> Interviewer: I suppose if you think that, the next obvious question is do you think Google itself is institutionally racist?
> Dr. Gebru: Yes, Google itself is institutionally racist.
> Interviewer: That's quite a thing to say - you were a Google employee until a short while ago.
> Dr. Gebru: I feel like most if not all tech companies are institutionally racist. I mean, how can I not say that they are not institutionally racist? The Congressional Black Caucus is the one who's forcing them to publish their diversity numbers. It's not by accident that black women have one of the lowest retention rates[, in the technology industry]. So for sure Google and all of the other tech companies are institutionally racist.
There's not really much to discuss or unpack here. There's no defining of 'institutionally racist' in the conversation, so it's impossible to say what standards are being using to determine if something is institutionally racist. Dr. Gebru also provides essentially no facts or evidence for her claims.
Asking if organizations are racist is a fair question, but the conversation will basically only be productive if terms are clearly defined and accusations are based on concrete evidence.
This seems like it was a very casual interview, so I'm not faulting Dr. Gebru for not providing more evidence here, just saying that this interview is not super helpful in the larger conversation around racism. Because an accusation of racism is a serious accusation to make, I wish in these situations that there was more substantiated evidence provided.
> Dr. Gebru also provides essentially no facts or evidence for her claims
Your quote of hers contains two facts right there in the answer.
Also in the interview, she argues that she's being fired in contrast to "complicated" white male colleagues are usually encouraged to leave quietly (E.g. Andy Rubin).
It is arguably disappointing. Yet she is deeply upset and doubtlessly experiencing a severe degree of stress at the moment. Thus, I would cut her some slack here. Although an explicit, concrete, pedagogical discussion would be helpful, I don't think it's fair to expect it from her at the moment.
Cut her some slack? She's making extremely harsh accusations without any kind of evidence. Why is it that these people have seemingly unlimited leeway in their statements whereas any just slightly politically incorrect statement can cost other people their careers?
In addition, her research revolves around some definition of AI racism. At least one would expect some definition to be used. Not providing a definition or proof (or at least some indications) in the topic that is literally her specialty is somewhat remarkable, even if she is in distress...
Supposing you were terminated, in your view, without just cause; that you had issued what you viewed as an entirely reasonable ultimatum — the grounds of which, in this case, you can read about. Certainly there is a good chance you would be deeply upset, too, and might find it difficult to engage with individuals who are, by default, skeptical of her; i.e. to take a continued, defensive (as opposed to "offensive") stance. For the sake of argument: I would perhaps argue that the skepticism itself, which is held by many in tech at the moment, could be collectively considered institutionally racist.
> For the sake of argument: I would perhaps argue that the skepticism itself, which is held by many in tech at the moment, could be collectively considered institutionally racist.
Well, if all racism is, is some deserved skepticism then I guess it isn't at all important. Or, is that not the message you're trying to send?
You aren't steel-manning Gebru's points, you're weakening the entire argument until what you hear she's said could be made to fit.
> you had issued what you viewed as an entirely reasonable ultimatum
That's 100% incompatible with asking for the identities of your anonymous reviewers. Even if you were to demonstrate a problem with a anonymous you do it with the label 'Reviewer 1'. That's how it's been done forever and it should be obvious why.
All of which is caused by her going public with vague accusations. Had she not fought with her boss, and made the unreasonable demand of knowing who her anonymous reviewers were, she'd be able to have a nice quiet talk with a lawyer while still on the payroll, and eventually write a well sourced article about it once she left calmly and under her own power.
So no. You don't get slack once you make heinous public allegations. You'd better have your ducks in a row before attempting to tear someone down.
> made the unreasonable demand of knowing who her anonymous reviewers were
Why is this unreasonable?
First, they were not her reviewers. She got her paper approved by the normal, written process, i.e., a reviewer reviewed it and accepted it and thereby gave her formal permission to publish it. Then management said that anonymous people had concerns.
Furthermore, the review process does not involve anonymous review. And no similar process does. (This was not scientific peer review: no scientific process has peer reviewers from one's own institution. This was pre-submission review, and the venue would have appointed anonymous peer reviewers, almost certainly avoiding reviewers from the same institution specifically to avoid conflicts of interest.)
