Unable to deal with Chrome Extension Team, Kozmos is shutting down

I have a theory about Google: They're using AI to handle moderation (among other tasks) and it's sort of effective but fails in a lot of frustrating cases, but they cannot admit that the lack of explainability is a serious flaw in their model or that they'd actually need to have humans in the loop to filter out the garbage experiences. Instead, they use humans but treat them almost like bots (probably easier to build more training datasets that way) and put up a hermetically sealed firewall between these moderators and the public.

Google briefly disabled my Google account last year, and it spooked me. I had created a "brand account" to use with YouTube, and for some reason the new account was flagged for suspension within the first day. I was annoyed, but it didn't bother me too much, and I submitted my appeal (aside: I always feel wonderful having no clue what I'm appealing - I'm pretty sure even the Soviet Union or Nazi Germany would have given me the dignity of knowing what I'd been convicted of before the trial + execution/Gulag). Two days later, I woke up to two emails from Google: the first indicating that my main Google account - the one where I have my email, several domain names and YouTube TV - had been suspended; the second indicating that the team reviewed my appeal and had found I hadn't violated any policies and my account had been reinstated, but that I'd need to login soon. I logged in posthaste and found that my brand account was still disabled, but at least I could read and backup my email. (At the time I also happened to be in salary negotiations with my then-client and now employer. Really wonderful timing.)

I tried to submit another appeal on the brand account suspension, explaining that I'd received a reinstatement email referencing the original ticket number, and that all their team had done was erroneously disable and reenable my main account. This time, I got a response saying that I already had an appeal pending so I'd need to wait for it to be resolved. If I really cared that much, I could have tried again to explain again that actually it was their mistake, yada yada, but it wasn't worth it. I didn't want my account to be suspended again with very little recourse other than trying to use my network to get something done internally. Like with a corrupt cop, the prudent strategy was to just walk away and avoid conversation.

I've since stopped using Chrome, and I find that I prefer DuckDuckGo to Google (DDG feels like Google Search used to before $GOOG started blindly trusting deep learning to solve everything). I have hardened my resolve against ever touching an Android device, and I use an adblocker zealously. I haven't fully moved my mail or DNS stuff away, but I am much more cautious. I didn't realize in 2004 when I signed up for Gmail that I'd be entrusting my life's records to an organization that makes the DMV look like an expert at efficiency and customer support.

I'd hope that OP and the various other developers here could maybe band together and maybe consult a lawyer. I have a feeling that given the legal realities of discrimination law and the like, Google's policies and enforcement procedures are bound to be in violation, especially considering the law in its home state. Google's behavior is not merely evil in these cases, but evil and incompetent.

One thing I've noticed is that there are a lot of automated decision making going on. In [1], the ICO offers guidance around what the GDPR says in regards to this subject.

What I don't understand is why the GDPR doesn't enforce the following:

1: Clear and concise information about the action being taken, and clear identification this was done using automation and did not involve any human oversight.

2: A process the user can invoke to request human intervention, and a confirmation email that a human will review the decision that was made within 30 days

3: Public statistics and transparency - any decision that was made that did not involve a human must be published, with stats on % of decisions made, number of cases flagged to human reviewers, and the success/failure rates (for example, number of cases resulting in an overturned AI decision by a human).

This could also be beneficial in other sectors too, like automated credit decisions and insurance policies, to publish statistics and data to afford transparency and identify possible biases. It should also be a requirement in law to preserve any code or algorithms should they need to be audited, including an AI system to be preserved "in time", so that the 2017 version can be audited in 2020 if an investigation is launched for example.

Right now, it's a complete free for all, too many edge cases and ways to game the system if you can figure its loopholes, and no requirement in law to provide a fair basis for users to appeal to a human without causing a PR shitstorm.

This early adoption of AI is quite bad, and I suspect we'll see such developments in the long term future as it matures.

[1]: https://ico.org.uk/for-organisations/guide-to-data-protection/guide-to-the-general-data-protection-regulation-gdpr/automated-decision-making-and-profiling/what-does-the-gdpr-say-about-automated-decision-making-and-profiling/

I feel Google, Facebook and Twitter need to take a big step back and stop antagonizing paying users: be it developers or advertisers. A majority of the issues stem from horribly strict enforcement of policies. What these sites expect from us mere mortals is to do a tightrope walk every time we create and launch something. And some of these policies are actually ridiculous. I understand the fallout of Cambridge Analytica has caused all these sites to give more importance to policy enforcement but the solution to that problem does not lie in automating it through use of technology alone. Business and people's livelihood depend on these policy enforcement. God forbid your business Ad account is shutdown for good because you used a word that causes some stupid algorithm somewhere trigger an automatic ban.

It has happened to me. And worst part is you have no recourse. I have spoken directly with Google Execs who were assigned to a client of mine for managing a big ad account. Even they were clueless as to why the system suspended the account. After a lot of digging around I realized that it was because I had used the same credit card in two different accounts. I mean really? Come on! I am lucky I had a Google Exec who was in contact with me because the client was high value. What would small businesses be going through if the same thing happened to them? I shudder to think.

All these sites need to relax with their enforcement of policies. Deploy humans for this. We already have a lot of unemployment due to this pandemic and this would be a great time to payback some of those profits by employing more support staff. AI is good but it is not good enough yet to have such levels of control.

Maybe they're still angry with him about leftpad.

Yes, it really is the same guy: https://kodfabrik.com/journal/i-ve-just-liberated-my-modules. And, joking aside it would be wise to bear in mind that we're reading only one side of the story here. As with leftpad, there's another side to this.

With leftpad he told Kik, "fuck you" (https://medium.com/@mproberts/a-discussion-about-the-breaking-of-the-internet-3d4d2a83aa4d#.ld8o5zqz7), and then wrought global havoc on npm users. Now he's claiming the Chrome Extension Team "continuously troll developers", and is pulling down something he's created... again.

I only have two data points, so the behaviour here is a coincidence rather than a pattern, but I will guarantee you whatever you think of Google there is more to this than meets the eye.

I'm not without sympathy for the author, but neither am I about to uncritically take his side.

I get it, Google loves to automate stuff to save money. Makes sense, I agree. But I'm seriously wondering if all the people in charge of automating such processes are those delusional ego-programmers who think they can solve anything with machine learning, aka "AI". Really, I cannot understand that there aren't basic safeguards in place like "hey this extension got repeatedly flagged and when a human finally reviewed it we found it was a mistake each time, so maybe set a flag on this extension to double-check next time". Or maybe, have such incidents automatically bubble up to the team responsible for the automatic screening. But why do that if you're a wunderkind programmer who never makes mistakes?

Sorry, this is the only explanation I have for this, I've worked with this kind of person twice. Once they got the first version of something running they are done, no further testing, no sanity checks, no asserts or logger.warn() for "this can never happen" branches.