[madacoo]

It's refreshing to see an assumption of incompetence rather than malice, since I seem to have come across the idea that Google is questionably motivated by ideology rather a lot recently.

One thing I don't understand is why Google continues to try to use algorithmic solutions for content moderation. Surely the frequent false positives ends up hurting potential revenue? And would it not be able to hire people to at least oversee the algorithms and respond to erratic behaviour as described in this article?

It seems one could even argue that Google's focus on algorithmic rather than human solutions is a kind of harmful ideology, albeit not the kind that probably hasn't hurt Techdirt in this case.

[natch]

After incompetence and malice, there’s a third option, or maybe a blend of two: arrogance and laziness.

They’ve been sitting on top of the world for a while. The results sometimes speak for themselves.

As a case in point, for over a full decade they have been flagging their own emails, that they send, that they know I signed up for, as spam. For a decade. Possibly as long as two decades.

[zerocrates]

You could interpret that in a positive way, that they're not just papering over problems in their filter by privileging their own messages.

Not super positive but still.

[donmcronald]

You could also interpret it in a negative way. They don't maintain _anything_ over there. How hard would it be to whitelist (can I still say that?) super high value, bulk sending, domains?

[natch]

I suspect that’s what they tell themselves. And that heuristics are beneath them. Better to have an elegant algorithm that does the wrong thing automatically than a set of human-tweaked rules that could come much closer (also automatically) to doing the right thing.

[natch]

Can’t edit. Not two decades, but ever since gmail added spam filtering and started flagging many Google Alert emails as spam.

[bryanrasmussen]

>It seems one could even argue that Google's focus on algorithmic rather than human solutions is a kind of harmful ideology

It's the kind of ideology that you often find in people who think they are smarter than everyone else.

[tmpz22]

Imagine being plucked from college at 22 and getting payed 200k+ total compensation with an introduction telling you how you're now one of the best at your field working with the smartest people in the world... how you're changing the world before you've even been on-boarded.

[asdff]

It's like being told you are heir to the throne, and many engineers carry that same princely arrogance obliviously. At least doctors and lawyers have gone through a second gauntlet after acing school to temper this somewhat. Those fields might not even start making good money until you are years above the entry level, even with a pricey professional degree.

[Nasrudith]

I would argue a second gauntlet only makes matters worse not better for "arrogance".

[asdff]

Maybe in the past where these professional degrees guaranteed highly compensated positions, but that's not the case now that these fields are saturated. A medical school grad still has a painful and overworked road to travel before they start making serious money and overcome their medical school debt. Similarly, if you are in a law school that isn't in the top 10, or you are not in the top 10 of your class, drop out while you can because you are looking at a career of terrible pay and insurmountable debt.

These sorts of gauntlets are much more life or death than a CSE program, where you are guaranteed at least a decent paying job if you make ok grades. On the other hand, pressure of not even knowing how many years before you can pay down your 300k+ professional school debt and start your life hanging over your head humbles you fast, compared to someone who turns in their capstone project and walks into 200k/yr.

[bryanrasmussen]

I think I was a pretty cynical bastard at 22 so I would actually have thought they were full of it.

[sli]

Sometimes there's a very fine line between cynicism and realism.

[WarOnPrivacy]

That line is the size of a Venn Diagram intersection.

[throwaway_1299]

As someone who is in a similar domain / company, no one knows how to do this at planet scale unless you narrow it down to a very short list (which is typically very large sums of money).

My back of napkin math from another comment:

I work in a similar space and it is significantly complex and expensive to do this. Back of the napkin math - * Lets say on average customers contact Google support once a year for each product they use. That's 0.25 tickets per user per quarter. * Consider Google has ~10billion monthly productuser combinations (9 products have 1B+, most have significantly more) That is 2.5M tickets/support requests a quarter. ~28M tickets a day * If we consider an average ticket take ~3 mins to resolve, thats ~155k hours a day * If we take an employee being productive for 7 hours a day, that's 22k employees * If you take a 1:10 ratio, that is 2205,220 and 22 - 1st, 2nd and 3rd line managers. * Take the cost to be an average of 30k,60k,150k and 300k for each of those layers, thats ($661, $132M, $33M, $6.6M) which totals to ~$833M per quarter * The real world costs for this will probably be anywhere between 2X to 3X of this because all of these people come with other costs like infrastructure, tooling, space, etc. So we are looking at ~$1.7B to $2.5B.