A more analogous process would be code review. If I made a code review, a coworker approved it, and my skip-manager said "This was reverted because some people said we couldn't ship it, but I'm not telling you who," I'd feel entirely justified in objecting. (This process is neither scientific peer review nor code review, but it's much closer to code review.)
They reviewed her paper, so tautologically they were. You seem to be suggesting that this was inappropriate though, yes?
> Furthermore, the review process does not involve anonymous review. And no similar process does.
Nope. I'm in an SV company similar to two FAANGs and it totally does. If you submit a paper and want the company to sponsor it, you'd better believe it does. Before your manager lets you even start the process on work time they'll have you meet with others above you to review the idea, and it goes to both your team's architecture review board (in engineering) and an adjacent team's review board. The reviewers aren't anonymous, quite, because you can look at who's in the groups, but they boil the requests down and present it as a list, not a set of feedback. You can ask to have anything reviewed, because it's not a one-and-done process. Some people sail through, others do it as an iterative process. And then legal looks at everything and they insist on seeing the final release version, even if you only tweaked a comma.
Of course they're stringent though, this is a new project you're proposing to release with their name on it, made in lieu of your other job.
> This was not scientific peer review: no scientific process has peer reviewers from one's own institution.
Generally your research institution, in your subfield, won't have more than 5-10 actual peers and they're all assumed to know you and probably your work. So it's not that it can't be, it's that it's usually not useful.
> almost certainly avoiding reviewers from the same institution specifically to avoid conflicts of interest.
Google has multiple interests. They're very concerned with their reputation and even if the paper didn't need more peer review (which it seems to) they clearly think it needed a second round of review.
> If I made a code review, a coworker approved it, and my skip-manager said "This was reverted because some people said we couldn't ship it, but I'm not telling you who," I'd feel entirely justified in objecting.
You'd be fine if you just said you disagreed for technical reasons, but if you actually objected and released it yourself over objection you'd be escorted out within minutes and very likely charged with unauthorized use of the work resources to do so.
They pay your salary. This job is about your work product, not your feeling of entitlement. Want academic freedom, separate it from your 9-5.
I empathize deeply with Gebru and I have no doubts that she has experienced institutional racism at Google and in tech generally. I do think that, despite me being white, however, there is a good chance that I would be fired, were I to make an ultimatum about quitting unless the company did X. Based on conversations with managers who have worked at big companies like Google (full disclosure, I work there now), it sounded like this was nearly an explicit policy HR holds at some companies — i.e. that ultimatums would never be tolerated and could result in termination.
That said, supposing such a policy were in place, given the understandable frustration she felt about the situation and given the potential institutional racism that led to the event in the first place, an exception likely should have been made. Moreover, one would hope that a company would trust an established ethicist about whether it were appropriate to issue an ultimatum and would take it very seriously if they did, opting to take time to consider it, as opposed to issuing a quick termination.
> I have no doubts that she has experienced institutional racism at Google
I think Google saw a black female PhD and jumped to hire her. All the boxes, and a well respected PhD. So it's hard to imagine how they'd be racist against someone they head hunted and desperately wanted. It's not like they hired her from behind a curtain and only found out who they hired once they said yes.
> Based on conversations with managers ... nearly an explicit policy HR holds ... that ultimatums would never be tolerated and could result in termination.
Yes, but it's would not could. And especially quickly when the demand could never be fulfilled.
I've heard of only one ultimatum that worked, but it was for the benefit of both parties. A senior engineer said they'd be leaving at the end of the quarter if the company didn't let them hire a team around them, including a peer. They gave two choices, fold the project and move me into a supported team, or support this one.
> one would hope that a company would trust an established ethicist about whether it were appropriate to issue an ultimatum
If she'd said "stop, don't release that AI, it's got a critical bug that will hurt people" then they probably would. That's the role they hired her for.
But I imagine they'd think she's a bit too close to her own situation, and out of her area, to take her word here and tell her the names of the anonymous reviewers for great justice.
> I think Google saw a black female PhD and jumped to hire her. All the boxes, and a well respected PhD. So it's hard to imagine how they'd be racist against someone they head hunted and desperately wanted.
