Thread by @realscientists

Okay, here are some personal thoughts on how to promote data ethics. Everything here is my own opinion.

First, I don’t think that data ethics should be left to the benevolence and wisdom of tech company executives or the algorithm creators.

https://twitter.com/realscientists/status/1372901544543277060

Most of the US legal framework regulating financial investments—having registered representatives for selling securities, rules around what they can and can’t promise, disclaimers for ads and prospectuses—came about as a result of the 1929 market crash.

I believe that the repeated mass-scale harm caused by Big Tech algorithms is ample proof that we need oversight and penalties—especially for automated decision making systems.

But for that to happen, we need technological literacy in Congress, + public awareness of where & how ML/AI are used.

It’s an embarrassment when Congress summons top tech CEOs, but legislators (+ their aides) can't come up with meaningful Qs—AOC being a memorable exception.

Such laws should address:

- guardrails, disclosure, opt-out options for what data can be collected & how it can be used.
- transparency & accountability.
- perhaps oversight for specific subsets of algos—e.g. those w audience of over 100M, news sources, financial institutions.

But this is much easier said than done. I think GDPR and COPPA are good starts.

In 2017, NYC passed a law to investigate possible regulations for algorithms. By 2019, the effort had largely failed.
vox.com/recode/2019/12…

(pausing while I deal w something else)

And let’s think about the room where it happens, where the decisions are made.

80-90% of the tech work force in the US is white and Asian. It’s also ~80% men.

wired.com/story/five-yea…

I’m *speculating* that Google Photos may not have labeled Black folks as “gorillas"; Twitter wouldn't've centered light faces over dark ones in photos; Zoom wouldn't've pushed dark skins into background… IF they weren't led by white+Asian teams focused on white+Asian user base.

I’m going to *guess* that Google search pushing sexualized search results for “black girls” and “asian girls” would have been caught much earlier if the teams behind the algorithm included more women of color.

So who should be in the room?
- Voices from identities that are most vulnerable to ML misuse (historically & insofar as it can be foreseen).
- Folks who know about the social impact of tech—esp. the internet and algorithmic bias. It shouldn’t be up to ML model creators alone…

the DS talent pool has had a bias toward math+physics. I say hire more social scientists and people with humanities experience.

I mentioned public awareness above. I think that can be used to pressure companies and brands publicly.

Call it “cancel culture” or whatever you want—I believe that there should be reckonings for decisions that eff up at that scale.

I also think that the developers, mid-career folks, and executives also need to reflect on and assert their moral positions.

You can’t be a “politically neutral” data practitioner—I believe you’re either okay with algorithms punching down and dehumanizing people, or you’re not.

And if you’re against those things, in an industry where such things have occurred so easily, we can’t be merely "not the bad guy"—we have to assert our moral positions actively and deliberately.

People in the privileged position to take jobs or awards from those companies can decline them, and make that decision public.

https://twitter.com/luke_stark/status/1372934350011043840?s=20

I’ve declined to speak to recruiters from FB et al because I have ethical concerns about how they run their businesses—and told them as much. I don’t think everyone HAS to do that, but I think those with moral objections should use whatever opportunities to voice them.

None of this is a solution. Just beliefs. I often feel powerless.

Maybe, later in my career, I will be in the room where it happens. Until then, I guess I’ll be tweeting, & exchanging beliefs w mentees & peers, in the hopes that some of them will find their way into that room.

I want to acknowledge and thank Avriel Epps-Darling @kingavriel and Matthew Finney, whose recent course on applying anti-racism to algo fairness, accountability, & ethics helped foment & solidify my thoughts.

I attended w/ a teammate; we shared learnings w/ the rest of our team.