to movie trailers, from Machine Learning Art to weird psychedelic clips, from the biggest LEGO thing to videos of marbles going down in labyrinths, from cat videos to animals in general, from "I will teach you how to play guitar in just 12 videos" ones to "how to build a house with your hands", from modern art to architecture in Ancient Rome and to "surviving in the forest alone"...

There is a whole spectre of videos out there, and right now we are focusing on (arguably, the most important one) educational videos. While it is most important, applying Tournesol algorithms to other categories of videos can be beneficial for their respective communities, and for us as well (see the business model proposal below)

Following a discussion on Discord, it was determined that while scientific videos have some sane criteria that we can agree upon, this is not the case for art: each category would have its own features and sliders, and each category might have its own ways to certify contributors.

For example, movie trailers could be rated on "reflects the actual movie accurately", music videos can be rated on "good atmosphere" and "interesting idea", and marbles can be rated on a "new and exciting" criteria, and LEGO videos can be rated on "height of the tower". For voting on music videos, it does not make any sense to only include institutional e-mail addresses, since music can be enjoyed by anyone. However, for educational videos on playing guitar, it would be nice to ask contributors to prove their actual guitar skills and moderate them manually, as it's hard for a beginner to assess the quality of a tutorial if they have no idea at all how to play guitar.

Instead of trying to design all possible features and add all possible sliders, instead, we can leverage the fact that art is subjective, divided into many forms, and requires specialists to design a proper voting mechanism.

Proposal: Tournesol-as-a-service

What if we allow paying a monthly fee of (say) $30, for which we set up a custom version of Tournesol (such as mostfunnycatvideos.tournesol.app or bestmarbles.tournesol.app or lanadelrey.tournesol.app). The $30 goes in part to the hosting provider to pay for another server, and the rest goes for our salaries.

• Setting up everything can be automated. We use hosting API to spin up a server, PayPal to process transactions, and a script to install everything. Additionally, we add a "setup" page in Tournesol that allows adjusting the quality criteria (for example, remove all of them and add custom ones), select a way to verify contributors (such as 'everyone is equal' or 'verified emails' or 'vouching for contributors' or 'manual moderation', or 'custom' -- see below), select the loss they want to use (for example, disable byzantine resilience, or use a 4-th degree instead of the 2-nd degree regularizer, ...), add some extra authentication methods etc...
• Additionally, we can accept payment to implement custom features. For example, a feminist music festival wants to decide what song was the best -- and they want a special version of Tournesol which a) only accepts people who actually visited the festival, via some ticket number for example and b) want women counted twice as important in the rating. Or, a conference (like AMLD) wants to give a prize for the best talk in the categories "Good presentation" and "Original research", so they ask us to implement university authentication, so that only people from EPFL can vote, they want professors to be counted with 10x weight, and they want to disallow to rate your own talks
• Custom losses are supported as well. For example, someone says "this university doesn't know what they are doing, they are not experts", so they pay us for a custom loss development, and we additionally exclude this university from participation in that custom version on a custom domain like mitisthebesteveryoneelseiswrong.tournesol.app.
• Databases from other custom versions, or from the main one can be re-used. For example, someone wants to clone the (original) science version, and make all domains to be accepted, and host it on communistscience.tournesol.app
• It is always possible to self-host Tournesol since the code is open-source, however, it's likely that subscribing to our service would be easier for organizers since it's only $30/month + whatever they need to develop additionally -- instead of hiring someone, designing everything themselves, etc)
• Some government might want to subscribe to their own Tournesol where only people from that country are allowed, and where only videos in the country's language are allowed
• Someone might want to host an "unofficial" government election, with candidate's videos as content and criteria such as "reliable and not misleading" (yep, this one is super important here as well :), "gets things done", "openness", "pro-lgbt", "good tax reform plan", "collaboration in foreign policy" and maybe even "kindness"... this custom version would only have N~10 videos, and at the end it would provide an overview of the candidates
• Some YouTuber might want to create their own Tournesol platform with all the videos from their channel already imported (and adding videos is not allowed), and they want to know whether their videos make the impact they want to have on people, so they add criteria (for example) such as "not overwhelmed by information" or "enjoyment" or "clarity" or even "audio quality"
• All the "extensions" (custom losses, authentication, contributor verification, ...) can be open-sourced so that when you want to subscribe to our service, you can add some "community extensions" to your version. For example, someone already needed "custom loss with great strategy-proofness" and "integration with facebook" -- so a new customer of Tournesol simply adds these existing extensions to their custom website in the Settings page.
• For the mobile app and the extension, visitors can choose which custom versions of Tournesol they want to load scores from. By-default, it loads the scores from the Science version (the main one), but, for example, a person interested in Machine Learning Art would add mlart.tournesol.app to the settings of their extension, and the scores from there would be displayed as well
• Custom Tournesols can be self-hosted for better privacy, and then added to the global catalog that we maintain, if the customer wants this
• The identity of the customer can be displayed publicly on their custom Tournesol, if they want to

Using this approach, we solve the "problem" of the diversity of videos on YouTube (and of culture in general) via allowing alternative versions of the platform, instead of trying to design one super-great one, which is likely impossible to do in a perfect way..., and at the same time, we get some money as well...

