11 ms·

How Google’s AI ethics meltdown could shape policy

27 points by realpanzer 6y ago

sojournerc 6y ago
Am I wrong or is the very first sentence of this article factually incorrect? I understand that Google accepted her resignation in the letter, and did not fire her outright.
- mlthoughts2018 6y ago
  You are correct. She attached an ultimatum threatening to resign to demands about the identities and circumstances of a paper review, and Google called the bluff. Gebru’s behavior seems wildly inappropriate, and my take on the actual research article is that it was indeed very weak, sanctimonious and legitimately deserving of the rejection for approval to publish externally until it had been edited significantly. Writing a workplace letter in which you tell your coworkers to stop working is just unacceptable in this circumstance. If Google had fired her, it would be 1000% the right move, but they didn’t, they just accepted her bluff resignation, and that is totally on her. Here were some of my thoughts on the actual publication, https://news.ycombinator.com/item?id=25314824 https://news.ycombinator.com/item?id=25314824
- itsdrewmiller 6y ago
  They framed it that way in their communication about it, but in practice they fired her: https://www.bizjournals.com/bizjournals/how-to/human-resources/2016/04/accepting-an-employee-s-notice-to-quit.html https://www.bizjournals.com/bizjournals/how-to/human-resourc... "Most state unemployment agencies consider a separation to be a voluntary quit (initiated by the employee) if the last day of work is within two weeks of the resignation date. So, if an employee gives two weeks’ notice and the employer accepts it immediately, the individual probably won’t be eligible for unemployment. However, if an employee gave two months’ notice and the employer accepted it immediately, the former employee may be eligible for unemployment."
- dragonwriter 6y ago
  > Am I wrong or is the very first sentence of this article factually incorrect? You are wrong in the simplest sense, a little bit more wrong somewhat more deeply, and possibly even more wrong even more deeply. > I understand that Google accepted her resignation in the letter First, you cannot resign with an completely unspecific effective date; a conditional statement that in should particular circumstances you would work with someone to set a final date is not a resignation. So there was no resignation to accept. Second, even though characterizing it initially as accepting her resignation, Google in the termination notice then immediately turned around and identified that she was being terminated immediately for conduct separate from the resignation. Thirdly, the fundamental dispute leading up to the whole “resignation/firing” end game was conduct that, from the descriptions throughout Google AI that have come out in response to the management story, is quite arguably the kind of targeted campaign of hostile treatment that defines constructive termination, which would make even an explicit resignation at the end still a firing for many legal purposes.
  - prepend 6y ago
    > First, you cannot resign with an completely unspecific effective date; a conditional statement that in should particular circumstances you would work with someone to set a final date is not a resignation. So there was no resignation to accept. Would you share some information on where this comes from. It sounds pretty authoritative. I’ve worked with people who said “I quit” and walked out. It seemed to succeed as they never came back. Could they come back the next day and say “well I didn’t specify a date therefore my resignation wasn’t valid?” I’m not familiar with labor law, but I thought that resignation was a loose term that could fit many patterns. This definition from feeling lucky on google [0] just says it’s the formal notice of voluntary termination and there are no laws governing resignations. But it sounds like you have more info on this and I’d like to learn more about this. [0] https://definitions.uslegal.com/r/resignation/ https://definitions.uslegal.com/r/resignation/
    - dragonwriter 6y ago
