Non-Fic Google is apparently altering search results

The essence is people despise being lead on or disadvantaged by unfair bias motivated result. I agree though since I've been using Yandex, Duck Duck go, and Google interchangeably to find what i needed.

 

Well... Even your friend or people around you can lead you on or put you in disadvantage position due to unfair bias-motivated words. So it wasn't that different from the real world. I think they trust Google too much... Even Wikipedia and Pubmed had their own fault...

 

The Muller report already said in its conclusion that there was no Russian collusion
And Google has already taken down the video talking about them and also a video talking about pinterest censoring right wing people and again an email came out showing their bias and intent on censoring them ,Google are not the good guys here come on let's not even kid ourselves we long knew this was coming and that big tech companies have been doing this for years

 

You're completely wrong on what the Mueller report says. Here's where you can read it. It does not exonerate them from anything. In fact, it even provides proof for Russian interference, and while it doesn't prove collusion it lines up a lot of very suspicious happenings that should at the very least make you alarmed. The investigation was actively hindered, and individuals from the campaign have been convicted for it. For that reason it also suggests further investigation and action. Anyway, while collusion isn't proven it is also FAR from disproven!

Let's take a look. First, influence on the elections by the Russians? Chapter II. Russian "active measures" social media campaign. This section literally starts off with: <The first form of Russian election influence ...> so this already says a lot, no? The Mueller report agrees on Russian influence on the elections. Then, Chapter III: Russian hacking and dumping operations. First thing it says: <the hacking in the DNC was done by the Russians>. Surprise! Is that not influencing elections? These are two things I quickly took from the report, but it's got more ammunition if you so desire.

Now, on to collusion. Interference in the elections is a fact as we've seen by now, but what about active participation with a foreign country, collusion? This is sort of like the villain in a novel getting support from a foreign kingdom. <The Office identified multiple contacts-"links," in the words of the Appointment Order between Trump Campaign officials and individuals with ties to the Russian government.> So the Trump campaign was in contact with people with ties to the Russian government throughout the election period. Doesn't that ring alarm bells? It's Russia, the USA's biggest enemy of the past half century or so, a country ruled by a dictator. I'm not going to transcribe the report here but it basically lists a whole bunch of contacts or stakes Trump and his campaign had with Russian-tied figures. It's a lot. E.g. secret back channel communication with russian agents: for example Jared Kushner being in contact with Sergey Gorkov, who graduated from FSB academy.

Here comes the part where you, some media outlets and some very relevant figures are so confused about. After describing what collusion would be like, the report says, <Based on the available information, the investigation did not establish such coordination.> What does this mean? It means there is no conclusive evidence of collusion, of working together explicitly. But here's the kicker: nowhere does the Mueller report say Trump or the campaign peons are not guilty either. It basically says, we weren't able to determine it conclusively. In fact, the Mueller said <the Constitution requires a process other than the criminal justice system to formally accuse a sitting president of wrongdoing>, which basically means it advises the Congress to either investigate further or to start impeachment. Shocker!

So, let's conclude this with an explanation as to why there was no real conclusion. Some quotes to go out.
<First, the Office determined that Russia's two principal interference operations in the 2016 U.S. presidential election-the social media campaign and the hacking-and-dumping operations violated U.S. criminal law.>
<Second, while the investigation identified numerous links between individuals with ties to the Russian government and individuals associated with the Trump Campaign, the evidence was not sufficient to support criminal charges.>
<Third, the investigation established that several individuals affiliated with the Trump Campaign lied to the Office, and to Congress, about their interactions with Russian-affiliated individuals and related matters. Those lies materially impaired the investigation of Russian election interference. The Office charged some of those lies as violations of the federal false statements statute.>
<The investigation did not always yield admissible information or testimony, or a complete picture of the activities undertaken by subjects of the investigation. Some individuals invoked their Fifth Amendment right against compelled self-incrimination and were not, in the Office' s judgment, appropriate candidates for grants of immunity. The Office limited its pursuit of other witnesses and information-such as information known to attorneys or individuals claiming to be members of the media-in light of internal Depa11ment of Justice policies.>
<Further, the Office learned that some of the individuals we interviewed or whose conduct we investigated-including some associated with the Trump Campaign---deleted relevant communications or communicated during the relevant period using applications that feature encryption or that do not provide for long-term retention of data or communications records. In such cases, the Office was not able to corroborate witness statements through comparison to contemporaneous communications or fully question witnesses about statements that appeared inconsistent with other known facts.>

Essentially, the goons from the campaign did their best not to cooperate: not talking, trying to get out of being questioned, deleting evidence, and so on. Interesting behavior! In the end there wasn't sufficient evidence to prove actual collusion, but opportunity and motive sure where there and there was a lot of contact. And a lot of "coincidences". As an earlier quote indicated, the Mueller report said they needed another institution, Congress, to take over the mantle of the investigation.

Again, every single piece of evidence you put here has been utterly worthless and your reasoning flawed. I still haven't seen any evidence of Google specifically targeting the far-right (mind you, being filtered out because of e.g. anti-vaccination talk or the likes would not be specifically targeting) in this thread and we're what, five pages in?

The article you first posted here showed a clear disregard for context, they were eager to make a collage of quotes and did not have any of the required technical knowledge. To top it off, they made it seem like they were investigative journalists while they're actually a site with a very clear purpose. Just like you using that Obama video, you ignored context, the time at which it was filmed and misinterpreted what he said.

