Blog: re: Meredith Whittaker
Meredith Whittaker has presented herself as an AI expert, ethics expert, ‘research lead,’ and martyr persecuted by Google for bravely speaking out as a whistleblower. She is none of those things, but instead fabricates credentials to fashion herself an academic career, and employs dishonest rhetoric to stoke employee sentiment and curry sympathetic press coverage.
I am a current Google employee. I’m not an executive, an HR person, an ML (machine learning) researcher, or in Meredith’s reporting chain. I know little about Claire Stapleton or the merits of her own retaliation claims. I believe strongly in many of the aims of the Walkout movement, including an end to forced arbitration and transparency about sexual harassment claims and executive payouts. Along with most Googlers, I don’t think any of the organizers or participants in the Walkout should be punished or harmed for their participation in the Walkout.
However, it is clear (and has been clear for over a year) that some of the Walkout participants saw the Walkout as simply the latest ‘outrage of the month’ at Google, constantly complain or agitate against all these outrages, and neglect their actual jobs in favor of their self-anointed ‘Ethics and Transparency Board’ or similar groups. Meredith’s actions over the past year are the most salient and visible example of this group, because of her public statements and affiliation with the AI Now Institute. She led these ‘ethics whistleblowers’ in six major protests against Google in the past year — not just the Walkout, but also protests against Maven, Dragonfly, ATEAC, contractor pay, and now an action today against her own role change. Sympathetic journalists have held her up as an example of a key ‘AI whistleblower’ whose treatment will set a precedent for how Google responds to ethics concerns, adding pressure on the company to tolerate her conduct.
This visibility and influence is problematic because, unlike her colleagues who simply neglect their day jobs and sometimes leak proprietary information, Meredith has invented credentials to proclaim herself an ‘expert’ in the emerging and important field of AI Ethics. She regularly exploits Google’s internal resources to coordinate activism and leaking by like-minded employees, and issues ideology-driven, dataless recommendations to industry and political leaders under the label of ‘empirical research’. The ‘retaliatory’ actions described by Meredith are a result of her harmful actions to Google, not of any involvement in the Walkout or other legitimate actions against sexual harassment/discrimination.
Meredith has embellished and fabricated credentials to pass as an ‘AI Expert’.
Meredith appeared earlier this month (with her AI Now cofounder Kate Crawford) on Kara Swisher’s “Recode Decode” podcast to talk about AI bias and the companies that contribute to it. In a telling exchange with the host, Meredith dances around the issue of her ‘expert’ credentials by insinuating that Google leadership is scared of real ethics accountability:
Kara Swisher: Now, Google has started its own advisory panel, right?…Talk about that. You’re not on the Google advisory panel, and you’re at Google, and you’re an expert in this. Is that correct?
Meredith Whittaker¹: I’m not on the panel.
Kara Swisher: And why is that?
Meredith Whittaker: You’d have to ask the people who put together the panel.
It’s unclear from the context whether Kara and Meredith are referring to ATEAC or some other Google advisory board, but Meredith failed to mention that Google’s research teams (which she is not part of) boast PhD-caliber researchers in ethics, law, and many areas of the social sciences who are simplify more qualified than she is to advise Google on ethical matters. In a statement to Googlers last week she claimed that Google “didn’t see my work at AI Now as worth supporting”, claiming that politics, not credentials, are what keep her out of the decision-making process in Google ML.
Meredith has a BA in Rhetoric from Berkeley. She has no technical degrees or formal technical training. She has no advanced degrees or experience in empirical, objective, or any peer-reviewed research. She claims to have “learned most of what she does by doing it.” She has worked as a Program Manager at Google since 2006, first in the Google Docs org, then on topics related to net neutrality, but never in any capacity related to ML/AI. (A Program Manager role at Google typically consists of keeping product/engineering teams on task and aware of deadlines, preparing external communications, and organizing events like conferences or symposia.) This is still Meredith’s current job title at Google, though in her recent statement to Googlers she derided such a role as ‘administrative work’.