      > I’ve worked with people who said “I quit” and walked out. It seemed to succeed as they never came back. There's a pretty big difference between “I quit” (a simple immediate present tense statement) and “I will work with you on setting a non-disruptive end date” (a statement of future intent to set an as-yet-unspecified end-date.) Here's some (Texas-specific, but broadly similar considerations apply in other US jurisdictions) relevant information: https://www.twc.texas.gov/news/efte/types_of_work_separations.html#2-weeknotice https://www.twc.texas.gov/news/efte/types_of_work_separation... > I’m not familiar with labor law, but I thought that resignation was a loose term that could fit many patterns. There are many patterns of resignation, and many patterns of firing, but the difference between resignation and firing has important legal effect and is frequently disputed, and thus regulators and courts have developed rules for determining which actually applies in a particular case. ...And that's before even getting into the deeper constructive dismissal issue. https://en.m.wikipedia.org/wiki/Constructive_dismissal https://en.m.wikipedia.org/wiki/Constructive_dismissal
d_theorist 6y ago
Gave up on this article after the first two paragraphs. Typical media fuzzy thinking, incorrect statements and nonsense.
loraa 6y ago
Crazy how we went for 'ultimate freedom' but only the dumbest most self righteous people will take advantage.
prepend 6y ago
This topic is interesting to me but it’s so hard to learn about. The reporting seems to have so much lack of detail it’s weird. It’s all summaries of statements without sources. I understand that Gebru may feel dehumanized but without seeing the email that she references, it’s hard for me to understand her statement. If the CEO of Google painted her as “an angry black woman” then that’s a really big deal. If she feels that way, then that’s not as unique to me as people frequently feel upset and about ways that are unique to them, but not interesting to me. Similarly with the fire/resign thing. These stories seem to all say “fire” even though that seems contested by Google for what happened. Without reading her actual ultimatum email then it’s hard for me to tell. So it’s frustrating that what I want to know, these stories don’t cover. They are statements from Gebru that seem vague, easily substantiated, but not substantiated. So the articles seem to just focus on someone’s feelings.
- Hnrobert42 6y ago
  All the details were hashed out in posts here a few weeks ago. I won’t summarize because one side or the other will downvote me, but you can search HN.
  - oli5679 6y ago
    Can you link?
    - fruffy 6y ago
      This is the most comprehensive discussion on this topic, which also deals with the fallout on Twitter. Includes (anonymous) comments by Google employees: https://old.reddit.com/r/MachineLearning/comments/k77sxz/d_timnit_gebru_and_google_megathread/?sort=top https://old.reddit.com/r/MachineLearning/comments/k77sxz/d_t... Otherwise this search will cover most of the HN debate: https://hn.algolia.com/?q=gebru https://hn.algolia.com/?q=gebru
  - prepend 6y ago
    I read all of those and never got the details. I could never find the actual email she sent. Just the email she sent to the listserv. I did end up finding Pichai’s email and while it was bland corporate speak, I don’t understand how it was particularly damning. I’m not sure how anyone could expect Google to apologize when they don’t think they are wrong, or especially through a company wide email. If anything it’s a positive that the CEO is taking such interest and spending such energy on this.
mellosouls 6y ago
Seems quite a deep review - but clearly takes a side from the get-go with little indication there are different perspectives on this issue (and no clear view as to exactly what happened yet to outsiders). Articles like this would hold more power if the bias was removed or at least acknowledged with pointers to alternative takes.
mfDjB 6y ago