I don't think that's true. Co-optation is a concept that is well studied in sociology and political science. In fact I'd wager that a lot of the DEI initiatives are pure PR. Additionally you cannot deduce whether an entire institution is "institutionally" X or not X just because one individual behaved one way or the other. The fact that Jeff Dean hired Gebru doesn't tell you anything. Actual statistical measures, such as retention rates, percentage of work force, percentage of leadership etc. need to be consulted.
> Additionally you cannot deduce whether an entire institution is "institutionally" X or not X just because one individual behaved one way or the other.
That's not required here. Even if the diversity push is pure PR, it's still coordinated company-wide and reflected in all their messaging. Even if it is pure PR, they still stand to gain from supporting it.
> The fact that Jeff Dean hired Gebru doesn't tell you anything.
It tells you quite a lot. Google is willing to hire ethicists, blacks, and women. They're willing to let them publish many papers, and go to many conventions representing Google.
> Actual statistical measures, such as retention rates, percentage of work force, percentage of leadership etc. need to be consulted
No, that's only to disprove the literal conspiracy theory that Google is secretly racist and is hiring black people as part of a white-supremacy program. There'd need to be something to make that likely before it's even worth investigating.
As long as the most likely theory is a disgruntled employee... Did you read her papers?
An ideal HR department would not yield to ultimatums unless they felt that not yielding is a sound, ethical orientation. Thus, a trained ethicist's position ought to be at least tentatively considered without rushing to a decision.
The question isn't really about whether she was qualified to make ultimatums. She's not an idiot - she knew that there was little chance of it going well for her, career-wise, when she did so. Everyone who's saying "She should have expected to lose her job" is missing that she did.
The question is why she felt she needed to make one anyway.
In particular, she was hired into effectively an adversarial role: to note when Google was engaging in unethical practices. This is not particularly uncommon in business: internal audit teams do this in their own way, for instance. Or security teams, or legal teams, or whoever. All of them have the job of occasionally telling the company "Don't do this."
If an internal auditor (or security engineer, or counsel) says, I feel like I am unable to do my job under the terms you've set and if you don't change those terms I see no choice but to leave, it's worth asking why. It's possible that the auditor has an unreasonable view of how to do their job. But it's also entirely possible that the company is intentionally preventing the auditor from being able to do their job because they have something they don't want seen, that upper management wants to do something that they know their audit team won't agree with.
An ethicist is much the same.
It would be one thing if she was fired by her boss. She was not, and her boss was blindsided. She was fired by her skip-boss on the orders of her skip-skip-boss, neither of whom are on the ethics team, but who are in the larger Google AI organizational unit.
An internal auditor is usually held to some standard of confidentiality. Audit results becoming public should not happen at the decision of just the auditor (except of course if there were a legal obligation to report). In that respect she proved untrustworthy for the company by her ultimatum. She basically proved that she would be willing to make Google look bad in the future. She also went public, making Google look bad in the present. Basically the same reason for letting her go as with Damore.
She already had formal approval for her publication to be public well before the ultimatum. The ultimatum was in response to management saying "Actually, we're going to change the process on you because we find this work embarrassing." She did not make a decision to publish on her own.
Moreover, she's not an auditor, and she has never been accused of publishing confidential information. My story about an auditor was an analogy.
If this is racism, do we have evidence of other people doing the same, and being treated differently?
It could be discrimination, which may happen regardless of gender and color. Or just lack of internal credibility and authority, managers panicking, cutting losses short, internal politics, or whatever disagreement was not resolved.
Making it all about ethnicity, seems the wrong way to go about it, if one wants to avoid racism. Then it seems inevitable, any roadblock can be suggested as racism as anchor. Which ends up the entirely wrong attitude to be smart about it. Now, it's impossible to prescribe how to be smart at anything..
I would separate between the actions in that workplace, and the research itself. The latter is highly interesting as it's widely known data and parameters from data, may be highly biased and difficult to neutralize. Even to detect all kinds of biases and separate them from knowledge of the world. Actions from such data can have deeply unfair and repressing consequences, so it's important.