Instead of accurately referring to her Google role as ‘Program Manager’, Meredith invented two titles for herself to lend an air of authority and expertise in AI Ethics. The first, “Founder and Lead of Google Open Research”, is the most concerning because there is no “Open Research” group at Google, other than Meredith. A cursory internet search for this group reveals a sponsored project about open Wifi networks and nothing else. Meredith’s chief role appears to be related to arranging funding — she led no research, did no research, and provided no technical expertise. Other references to the non-existent ‘Open Research Group’ serve as a credential for Meredith or AI Now and have succeeded in getting her labeled by both conservative and liberal outlets as a ‘prominent AI researcher”.
The second, “co-founder of M-Lab”, is almost as problematic. Meredith indeed worked as a Program Manager at the Measurement Lab (aka M-Lab), but held no executive or research role. Calling herself a ‘co-founder’ is a huge distortion. Her claims to have “advised…governments and civil society organizations on artificial intelligence” are equally suspect — her first apparent contribution as an ‘AI Expert’ was in organizing the original 2016 AI Now symposium, where she contributed no technical expertise and mainly just introduced the presenters. Since she started working with Kate, she has gained a layman’s knowledge of the current ethics issues in the field, and has participated in many panel discussions and invited talks on AI ethics where she offers that knowledge as a ‘co-founder’ of AI Now, along with a healthy dose of ideology and snark. However, as shown in the Recode Decode episode, her grasp of technical issues and specifics in the area is shallow and vague, closer to that of the journalist Kara Swisher than that of Dr. Crawford (who has published dozens of peer-reviewed papers related to AI and AI Ethics) or other, more accomplished researchers in the field.
Because of these two invented credentials and her work with Kate, Meredith holds the honorific title of ‘Distinguished Research Scientist’ at the NYU Engineering school. To put this in perspective, another holder of this title is an IBM SVP with 28 years of experience, who helped lead the Watson project as Director of Research there.) It appears that she has remained at Google as a full-time employee for the last several years (until the recent role change that has led to today’s protest), and presumably not paid by NYU or with the grants AI Now has received².
Meredith traffics in activism and ideology, not empirical research. Her work aims to promote broad acceptance of her personal views.
Part of Google’s identity is that of a ‘data-driven’ company — where business decisions are made after consulting the best available information. Ethics is not a data-driven science; it deals with norms and values. Nonetheless, it is certainly possible for ethics research and ethics-based decision-making to be rigorous, and this is crucial in fields that profoundly affect most people — like technology, medicine, or government. This is part of the stated purpose of AI Now, and the reason for its major funding — a grant from the MacArthur Foundation to produce “rigorous empirical research” in bias/inclusion, civil liberties, safety, and future labor impacts of AI. Over the past two years Meredith and Kate have identified research fields of focus and started hiring graduate students and postdocs to perform research in these fields. In the meantime, Meredith has published an annual AI Now Report and other reports, none of which have gone through peer review.
These reports, especially the 2018 report, are anything but ‘rigorous’ or ‘empirical’. They are review papers that begin with strongly-worded ‘recommendations’ which are as informed by author ideology as they are by any of the reviewed scholarly sources. As one very relevant example, the 2018 report claims that
“Technology companies should provide protections for conscientious objectors, employee organizing, and ethical whistleblowers. Organizing and resistance by technology workers has emerged as a force for accountability and ethical decision making. Technology companies need to protect workers’ ability to organize, whistleblow, and make ethical choices about what projects they work on…”
This recommendation aims to justify employee leaks of proprietary business information such as the plans for Maven or Dragonfly, and employees’ choices to spend work time and resources to promote their political views. The relevant section of the report mainly cites two types of sources: previous editions of the AI Now report, and news articles about events at large tech companies. Of the articles, each article concerning Google involved protests that Meredith instigated.