One might be tempted to say that money can be saved vs my estimates but keep in mind the challenges of localization, time zones, compliance etc is also significant and will probably mean an even larger expense.

So yeah, it would be ~40% of the quarterly profit.

Sure this is an expense so tax etc can be changed but my argument would be that we are severely underestimating the complexity and challenge at each step.

So yes, I do think it will never make economic sense unless you are on the platform with sufficiently high spend. Just like every single other economic system we have out there.

[grahamburger]

> Lets say on average customers contact Google support once a year for each product they use.

Meaning that if someone uses 4 Google products they are contacting Google support on average once a quarter? That seems really high. I would have guessed at most a tenth of that.

[throwaway_1299]

I was averaging it out assuming there will be an overhead of users using the channel for many other reasons too, eg. they dont know how something works, there are language barriers, other errors get attributed to google (eg. android vs manufacturer) etc.

But this is a fair point, all the assumptions I make (eg. how many products a user uses, number of requests for support, etc.) are hypothetical and might be incorrect if this was actually implemented. Based on my judgement though, while variable might change, the end numbers will probably remain the same.

[donmcronald]

Is 3 mins a ticket realistic? That seems super low for anything that isn't dead simple.

I think the old internet was a lot better because the ratio of moderators to users was much, much higher and the moderators tended to be well informed enthusiasts that were usually running the site and had the ability / authority to fix anything / everything.

The easy solution to me is just add money. Let me pay on-demand for support. I've had a GSuite account for ages and never needed help with anything, but if my account got locked and I needed to talk to a human with the ability / authority to unlock it I wouldn't hesitate to pay $50-$100.

IMO part of the problem at Google is likely years and years of bragging instead of building and the tooling is probably absolute garbage to the point that a human wouldn't make a difference. That's why I say 3 mins seems low. I bet there are a lot of things that a human couldn't fix even if they wanted to.

It's pure speculation, but makes sense to me.

[luckylion]

I believe your guess for frequency is off by a lot, my guess would be more in the vicinity of a tenth of those assumption.

However, more importantly: going in with a "we need to support this" mindset alters incentives. Suddenly, useless error messages and lacking documentation aren't okay any more.

A different approach to solve the issue, if it even exists at that scale: make them pay for it. If you want service, pay $10. That's more than enough cover Google's cost and it's cheap for anyone that really requires service because e.g. their account has been locked, or Google's systems have disabled ads on their site. At the same time, it's expensive enough that you don't get flooded with "why do you show me those videos on the front page?" requests.

[throwaway_1299]

2/3/4 billion users paying $10 a year would certainly change things. I think the closest I have seen is the Google one program but not sure if people are aware or would want to pay for it.

For power users, I completely agree it might be the only feasible idea.

The closest parallel I can think of Amazon where there is cost of products involved (and >> $10 year I suspect) - they do a good job but even they work hard to make sure you cant easily reach a person.

[thewebcount]

I don't think they need a human to handle every single case you mention above. If advertisers had specific actionable steps to take when there's a problem and a way to say, "We've done those things," (and I'm sure Google can automate checking that they actually have), "and it's still not working," they probably wouldn't need to have as many humans working on this as you're saying above.

A lot of the problem is that when Google screws up, there's no recourse. We don't hear much about when Google gets it right because those people's problems were solved. Surely some significant portion of the incidents you mentioned above fall into that category.

[paulryanrogers]

My guess is the vague responses, or no response at all, is because scammers will game the system using any information they disclose.

One would hope transparent, clear rules and enforcement would prevail. But ambiguity and silence are far less expensive.

[madacoo]

Providing tech support by responding to customers is a different beast entirely to content moderation, however. It may be the case that human oversight of algorithmic moderation is also unreasonable, but I find this fairly unlikely. I wouldn't expect it would require quite so many man hours. But again, could be wrong.

[novok]

A possible solution would be to charge for support for people who get paid by you, although it probably creates a whole host of other problems too that would be difficult to balance. Also competition so those same people have someone else to go to.

[CydeWeys]

> That is 2.5M tickets/support requests a quarter. ~28M tickets a day *

What happened to the math here? I'm not following this step?

[nraynaud]

does that means that we have to throw the baby with the bathwater and just forbid any company from being this big, because they could not possibly add value to the world in general past a certain scale?

[ngold]

Yes google should have as many people working for it as the advertising industry that it savaged. It should at least have customer service. You know a phone number you call to talk to someone.

