Project Veritas, a non-profit conspiracy theory outlet, today dropped the ‘bombshell’ that it’d obtained hundreds of leaked internal documents from a whistle-blower showing Google’s bias against Conservatives. Unfortunately, as is almost always the case with Project Veritas, the claims are crap.
As TNW’s Ravie Lakshmanan pointed out, the documents are probably authentic. I’ve pored over every single one in the document dump on Project Veritas’ website and it’s my opinion that almost all of them appear completely legitimate – the exception being screenshots of a Google employee’s Twitter and Facebook accounts, which I’m guessing were taken by the ‘whistle-blower,’ as opposed to being official company documents.
Why am I so sure these are real? Because, aside from personal information of dozens of Google employees, none of it matters. There’s nothing in any of these documents that even the most obtuse technology journalist doesn’t already know. In fact, I’d go a step further and say that this data-dump actually makes Google look really good.
I’m intentionally not using the name of the whistle-blower or the names of the conspiracy peddlers at Project Veritas because the source of the information doesn’t matter; I don’t care about the messenger if the message is true. And in this case the message is that Google has diversity training, holds lectures and training on bias in machine learning, and thinks Breitbart and Infowars peddle in hate speech and fake news.
Here’s the TL;DR: It’s almost certain that this is just everything the whistle blower could find by doing a quick search on the employee share drive for “inclusion,” “diversity,” and “bias.” It’s almost all just Power Point slides and papers — the kind you see at your own weekly corporate meetings. You’ll find more shocking information on Google by visiting its official blog. This, in my opinion, is nothing more than Project Veritas disguising an effort to dox Google employees as “a leak.”
Here’s the breakdown of every file Project Veritas has made available (note, in instances of text documents over one page long I’ve selected a screenshot indicative of the documents contents):
There are 15 documents here. They appear to be randomly selected from an email chain discussing deprioritizing Breitbart because it spreads hate speech. Interestingly, the documents clearly show that Google is concerned about fake news. There’s even a chart saying the company trusts the Wall Street Journal the most, CNN and Fox News about the same, and The Young Turks only slightly more than Infowars. Here’s the documents (edited to remove personal information):
There’s also a random slide featuring Trump that appears later in the data dump in a slideshow (apparently presented to employees as part of training about bias):
Ironically, there’s nothing in the censorship folder that pertains to censorship in any way shape or form.
Lol. This folder has three documents showing that Google tried to make an Easter egg that translated the word “covfefe” to an ASCII depiction of a person shrugging. Apparently it was too difficult to get it to work with something in the code and they called it off – or maybe it appeared in testing? Either way, there’s simply no possible way to equate a “covfefe” Easter egg as Google engaging in election interference. Have a gander:
This folder contains five documents exposing the personal information, including their resumes, of Google employees and twenty-one files with information about the company’s efforts to combat fake news. There’s nothing in here that would indicate Google is biased towards any group – merely against disinformation.
The most interesting document here – and perhaps in the whole archive – is titled “news blacklist site for google now.” It’s a list of websites with a high user block rate, or flagged for peddling hoaxes, that appear to be blocked on the mobile Google Now app. Here are the documents with personal information removed:
Here we’re treated to employee handouts and powerpoint slides about Google’s diversity practices. There’s about 20 files here – all of them read like boilerplate diversity jargon: there’s nothing interesting to see here except maybe the “Coffee Bean” analogy, but it in no way indicates that Google is up to anything nefarious. It’s just a way of explaining the demographics of the workforce. We already knew that Google is a company that believes diversity is a good thing. Nothing to see here, but here’s shots of the documents:
Here we have a handful of documents related to training leaders on avoiding unconscious bias. None of this has anything to do with the alleged bias against Conservative views, but on inspection they appear to show some pretty good practices for employee evaluations and communicating diversity and inclusivity training requirements. So, kudos to Google for that I guess, but it’s tough to figure out why they’re in this dump. Here you go:
Machine Learning Fairness
Mother of God. This is it. Dozens upon dozens of documents proving that Google knows about bias in machine learning algorithms and artificial intelligence models. Sound the alarm. Punch the donkey. Run around in circles screaming the end is nigh. Oh wait. Everyone knows machine learning models are biased. What Project Veritas has done here is provide us with a whole folder proving that not only is Google aware of the very real bias inherent in AI, it’s doing a lot to combat it. We’ve got employee lectures, research papers, and slides galore proving it. Thanks Project Veritas! Despite the fact that Google’s in bed with China and helped the US military develop weaponized AI, I’m starting to think maybe it’s not so evil after all.
Note: there are more than 60 files in this folder and they’re all about either bias research, or screenshots related to an email thread about Google’s algorithm showing bias toward African-American inventors when a search for “American inventors” was made. Feel free to visit Project Veritas to see all of them, but for the sake of space we grabbed a representative sampling:
Here’s another folder containing the personal information of Google employees, 9 of the 20 files are just personal information and the others related to a Google protest group (a guide for protesters that includes best practices) and what appears to be an email chain from Google employees who are distraught that Hillary Clinton lost the 2016 election.
While it appears they were very sad, and needed to take some time off to deal with it, they also claimed to take great pride in running a bipartisan effort to get out the Latino vote. That’s it. No mustache-twirling or threats to tank the system. Just some sad emails. Here’s the shots:
Two files. One’s a corrupt PDF and the other’s a diversity lecture. Nothing to see here. Like all the other documents on diversity and bias, these are training and research documents that employees would read at lectures or would put together for meetings. Here:
Aside from the personal information of its employees, it’s a safe bet that none of the information in the archive is even slightly bruising to Google. It’s rife for hot takes from people who don’t take the time to download and pore through all of it, but more importantly: it’s a trove of information for bad actors to pester and harass individuals at Google and little else.
Conclusion: Doxxing tech employees is wrong and Project Veritas, as is almost always the case, is full of shit.