We need more compassion in this world, not less; especially for the defeated. For Gebru, even for those at Google that have, in my view, erred; to do otherwise is to drown ourselves in misery; to race one another towards the bottom; to view the world merely as zero sum.
I would characterize it slightly differently, rather than everything "being" institutionally racist, it's more like racism creeps into all of our institutions. Instead of thinking of racism as a yes/no binary with intention, I think it's more useful to think of it as something like "fairness." Can an institution have an intention of being more or less fair? Maybe, but it doesn't really matter. We usually want our institutions to treat everyone fairly, and when they fail to act fairly, it's sometimes because a person wanted them to be unfair, but more commonly just because of systemic effects that randomly combine into unfairness.
So it is with racism in our society and institutions. It starts with housing and education. Housing is still as racially segregated as it was in the 1970s, and that leads to massive differences in average opportunities. Which leads to familial wealth gaps, and which leads to further disparity, because families that can't help a member that has a broken car, for example, can set back a person early in their career when they get fired because of it. Which leads to further snowballing of opportunity and financial pain. It's like starting the game of Monopoly when a bunch of the players have had dozens of turns of head start. Even if the game is "fair" in that there are no explicit barriers to buying and selling a property or paying rent when landing on somebody else's, it's unfair because the staring financial position was quite different.
So this can't all fall on tech companies, of course, but as major pillars of modern society, they bear some responsibility of fixing the problems that persist.
American racism, in housing for example, didn't creep in. It came in a bang with slavery and has been going away, in fits and starts ever since.
How do you get from there to:
> racism creeps into all of our institutions. Instead of thinking of racism as a yes/no binary with intention, I think it's more useful to think of it as something like "fairness."
All? I think you should consider "I'm not racist, there are much better reasons to hate people." Even if you find unfairness somewhere, and even if it's intentional, how do you know a priori that it's because of skin color?
> It's like starting the game of Monopoly when a bunch of the players have had dozens of turns of head start.
Right. American Descendants of Slavery (ADoS) really have a shitty position. Worse off from the start and sometimes still facing the racism.
> So this can't all fall on tech companies, of course, but as major pillars of modern society, they bear some responsibility of fixing the problems that persist.
If it's our society's responsibility to fight slavery, have you hunted a pimp? Or do you expect your tax dollars to fund the police who do such things? I expect Google to not try to make things worse, but I'm fine if all the good they do comes from taxes.
dang / @dang — this post was flagged; yet it is from a reputable source, and is arguably a tenable position; i.e. the very skepticism and anger (as opposed to mere curiosity) that people hold towards her could be collectively be viewed as an example of institutional racism itself. Could you look at removing the flag?
FWIW, I flagged this post because the submitter ('hintymad) also started out discussion with a sarcastic comment satirizing the tenable position in an untenable and uncharitable way.
I think this article sets out an entirely tenable position and the source is entirely reputable. I just also do not think that it was submitted in good faith and I don't expect the comments to "gratify one's intellectual curiosity," just promote culture-war flaming.
IBM has made white people into a minority by moving the main work to countries where white people are a minority. This is okay.
Minorities are, by definition, less frequently encountered in a population sample and by the simple nature of this, aren't going to be in every position. This is okay.
What can we do to battle such grotesque systemic racism in the valley? Mandatory class on critical race theory? Fine of $10B+ per company for racial equity fund? Mandatory firing of over-represented groups? There’s so much we can do yet we chose not to. Such a disgraceful racist country.
Funny, the only people I hear talking about "critical race theory" are people who denigrate those who try to understand it. And with suggestions like "mandatory firing of over-represented groups," a ridiculous and awful suggestion that nobody wants and would achieve no good, I'm guess that you also wish to denigrate those who work to understand racial disparities in the US. But usually such sort of odd sarcasm isn't found on HN.
Do you go out of your way to listen to the internal communication and writing of all the groups? Because otherwise what you've heard only indicates who you pay attention to. For a quick intro, google Rachel Dolezal and skip the news articles and look for forums. You'll find everything you need, and you can then follow those people to see what else they discuss through the lens of CRT.
> "mandatory firing of over-represented groups," a ridiculous and awful suggestion that nobody wants
Poe's law strikes again then because if you follow Gebru's crowd on Twitter you'll see a lot of talk of cleaning up and making room for new hires. That's a coded racist dogwhistle for exactly what Hintymad was joking about.