These apologies to mega corporate is very sad. I am disappointed in you.

[Nasrudith]

That seems a bit ironic of an accusation given all of the people who don't do even back of the envelope math about what their demands for scaling would cost to try to do manually. Isn't that like saying plumbers are arrogantly obsessed with imprisoning water using pipes instead of free flowing streams of water while not considering the evaporation and contamination of said approach?

Scalability of algorithims is how they became a behemoth in the first place. I would call it being ruthlessly un-ideological in a sense. While there are implicit goals the approach is more "whatever works" for the system as a whole. While annoying it isn't a breaking point that gets people to exodus.

It is ultimatelty "democratic" in being what is wanted instead of what makes sense. Like the yotube advertiser exoduses where they stupidly think they need "advertiser friendly" content or it will harm their brand when nobody gives a fuck that they advertise on content where people swear or talk about sex.

[megous]

Techdirt doesn't sound like a small website, so presumably it generates quite some money for Google.

How hard it is to gate algorithmic decision making by (sum_of_ad_money_from_customer(x) > SOME_AMOUNT) goto human_review;

[alasdair_]

Or even just have a “pay us $150 for a month’s support” option.

[nomel]

> One thing I don't understand is why Google continues to try to use algorithmic solutions for content moderation.

I don't mean to sound condescending, but I think the answer to this question would be much clearer to you if you took a moment to try to comprehend the scale that google operates in.

[tpxl]

Some random source on the internet tells me Alphabet made around 7 billion profit in a quarter last year. That's 28 billion per year. At 100k/year per person you get 280 000 people doing moderation for you to breakeven. According to another random source Alphabet has about 120k employees.

The scale is not the problem, the profiteering is.

[pwinnski]

Somewhat ironically, the very site at issue has long claimed that content moderation at scale is impossible[0].

[0] https://www.techdirt.com/articles/20191111/23032743367/masni...

[edmundsauto]

Human solutions have problems too. Inconsistency, monitoring, access control. Every employee is a potential leak or Twitter takeover. Building systems to manage these things are difficult, painful, and could make Google do work they aren't very good at.

Also, we have no idea how their support teams metrics are doing. Maybe the automation is actually superior to human customer service.

From the outside of the castle, what we see is more a result of our projections rather than what's going on inside.

[InfiniteRand]

One thing I imagine is if you scale up assuming that algorithmic solutions will be good enough, and then find out that they aren't, you find yourself in a hole, because you haven't budgeted enough to hire human moderators now, and you have enough visibility and investors that it is hard to scale down.

There was some post on HN I think about a Morrocan links site that basically got into this situation, his solution was to freeze sign-ups and look for someone to buy his domain, which is probably not an appetizing option for Google. I think MySpace's experience was also somewhat like this.

Of course, in reality the situation with Adsense is not that they cannot function while using algorithmic moderation, but that they cannot provide a high quality service, so whether they like or not the most viable option is to provide a lesser quality service.

[benlivengood]

My guess is that it's pretty simple. There are ML models predicting the ad quality score; when the predicted outcomes for a site/account don't match reality it is a strong signal that something is wrong. Other inference models determine what is likely being violated on the site or by the account. A human has no chance of digging through the model to figure out exactly what causes the flagged outcome. The only solution is trying various changes on the site and seeing what the models and clickers/consumers think of it unless you want to try to pay Google to deepdream your site for you with their inference models; good luck with that.

It's not worth it for Google to fix rare edge cases; fixing them manually may even bias the models to cause problems for a greater number of sites. I am sure the models are trained as well as humanly possible; there are millions of dollars on the table at the fringes. Every employee's time is most valuably spent improving the models as a whole and not chasing down one-off edge-cases unless it's for a very high profile client. I am sure there are whitelists (and blocklists).

Accept AdSense for what it is; a highly-profitable-for-Google way to use ML to turn HTTP into $$$. It will not work for every site, only 99.99xx% of them.

[throwaway45349]

> I am sure there are whitelists (and blocklists).

Off topic, but why is whitelist acceptable but not blacklist?

[y0ssr3n]

I think the preferred term is "allowlist" not "whitelist". I like Allowlist personally because it's immediately obvious what it means esp for less technical folk. It takes time for new terms to become commonplace.

[throwaway45349]

Post hoc rationalisation, but okay. My non-technical friends know exactly what a whitelist/blacklist are but none of them have ever heard these new politically correct terms.