I find the term "ethics" has had its meaning warped lately, ironically for the purposes of "ethics-washing" which is what this article is claiming is happening when big corporations engage in AI ethics. When I worked at a FAANG company in the machine learning field, there was a group of people interested in ethics, but they never seemed to discuss different views on ethics. They often seemed to use the word as a tool to hide behind their collective action to force the company to change course. For example this article says the following: "Members of Google’s Ethical AI team sent additional demands to Pichai, calling for policy changes, among other things.". If you look at these demands it includes the following: * The dismissal of the company vice president. * The reinstatement of Gebru at a higher level than she previously had. The question as to whether these demands are ethical is not clear at all to me. Another case in point, anecdotally I saw this a lot at the place I used to work, was the call for regulation to control AI research, and indeed this article talks about this in the section: "Scrap self-regulation". Taken from a different political perspective, someone might argue that giving more control to the state for the purpose of pursuing legitimised violence against AI researchers is unethical (note this is not my view it is an example). I saw people air views like this in the ethics discussions and given very unfavourable treatment, bordering on unethical. It would be great if AI ethics was a field that produced solutions to problems such as verified datasets, best codes of practice and testing tools. Instead it just seems to be a way to highlight issues and advocate for political positions, which isn't in itself worthless but it feels like the first step in what should be a journey. Unfortunately highlighting problems is really easy, fixing them is the challenging part. I often got the feeling that the people interested in this just cared about directing the labour of other people for their own interests, e.g. leading from behind rather than leading from the front. All seemed rather unethical to me.
- mlthoughts2018 6y ago
  It reminds me a lot of Robin Hanson’s post “Inequality talk is about grabbing.” https://www.overcomingbias.com/2013/08/inequality-is-about-grabbing.html https://www.overcomingbias.com/2013/08/inequality-is-about-g...
- chokeartist 6y ago
  30 years ago, "ethics" generally translated to don't be a dick and treat people how you would want to be treated. Compared to now, "ethics" is essentially a super-big tent of virtue signaling.
mam2 6y ago
There's no meltdown, everyone will have forgotten the story in 2021
mark_l_watson 6y ago
That is one of the better articles I have read on Google’s ethical missteps (that are probably correct from hard-nosed business optimization). We need government action to apply reasonable guardrails on the actions of the largest tech companies, a clear legal structure that puts society first and still allows growth and profits. I am a member of the Libertarian Party (recent, basically my disgust with the DNC and RNC), and I feel a bit like a hypocrite for taking a non-Libertarian view of the current situation with Google and Facebook.
- jariel 6y ago
  We do not need government intervention to handle relations between companies and their researchers. There's no 'whistle blowing' there, this is not even remotely on the 'needs intervention' radar. 'The Current Situation' with G and FB is complex, just because there might be some anti-trust issues over here, or some 'bad thing over there', does not mean that 'this issue here' needs intervention or even remedy. With respect to 'Bias and AI' - companies are pretty self aware of this, and don't mind adjusting. I suggest the best way is to have independent researchers publishing, and letting those facts + some voices do the work. The researcher in question here actually published some work highlighting the challenges of inequality in data and G responded well to that. So that ironically might have been a better model than what they are doing now.
listless 6y ago
The level of misinformation here is astonishing. Blocked and Reported did a deep dive on this recently with Ars Technica co-founder. https://podcasts.apple.com/us/podcast/blocked-and-reported/id1504298199?i=1000502440747 https://podcasts.apple.com/us/podcast/blocked-and-reported/i... And by “misinformation” I mean it’s either in the podcast or in this article. You be the judge.
giardini 6y ago
Google should close the "ethics" department and open another one dealing with etiquette. It seems we've lost our manners and need someone to help us find them again. So sorry to see you go, "ethics", but we're just not ready for you yet. Perhaps someday when we're more evolved!8-))

