Disable word2vec and add explanations

[bomanimc]

We are disabling the word2Vec function to reduce the risk that it produces harmful language using model files we included with our examples. See issue #1238 for additional context on this change.

Reviewers: Please help me catch any spelling or grammatical issues in the notices. I've also left some inline questions.

Images of Updates

[shiffman]

Looks great! I added the word "pre-trained" in one spot as a suggestion but this is minor point, it is fine either way! Thank you for your thoughtful approach to this critical work!

[tlsaeger]

We could maybe add the link to the Twitter thread instead of the whole Twitter account. It's a little bit a snake biting its own tail, so we need to wait till we have this thread so we can update again? Does that make sense?

[joeyklee]

Hi @bomanimc ! Whoa - thanks for catching this and for taking action. I didn't realize that these were parts of the pre-trained models!

✅ on my side to disabling and reevaluating how we might continue.

A few non-blocking comments/questions:

1. Would it make sense at all to delete the pre-model files and prompt the user: "in order to use word2vec you must supply your own pre-trained models?" instead of removing the word2vec feature all together? -- this assumes we keep the word2vec feature as is with the assumption that it is the models that are problematic and not the ml5 functionality itself.
2. Another provocation would be to imagine what word2vec looks like w/ some filtering built in where we use something like https://github.com/dariusk/wordfilter to remove words we don't really want to show up in a "bot" like environment. This could be something we integrate on the ml5 side 🤔

Generally speaking, I think this approach of disabling a feature > making space to think and evaluate > reapproaching the topic/feature feels like it reflects the values of the project to forefront the "thinking and processing" part of this work over the "just doing" part.

Thank you! 🙏

[bomanimc]

Thanks for comments!

@tlsaeger the suggestion to link directly to the thread makes sense! Maybe we can come back and update these links after posting, since it'd be good to have something in place before we post.

@joeyklee thanks for these comments. These were ideas we consider as well (the first discussed in the Discord #maintenance channel) and the second @c-dacanay and I discussed with Allison Parrish. Overall, it seems like the team leans to this approach of disabling to reduce the immediate harm that could be posed by cases where people are using the example pre-trained model that includes slurs, but would love to continue the discussion with your helpful thoughts!

[joeyklee]

@bomanimc -- Thanks so much for the update and the explanation. Apologies on my part for not being tuning in! This sounds good! 🙏 👍