[SpicyLemonZest]

It's a hard problem, because the alternative to algorithmic solutions is human solutions, which people are quite unhappy with as well. Facebook for example has gotten a ton of negative press over the years for their human content moderators.

[asdff]

The problem with facebook isn't that they use humans, it's that they treat them very poorly. Treat people with dignity and respect, and the good press will follow.

[disgruntledphd2]

In this space, it doesn't matter how many times you get it right, just how many times you got it wrong.

Also, it was probably a bad idea for Google and Facebook to be so cavalier while they destroyed print advertising (i.e. the press).

Never pick a fight with people who buy ink by the barrel, as they say.

[kiba]

Who knows? They had decided long ago that genuine tech support is not in their business model, and that's the end of things.

[jariel]

"I don't understand is why Google continues to try to use algorithmic solutions for content moderation"

That's really easy to understand.

Humans cost money, algorithms are free.

Also, Google is full of the kind of people that 'don't understand humans' and 'understand algorithms very well'.

To Google, the answer to all of life's problems is probably 'an algorithm'.

[M2Ys4U]

> It's refreshing to see an assumption of incompetence rather than malice

To mis-quote Arthur C Clarke, any sufficiently advanced incompetence is indistinguishable from malice.

[alehul]

> why Google continues to try to use algorithmic solutions for content moderation

When people assume malice so often as you mentioned, it, at least intuitively, feels easier to fault human moderation than algorithms.

Algorithms offer a black box which protect against accusations of bias. If accused of bias, an algorithm's biases can just be labeled unintentional accidents, whereas a human's biases are malicious.

[srndpty]

If no one tries it, no one will ever get it right. Sure, they could do it alongside real support in a simulated way, but that doesn't seem like it fits with their MO.

[mtgx]

The explanation could be that Google would prefer to use the excuse that "it's our objective algorithm that downranked our competitors, not us!"

And before you say "Aha! There you go assuming malice instead of incompetence" - keep in mind that Google has used this "our algorithm is objective so we don't manipulate what appears on our search pages" for years and years. It's the go-to strategy they use to appear neutral when called upon by governments.

Facebook has also attempted something similar in recent cases.

[syshum]

At some level though incompetence has to transition to malice.

Do we really believe Google does not have the resources to hire and retain competent people? or are they willfully incompetent as a shield or cover for their malice

How many times can they play "oops our bad it was just a mistake"

[throwaway45349]

An organisation the size of Google's probably fucks up thousands of times daily. I reckon <1% of issues are actually complained about, so yeah, I'd believe it's a mistake.

[john_moscow]

Because Google has a skin in the game. It is a de-facto monopolist in the online advertisement business, and online search. And it's only the political goodwill that keeps it safe from antitrust scrutiny.

So, it's a kinda wink-wink nudge-nudge pact, that Google politically aligns with the Democrats, and the Democrats turn a blind eye on Google's monopolistic state.

There are plenty of examples, just try searching "Newsom Estate" on Google [0] vs DuckDuckGo [1] and count critical vs. approving articles out of first 10 results.

[0] https://www.google.com/search?q=newsom+estate

[1] https://duckduckgo.com/?q=newsom+estate

[rubyn00bie]

Critical vs approving articles is a very weak proxy for antitrust. In this case most of them come from the same source (a conservative media company). It seems to me Google is better about removing useless results, or that which are regurgitated (copied) content, than DDG.

I'm not saying Google at this point doesn't have a monopoly. I am saying that the comment above reads more like a conspiracy theory than truth. Doubly so when you read the sources listed as they're all regurgitating the conspiracy-fodder into one another.

[john_moscow]

I would agree with it if Google results showed one page from that conservative media company, while dropping others as redundant.

Instead, from what I can tell, it heavily penalizes sources that compete with the mainstream media, while filling the gaps by doing looser search (like replacing 'estate' with 'property' and suddenly finding 'How Gavin Newsom plans to close California's huge budget gap').

So if you google for common divisive topics, you get an impression that the public unanimously supports the narrative pushed by the Democrats (that happens to actually cause more division), while in reality there are many critical voices calling it out, that Google conveniently omits.

[joshuamorton]

I'm confused, the two main headlines, one about property tax, and one about being gifted the estate, appear on both engines.

DDG cites redstate for both. Google cites the californiaglobe for the first and rightondaily for the second.

The root of the issue seems to be that RedState isn't considered high quality by Google, which is reasonable since it's about as partisan and reliable as OccupyDeomcrats is.