[free_rms]

There are all sorts of things that people broadly agree are offensive that can be or have been produced by ML models. Why not stick to those and discuss mitigation strategies? The paper could be vastly improved and still make it's point in a less divisive way by sticking to things everyone agrees are beyond the pale.

And that's not even getting into the whole "global warming is racism" digression. Maybe it is, maybe it isn't, but it's definitely not a topic for an AI paper. Just say carbon is bad!

It seems like the goal of the paper is to make a bunch of sociological statements and impute the authority of "AI Ethics Lead Researcher" to them -- they're just political opinions, everyone has those.

[charlescearl]

Yes, we can use data science to discern the values (i.e. politics) inherent in AI research [1]. Also, "beyond the pale" [2] stems from the area enforced during the tsarist era of Russia (1791 - 1917) beyond which Jews were not allowed settlement. Tsarist apartheid in other words.

[1] Values of Machine Learning, William Agnew, Abeba Birhane, Dallas Card, Ravit Dotan, Pratyusha Kalluri at Resistance AI Workshop 2020, https://drive.google.com/file/d/1tjrm3Bf1hxV8iuPSiCcM1IazITG...,

[2] https://en.wikipedia.org/wiki/Pale_of_Settlement

[free_rms]

This is a cool example of how difficult life can be if every figure of speech is interpreted to try and find offense, somehow, whether historically accurate or not.

I agree with sister comment that the English history probably informs the etymology of an English phrase, especially given that the phrase was in use by the mid-17th century, before your timeframe.

https://en.wiktionary.org/wiki/beyond_the_pale

Anyways, that paper scans more like a sociological research piece about data scientists, based on how they write, than an examination of how 'biased' generated speech is or is not, in case passer-bys are curious. Their conclusion is that most papers write about the neat things you can do with a new technique and don't always include a section on societal impact. It's more an absence of politics (I've got GPT-2 ready with your reply to this sentence ;)).

It's also worth taking into account that given their methods involved a bit of subjectivity, what bias did they bring. That said, it's easy to believe lots of papers were like "here's this thing i did!" without including a section on societal impact.

[gundmc]

> Also, "beyond the pale" [2] stems from the area enforced during the tsarist era of Russia (1791 - 1917) beyond which Jews were not allowed settlement. Tsarist apartheid in other words.

This doesn't seem to be true based on pretty much every source I've seen.

https://en.m.wikipedia.org/wiki/The_Pale