27 points by realpanzer 6y ago

Am I wrong or is the very first sentence of this article factually incorrect? I understand that Google accepted her resignation in the letter, and did not fire her outright.

mlthoughts2018 6y ago
You are correct. She attached an ultimatum threatening to resign to demands about the identities and circumstances of a paper review, and Google called the bluff. Gebru’s behavior seems wildly inappropriate, and my take on the actual research article is that it was indeed very weak, sanctimonious and legitimately deserving of the rejection for approval to publish externally until it had been edited significantly. Writing a workplace letter in which you tell your coworkers to stop working is just unacceptable in this circumstance. If Google had fired her, it would be 1000% the right move, but they didn’t, they just accepted her bluff resignation, and that is totally on her. Here were some of my thoughts on the actual publication, https://news.ycombinator.com/item?id=25314824 https://news.ycombinator.com/item?id=25314824
itsdrewmiller 6y ago
They framed it that way in their communication about it, but in practice they fired her: https://www.bizjournals.com/bizjournals/how-to/human-resources/2016/04/accepting-an-employee-s-notice-to-quit.html https://www.bizjournals.com/bizjournals/how-to/human-resourc... "Most state unemployment agencies consider a separation to be a voluntary quit (initiated by the employee) if the last day of work is within two weeks of the resignation date. So, if an employee gives two weeks’ notice and the employer accepts it immediately, the individual probably won’t be eligible for unemployment. However, if an employee gave two months’ notice and the employer accepted it immediately, the former employee may be eligible for unemployment."
dragonwriter 6y ago
> Am I wrong or is the very first sentence of this article factually incorrect? You are wrong in the simplest sense, a little bit more wrong somewhat more deeply, and possibly even more wrong even more deeply. > I understand that Google accepted her resignation in the letter First, you cannot resign with an completely unspecific effective date; a conditional statement that in should particular circumstances you would work with someone to set a final date is not a resignation. So there was no resignation to accept. Second, even though characterizing it initially as accepting her resignation, Google in the termination notice then immediately turned around and identified that she was being terminated immediately for conduct separate from the resignation. Thirdly, the fundamental dispute leading up to the whole “resignation/firing” end game was conduct that, from the descriptions throughout Google AI that have come out in response to the management story, is quite arguably the kind of targeted campaign of hostile treatment that defines constructive termination, which would make even an explicit resignation at the end still a firing for many legal purposes.
- prepend 6y ago
  > First, you cannot resign with an completely unspecific effective date; a conditional statement that in should particular circumstances you would work with someone to set a final date is not a resignation. So there was no resignation to accept. Would you share some information on where this comes from. It sounds pretty authoritative. I’ve worked with people who said “I quit” and walked out. It seemed to succeed as they never came back. Could they come back the next day and say “well I didn’t specify a date therefore my resignation wasn’t valid?” I’m not familiar with labor law, but I thought that resignation was a loose term that could fit many patterns. This definition from feeling lucky on google [0] just says it’s the formal notice of voluntary termination and there are no laws governing resignations. But it sounds like you have more info on this and I’d like to learn more about this. [0] https://definitions.uslegal.com/r/resignation/ https://definitions.uslegal.com/r/resignation/
  - dragonwriter 6y ago
    > I’ve worked with people who said “I quit” and walked out. It seemed to succeed as they never came back. There's a pretty big difference between “I quit” (a simple immediate present tense statement) and “I will work with you on setting a non-disruptive end date” (a statement of future intent to set an as-yet-unspecified end-date.) Here's some (Texas-specific, but broadly similar considerations apply in other US jurisdictions) relevant information: https://www.twc.texas.gov/news/efte/types_of_work_separations.html#2-weeknotice https://www.twc.texas.gov/news/efte/types_of_work_separation... > I’m not familiar with labor law, but I thought that resignation was a loose term that could fit many patterns. There are many patterns of resignation, and many patterns of firing, but the difference between resignation and firing has important legal effect and is frequently disputed, and thus regulators and courts have developed rules for determining which actually applies in a particular case. ...And that's before even getting into the deeper constructive dismissal issue. https://en.m.wikipedia.org/wiki/Constructive_dismissal https://en.m.wikipedia.org/wiki/Constructive_dismissal

d_theorist 6y ago

Gave up on this article after the first two paragraphs. Typical media fuzzy thinking, incorrect statements and nonsense.

loraa 6y ago

Crazy how we went for 'ultimate freedom' but only the dumbest most self righteous people will take advantage.

prepend 6y ago

This topic is interesting to me but it’s so hard to learn about. The reporting seems to have so much lack of detail it’s weird. It’s all summaries of statements without sources. I understand that Gebru may feel dehumanized but without seeing the email that she references, it’s hard for me to understand her statement. If the CEO of Google painted her as “an angry black woman” then that’s a really big deal. If she feels that way, then that’s not as unique to me as people frequently feel upset and about ways that are unique to them, but not interesting to me. Similarly with the fire/resign thing. These stories seem to all say “fire” even though that seems contested by Google for what happened. Without reading her actual ultimatum email then it’s hard for me to tell. So it’s frustrating that what I want to know, these stories don’t cover. They are statements from Gebru that seem vague, easily substantiated, but not substantiated. So the articles seem to just focus on someone’s feelings.

Hnrobert42 6y ago
All the details were hashed out in posts here a few weeks ago. I won’t summarize because one side or the other will downvote me, but you can search HN.
- oli5679 6y ago
  Can you link?
  - fruffy 6y ago
    This is the most comprehensive discussion on this topic, which also deals with the fallout on Twitter. Includes (anonymous) comments by Google employees: https://old.reddit.com/r/MachineLearning/comments/k77sxz/d_timnit_gebru_and_google_megathread/?sort=top https://old.reddit.com/r/MachineLearning/comments/k77sxz/d_t... Otherwise this search will cover most of the HN debate: https://hn.algolia.com/?q=gebru https://hn.algolia.com/?q=gebru
- prepend 6y ago
  I read all of those and never got the details. I could never find the actual email she sent. Just the email she sent to the listserv. I did end up finding Pichai’s email and while it was bland corporate speak, I don’t understand how it was particularly damning. I’m not sure how anyone could expect Google to apologize when they don’t think they are wrong, or especially through a company wide email. If anything it’s a positive that the CEO is taking such interest and spending such energy on this.