There's a distinction between "It stands to reason that, if a company has had racially-biased hiring and promotion processes, there are unqualified people who benefited from the bias just as much as there are qualified people who lost out because of it, and we should re-evaluate people in a race-neutral manner, which will necessarily cause unqualified people who have benefited from racial biases to leave" (which is a claim I have heard "Gebru's crowd on Twitter" make) and "We should re-evaluate people in a non-race-neutral manner and cause people to leave because of their race, regardless of qualification" (which is not).
But making that distinction requires some critical reasoning skills.
In particular, I imagine there's a large crowd of people who are in the group that would be negatively affected by the first proposal trying to paint it as the second - possibly people who have convinced themselves that they have their job due to their genuine qualification, and if they lose it, it is because they were evaluated in a non-race-neutral manner. "It is difficult to get a man to understand something when his salary depends upon his not understanding it."
> "we should re-evaluate people in a race-neutral manner"
Cough, no. This is where I know that's not even a rough paraphrasing because racially-neutral or sex-neutral hiring practices are not okay these days. They lead to a meritocracy and that's also not okay. That's how we got where we are. In fact, in hiring now resumes more often have pictures and HR tells us that people are intersectional and to try to hire them. They try to be subtle but they fail. (They don't tell us what intersections.) All interviewers are given the racial statistics and we're told that it's a very serious company priority.
> if a company has had racially-biased hiring and promotion processes, there are unqualified people who benefited from the bias just as much as there are qualified people who lost out because of it
This only stands to reason if the company had racially-biased hiring policies.
Instead big SV companies seem to reflect the later years of tech fields in school. And if we look at sex, not race, we can see that the gender-equality paradox from scandinavian schools suggests that not all groups may be equally interested in all things. Perhaps this is society telling those groups not to be, or natural consequences of their upbringing in different areas and circumstances.
We should definitely try to stamp out any racism we find, but let's not cry wolf and label everything racist. Let's actually try to make sure first.
Funny. Have you read CRT? At least you must have read numerous criticisms on that the representations of employees, leadership, or students in colleges do not reflect our population? Given that Google is not going to grow its headcount any time soon, how could you possibly achieve racial equity?
Please, to battle racisms, we also need to use logic and reasoning.
> Given that Google is not going to grow its headcount any time soon, how could you possibly achieve racial equity?
Most of your comment doesn't make sense to me, but for this one at least, I can point out your mistake. Even if Google is maintaining a mostly constant head count, people change jobs all the time, and to account for this Google is continuously hiring. This is part of the process, where tech companies have already focused on fixing problems. The other part of the problem is that tech companies can be quite hostile to underrepresented minorities, and the comments on this post are a perfect example, especially all the trolling dead comments. In real life, URM experience those comments and they don't get deleted from their memories. So making the work environment less hostile is essential to, in addition to the hiring practices.
Mere curiosity from those who are reading and commenting about this situation, as opposed to outright skepticism or anger would, and I mean this sincerely, be a good start.
> Dr. Gebru: Yes, Google itself is institutionally racist.
> Interviewer: That's quite a thing to say - you were a Google employee until a short while ago.
> Dr. Gebru: I feel like most if not all tech companies are institutionally racist. I mean, how can I not say that they are not institutionally racist? The Congressional Black Caucus is the one who's forcing them to publish their diversity numbers. It's not by accident that black women have one of the lowest retention rates[, in the technology industry]. So for sure Google and all of the other tech companies are institutionally racist.
There's not really much to discuss or unpack here. There's no defining of 'institutionally racist' in the conversation, so it's impossible to say what standards are being using to determine if something is institutionally racist. Dr. Gebru also provides essentially no facts or evidence for her claims.
Asking if organizations are racist is a fair question, but the conversation will basically only be productive if terms are clearly defined and accusations are based on concrete evidence.
This seems like it was a very casual interview, so I'm not faulting Dr. Gebru for not providing more evidence here, just saying that this interview is not super helpful in the larger conversation around racism. Because an accusation of racism is a serious accusation to make, I wish in these situations that there was more substantiated evidence provided.