The conclusion of the report asserts that employees should be protected for ‘ethical whistleblowing’ where ‘ethics’ presumably refers to the ethics of the employee (or, more likely, of the author). This conclusion cites a UCLA Law Review article in support of this claim — although the article is limited in focus to whistleblowing of illegal activities such as discrimination (in particular, algorithmic accountability to avoid discrimination) or harassment. This misrepresentation of a scholarly source to justify an a priori recommendation rooted in ideology is, for lack of a better word, unethical. It is far too common of Meredith’s written work, in particular the 2018 report and the protest demand letters she has authored.
While none of Meredith’s AI Ethics work has passed peer review, her review papers have been cited in peer-review work. The citations did not examine the report or its recommendations (here’s one example) but took them at face value. Although the 2016 and 2017 reports had recommendations more grounded in fact than in the author’s own biases and actions, and the 2018 report has not yet (as of May 1 2019) been cited in any peer-reviewed articles, it doubtlessly will, which will lend perceived credibility to the report even if many of the citations do not support or consider its more controversial/biased recommendations.
On Monday she will deliver an address at ICLR (a major machine learning conference) entitled “Why AI Needs Collective Action.” This continued ideological activism under the guise of ‘research’ is likely to continue as long as qualified researchers, Google senior leadership, and the tech press fail to call her out on it.
Meredith claims ‘retaliation’ for the Walkout as cover for her other, harmful actions against Google.
Meredith has been involved in pretty much every large protest or pushback at Google in the last year. Two of these — the Walkout and the contractor protest — were primarily about working conditions, and the Walkout involved coverup of deplorable, illegal activity. Organizing for better working conditions is an important reason for protest — you may think that Google employees have pretty great working conditions, and you’d be right, but sexual harassment is unacceptable in any workplace and our contractors often work in much less favorable conditions.
The other protests Meredith has organized and the leaks she has coordinated have harmed Google’s business interests and its culture. The least controversial of these protests, against Project Dragonfly, was unnecessary at best — the project had faced stiff internal opposition for over three months, was unlikely to ever launch, and had already had the bulk of its assets re-allocated. Meredith’s publicized Medium post exaggerated the scope and status of the project, allowing it to serve an example in her 2018 report of alleged failure on the part of Google leadership to uphold its own AI Principles, and a positive example of employee organization.
The protests against Maven and ATEAC were successful in kneecapping Google initiatives that did not align with her ideological aims. Both these protests enjoyed favorable press coverage claiming massive participation among Google employees. This participation frequently involved signing petitions or rhetorical pieces authored by Meredith and promoted on internal message boards, or attending meetings armed with talking points she authored.
Maven: In March of 2018 a Google executive invited employees to attend discussion sessions about Maven and other AI ethics issues facing the company, including concerns with Youtube recommendation patterns, discrimination in Photos, ad targeting, traffic routing by Maps and Waze through residential streets, and others. Maven was easily the most-discussed topic at that time, and Meredith seized this opportunity to ensure that the meeting would be about pressuring executives on Maven, and nothing else. She recruited like-minded employees to sign up for and attend the (limited-capacity) sessions to amplify her own views and drown out opposing ones:
“You likely received [executive]’s email, which lists the times and locations of the Maven (sic) discussion sessions (pasted below). These will be structured as small-group conversations. You’ll need to sign up below if you want to attend… From [executive]’s email it doesn’t looks like we’ll be getting an update about Maven’s status. However, the small group format should be conducive to talking about the open letter and 7 key concerns [link to talking points] (thanks for you contributions, everyone!) In the spirit of a productive conversation, we encourage those attending to prepare in the following ways:
Identify which of the 7 concerns (or which others) are most pressing for you, and come ready to respectfully expand on your reasoning.
Think constructively about what policies and principles the company could adopt to ensure that your key concerns are adequately addressed.”