mellosouls 6y ago

Seems quite a deep review - but clearly takes a side from the get-go with little indication there are different perspectives on this issue (and no clear view as to exactly what happened yet to outsiders). Articles like this would hold more power if the bias was removed or at least acknowledged with pointers to alternative takes.

mfDjB 6y ago

I find the term "ethics" has had its meaning warped lately, ironically for the purposes of "ethics-washing" which is what this article is claiming is happening when big corporations engage in AI ethics. When I worked at a FAANG company in the machine learning field, there was a group of people interested in ethics, but they never seemed to discuss different views on ethics. They often seemed to use the word as a tool to hide behind their collective action to force the company to change course. For example this article says the following: "Members of Google’s Ethical AI team sent additional demands to Pichai, calling for policy changes, among other things.". If you look at these demands it includes the following: * The dismissal of the company vice president. * The reinstatement of Gebru at a higher level than she previously had. The question as to whether these demands are ethical is not clear at all to me. Another case in point, anecdotally I saw this a lot at the place I used to work, was the call for regulation to control AI research, and indeed this article talks about this in the section: "Scrap self-regulation". Taken from a different political perspective, someone might argue that giving more control to the state for the purpose of pursuing legitimised violence against AI researchers is unethical (note this is not my view it is an example). I saw people air views like this in the ethics discussions and given very unfavourable treatment, bordering on unethical. It would be great if AI ethics was a field that produced solutions to problems such as verified datasets, best codes of practice and testing tools. Instead it just seems to be a way to highlight issues and advocate for political positions, which isn't in itself worthless but it feels like the first step in what should be a journey. Unfortunately highlighting problems is really easy, fixing them is the challenging part. I often got the feeling that the people interested in this just cared about directing the labour of other people for their own interests, e.g. leading from behind rather than leading from the front. All seemed rather unethical to me.

mlthoughts2018 6y ago
It reminds me a lot of Robin Hanson’s post “Inequality talk is about grabbing.” https://www.overcomingbias.com/2013/08/inequality-is-about-grabbing.html https://www.overcomingbias.com/2013/08/inequality-is-about-g...
chokeartist 6y ago
30 years ago, "ethics" generally translated to don't be a dick and treat people how you would want to be treated. Compared to now, "ethics" is essentially a super-big tent of virtue signaling.

mam2 6y ago

There's no meltdown, everyone will have forgotten the story in 2021

mark_l_watson 6y ago

That is one of the better articles I have read on Google’s ethical missteps (that are probably correct from hard-nosed business optimization). We need government action to apply reasonable guardrails on the actions of the largest tech companies, a clear legal structure that puts society first and still allows growth and profits. I am a member of the Libertarian Party (recent, basically my disgust with the DNC and RNC), and I feel a bit like a hypocrite for taking a non-Libertarian view of the current situation with Google and Facebook.

jariel 6y ago
We do not need government intervention to handle relations between companies and their researchers. There's no 'whistle blowing' there, this is not even remotely on the 'needs intervention' radar. 'The Current Situation' with G and FB is complex, just because there might be some anti-trust issues over here, or some 'bad thing over there', does not mean that 'this issue here' needs intervention or even remedy. With respect to 'Bias and AI' - companies are pretty self aware of this, and don't mind adjusting. I suggest the best way is to have independent researchers publishing, and letting those facts + some voices do the work. The researcher in question here actually published some work highlighting the challenges of inequality in data and G responded well to that. So that ironically might have been a better model than what they are doing now.

listless 6y ago

The level of misinformation here is astonishing. Blocked and Reported did a deep dive on this recently with Ars Technica co-founder. https://podcasts.apple.com/us/podcast/blocked-and-reported/id1504298199?i=1000502440747 https://podcasts.apple.com/us/podcast/blocked-and-reported/i... And by “misinformation” I mean it’s either in the podcast or in this article. You be the judge.

giardini 6y ago

Google should close the "ethics" department and open another one dealing with etiquette. It seems we've lost our manners and need someone to help us find them again. So sorry to see you go, "ethics", but we're just not ready for you yet. Perhaps someday when we're more evolved!8-))