Instead of a single sunflower, we have the whole field of tournesols, each of them is unique in their own way, and they additionally share information to collaborate (such as databases of public ratings for common criteria, and maybe seamless authentication)

Since this issue requires a lot of discussions, feel free to either use issues on GitHub (here below), or write in the following document (please respect others!): https://docs.google.com/document/d/1-bwhM7oRe_0h3oEdYRbw7RwyyBkmJ3Nf629kCxoFdr8/edit?usp=sharing

@lenhoanglnh @lfaucon @srlt @Lbadikho @tchaumeny @TheDuckWhisperer what do you think? :)

[TheDuckWhisperer]

Great ideas @sergeivolodin ! Below are some of my thoughts on the matter.

Assuming Tournesol will at some point be broader than just videos, with the ability to compare any media item on the internet (or off the internet? Books for instance?), we would now also propose to have Tournesol be platform independent, to be used for any goal an individual or party has in mind? It is an interesting proposal and a possible business model but it also makes me think about what makes Tournesol valuable and what truly gives the product its identity.

What is the core of Tournesol, the identity, the thing we are trying to sell? Is it the algorithm? The comparison interface? The database of comparison data? How hard is it for another company to set this up themselves? Where is the value that we're trying to sell here?

Talking in terms of business models I think we're currently mainly trying to sell an ideology. We want to make better recommendation algorithms and help people find higher valued consumer content. This works very well with learning subjects, scientific subjects, subject that have a right or wrong, and media with a goal. If however we are talking about art in the broadest sense of the word, I think the 'value' of the content is too subjective to measure. Someone might love cat videos and watches them every day to get over their anxiety attacks while another person detests them and believes most cats in the videos are subject to animal abuse.

As a business to business proposal I think a platform independent Tournesol definitely has potential, however we should maybe work towards specific use-cases that other companies would commonly want to use Tournesol for, so we can optimise towards those use-cases and have more value for them as a product (and better direction for marketing).

For the reasons mentioned above, I think as a business model Tournesol might have more potential for content of substance than subjective art related content.

Besides the business to business, I also believe Tournesol as a service has a high potential for business to consumers. There are a lot of people who want to have a more accessible internet experience where there is less doubt about wether the media they are consuming is of value. We could for instance have the recommendations on the Tournesol platform be free of charge yet make the integration of Tournesol scores in popular platforms like Youtube be a paid membership feature of the plugins.

I realise I've been talking more about the subject of a business model than about the art aspect of your proposal. I think this is because they can be seen as two separate matters in my opinion.

So coming back to 'Tournesol for art' minus the business model side of things: Like I mentioned art is very subjective, I might consider a more simplistic approach for art instead of making things more complicated with a seperate approach for every category within this subject. Something along the lines of rating media content individually instead of vs another item might solve most of the issues. You can still go in depth with multiple parameters and criteria to rate but it solves the problem of massive diversity and high subjectiveness within the art subject.

[lfaucon]

[I have not read the above comments]

Discussed quickly with Lê last Tuesday: It would be amazing to choose allow someone to choose feature, choose a list of contents and automatically create a Tournesol platform for these features and content

[lenhoanglnh]

I think that we should definitely move to a customizable service in, say, about 6 months. @aidanjungo had the great idea of instantiating a Tournesol vote for the 2022 French Presidential election, where candidates would be rated by "who should be President", but also "who's best for economy / international relations / culture... Etc". I give an important probability (p ≈ 75%) that this can lead to a buzz, and invitations in big medias.

Besides comparing movies, movie trailers, books, papers, podcasts... I think that another important application could be to rate advertisements. It's a long shot but I think that there should be an ad-pricing regulation, with additional taxes for more problematic ads (misinformation, bad habits like tobacco or fast food...). A longshot potentially extremely lucrative role that Tournesol may obtain is to some day become the main tool for scalable ad regulation.

Having said this, especially for now, I think we should rather steer away from Tournesol as a business. First because we legally cannot be a business (we need to register to the Chambre des Commerces). But also because this is not the main goal of Tournesol and could confuse potential donators. Finally, we should reflect carefully about the implications in terms of incentive structures. As we grow, we run the risk of providing poor incentives to our collaborators and to ourselves.