This advocacy was productive in derailing the internal meeting from a broad discussion of AI principles and ethics to a Maven roast, where each of her seven talking points were parroted multiple times in the small group sessions and in the larger format. Employees who supported Maven, or who were neutral about Maven and did not agree with the tactics of the ‘Cancel Maven’ group, were met with derisive comments from those organized by Meredith. Several weeks later, after more actions from Googlers including leaks to the press, Google announced it would not renew the Maven contract.
There were many reasons for Google to participate in Maven — not least of which being Google’s hoarding of US tech talent, many of whom would consider defense industry jobs if compensation and benefits were competitive with that of Big Tech. Meredith’s organizing succeeded in amplifying the anti-military sentiment in Silicon Valley (and other countries, although about 85% of Google’s full-time engineering staff is US-based) to dominate the discussion in a company with diverse views about the role of Google in assisting the US military, and to control the headlines in a country whose views are very much at odds with that sentiment.
ATEAC: More recently, Meredith’s rhetoric and activism led to the dissolution of the recently-created Advanced Technology External Advisory Council (ATEAC) for its inclusion of Kay Coles James, claiming her political views (based on three tweets) were not “a valid perspective worthy of inclusion in its decision making”. The views expressed in the cited tweets are, in fact, fairly mainstream for conservative America, which James’s foundation represents. The views condemned by the letter are part of very active current debate and held by a significant number of Americans (and most conservatives). If James’s is not a valid perspective, that means American conservatism is not a valid perspective (unless limited to certain areas).
This is chilling. While it’s true that much of the criticism of tech companies from the American right is overblown, often complaining about ‘censorship’ or negative stories about conservatives (like Alex Jones or Rep. Steve King), it’s also true that conservative viewpoints have been shut down on platforms like Youtube even though they don’t violate any terms. Conservative voices in the room are also important for discussing the nature of self-regulation, government regulation, the military, spending, criminal justice, and so on. These viewpoints deserve a place at the table, and efforts to silence them are flat-out wrong. Joanna Bryson (a professor who works in AI ethics and is far more qualified than Meredith to speak on the matter) said that the dissolution of ATEAC was damaging to discourse across political viewpoints and that “bullying and shunning [of James] are problems, not solutions.”
Besides currying sympathetic press coverage, Meredith again solicited signatures and activism on Google’s high-traffic email lists. Her activism in cancelling Maven and destroying ATEAC not only harmed Google’s business interests, it shut out viewpoints that do not align with her own, and diminished Google’s role in being able to influence policy across the US political spectrum. These negative outcomes were likely the final catalyst for her change in role, because of the timing, but that is as much ‘retaliation’ as is getting fired for chasing a customer out of a store because of their political views.
The question about Meredith’s role change is not whether it constitutes ‘retaliation’. The questions are:
- Why did Google allow Meredith to draw a salary for years with little to no oversight of her work?
- Why did Google condone her inventing a ‘research group’ with no members and no research, to qualify herself as an academic expert?
- Who gave her alleged strong performance reviews for her AI Now work, given that her manager and org are not involved in AI Ethics or academic research, and given that her work is highly biased and not scholarly?
- Why did it take so long for Google to change Meredith’s role back to Program Manager in the first place?
A Program Manager at Google can easily make over $125,000 a year (Meredith has been at Google a long time and likely makes much more than that.) If you or I were to spend our work time and work resources to advocate for our favorite political views, how long would it be until we got fired? Six months? A year? That’s a long time to get paid six figures while doing nothing for your employer.
Yet Meredith labels what she has sown and reaped herself as ‘retaliation’. That’s not only dishonest and unethical per se, but distracts from the very real issues of harassment and retaliation that are (and should be) the focus of the Walkout and today’s actions.
: The transcript provided by Recode misattributes Meredith’s comments to Kate.
: This is accurate to the best of my knowledge. If you have more correct information (about this or other details), please email me at b17k [at] protonmail.com.