EPIC: Reporting, obsolescence, violence, child mode...

[lfaucon]

Typically a video about COVID would be important between 2020 and 2023 but not later (hopefully)

[glerzing]

2 suggested solutions for now, here is my subjective analysis:

1 - Guess via machine learning if an entity is getting obsolete, (e.g. through decreasing importance score, diminution of the number of comparisons)

Advantages:

• no need to gather data and to modify the user interface

Drawbacks:

• indirect, guessed
• probably not enough data for each entity to make a reliable decision

2 - Add some kind of button to signal an obsolete entity. The entity will not be recommended anymore to the signaling contributor. A threshold could be set such that, if this entity is signaled by a certain number of trusted contributors, it is considered obsolete, and not recommended anymore.

Advantages:

• Explicit
• May be interesting for ethics research

Drawbacks:

• This opens the possibility of trusted contributors signaling entities that they don't approve, even if it is not obsolete.
• If there is an obsolescence reporting threshold, it could be fair to set it higher for someway popular videos, that are thus more susceptible to be reported.

[glerzing]

A discussion occurred in the french questions Discord chat about how to penalize problematic channels.

I proposed to generalize the concept of reporting, with other types of justifications than just obsolete content (e.g. violence, channel to avoid...). Obviously, different types of reports would be handled differently. The consequences could be for example showing a violence pictogram, marking a content as "unsafe", or recommending it less...

To reduce the risks of abuse, we could limit the number of reports that contributors can do per week, and consider only reports from "trusted users" when taking action. There could be further reflection on how to protect whistleblowers or channels that talk responsibly about controversial subjects from undeserved penalties.

This could also be interesting to improve the Tournesol database. It could help the automatic learning of the detection of different kinds of problematic content. It would be possible to aggregate reports anonymously, but we could also chose to leave it as a choice to contributors whether to make it private or public, if the benefit is worth the additional complexity.

[glerzing]

If this is implemented, reporting benign and frequent issues like obsolescence (maybe even allowing to set a duration before obsolescence) should be more accessible in the user interface than reporting more serious issues.

If we are willing to make someday a child mode (maybe even with different ages), it would be interesting to think about how we would implement it and if it can reuse some of the logic of this suggested "reporting" feature. For example, would it be implemented as a black-list of entities, or as a white-list. If implemented as a white list, we could imagine reusing the "reporting" feature to enable reporting positive aspects (e.g. being suitable for children with more than xx years old)...

[glerzing]

I was mentioning the possibility of a child mode. There is a Youtube Kids for children of 3 categories (age<4, and 4<age<8, and 8<age<12). The Youtube API has a status "madeForKids". For children between 12 and 18 years old, it looks mostly based on blacklisting (maybe the API attribute ytAgeRestricted can do the job, more details in https://developers.google.com/youtube/v3/docs/videos).