As I've said throughout this thread, I'm all for checks and balances on the methodology of such an influential company. But do not assign guilt before there's even evidence of anything, without context.

 

I never listened/watched Jones, and I can't honestly say that I didn't want him to stop spouting his nonsense. However, in the US there is a freedom of speech thing, so if the guy wants to go out and talk about stupid crap - he can. And google or facebook should not be able to silence him, at least not in the case if they are truly neutral as they legally claimed and got some protection from the law because of it - they are not responsible for what is posted or said on their platforms unless it's outright illegal stuff. If they want to censor people according to what they like - they should change the terms of service, start to carry responsibility for the content and being regulated.

---
Among other problems with neural nets there are two prominent ones that makes them unsuitable for certain tasks - they are unsound and no one knows why they work the way they do.
Neural networks are basically approximations of computations, therefore they are by definition not precise. They work for things like machine translation because you essentially map one language to another one to one. However, if you try to generate any reasonable text from non-linguistic input using nn, you are in for a load of ungrammatical sentences. Half of the time you would get a sentence that is somewhat readable, another half of the time it will be just gibberish. One doesn't need to go far, but to just go ahead and read some novels using MTL to see just how "good" it is. And all of this doesn't even take into account that if you want to increase the functionality/add more words in the dictionary of nns, you will need to retrain your network every time. This is way different from a science of computational linguistics, which creates formally defined and sound models of computation for languages. They are not only precise, they guarantee, unlike neural networks, that if they do produce a result - it will be a correct result. You will not get ungrammatical crap with formal methods.

The reason why NNs became so popular, they perform quite well on a subset of tasks that involve probability models. Traditionally scientists just used Bayesian, ngrams, or regression models. NNs take those and improve on them. But the number of tasks is limited to things like POS tagging, semantic role labeling, and word embeddings. These tasks are far from a complete list of problems that NLP attempts to solve. In addition, other methods, like SVMs work just as well on these tasks as NNs. And in some cases a simpler Bayesian model performs just a tat worse, which is not significant, but it runs much faster. Plus, NNs use pretty much the same stuff as previously staple probability models, such as softmax functions. NNs are just a bit better at working with vector inputs, that's all.

On top of all that NN's are task-specific, unlike general formal methods for NLP. Every little taks will require a separate NN. On the other hand, state-of-the-art parsers can do everything in one step. After RNNs became popular, they came close to the formal methods in some ways, but still have quite a lot of disadvantages. But they still perform overall much worse that formal parsers and generators.

With all that in mind - what google does, is design several neural nets for specific tasks just so that can propose a solution to a complex problem. It is a partial and imprecise solution, but when there is a lack of any - it will still do. Their work only involves improving the NNs, which will get google some money and generate some hype amongst people who only care about fancy words and don't know how stuff actually works. I will give them props for a solid machine learning research and achievements. However, from the point of NLP science - what google does is shit. I haven't seen a single paper from google about formal methods in computational linguistics.
If you wanna read about how NLP actually works - read Chomsky, Joshi and Steedman to start with.

 

Oh, but what I'm talking about isn't outright silencing. It's basically lowering the rank due to being an unreliable, low-quality source, in this hypothetical because of anti-vaccination content by Jones (which endangers public health and is factually wrong). Lowering the rank is de facto filtering. They could also exclude him from some parts of the architecture, making it more difficult (but not impossible) to find his content through the search engine. This is totally fine from the perspective of serving the user - as long as their methodology is sound, i.e., they try to prevent bias and do their best to draw the line properly. This is one of the aspects I'd like to see checked, if it doesn't already happen.

In the end I still haven't seen any proper proof of any intentional censorship, so why proclaim guilt immediately? A healthy dose of cynicism is as always recommended though, which is why I want control on all companies too large/influential and their activities including Google. Lastly, I don't see much incentive for them to take a huge risk over something they can't reap profit off of.

As for disabling specific forms of filtering on a user-by-user basis, it's total madness from a practical and commercial standpoint. Not doable if you look at the graphic of the (very) basic concept of their architecture. If you want to use their product you are stuck with the way they filter. Again, no proof they do it wrongly but it wouldn't hurt for some independent party to be able to check now would it.

Most of the technical talk I agree on. SVMs and NNs have their own strengths, although overall I'd give it to NNs. Yes, (R)NNs are primarily used for specific tasks like embeddings of media or semantic hashing or whatever. Yes the best method differs problem by problem and it's certainly isn't always NNs. It's one tool in the toolbox, but in the end we are stuck with e.g. that MTL and we are stuck with that toolset. And it's a certainty that Google has their own uses and tools to work on the problems they try to solve. Tools, ways of working, architectures they want to keep from other big data companies.

You saying they don't do anything cutting edge, sorry, doesn't hold any weight. There's a reason they don't go open source and it probably is to not give competitors their entire blueprint, their R&D, and so on. Yes, there's a lot of valuable things in there that the competitors would like to see. Then again, you think Microsoft wouldn't make a loss open sourcing MS Office, so ... there's that.

 

Dan Crenshaw is roasting Google in a hearing right now.....

 

Because many of the censored or downranked channels/videos are neutral or don't contain any crap like anti-vaccine. I am not going to give any examples. Go on youtube and search "my video was demonetized" or some variant of that search. Before you go and claim that the videos they downrank are low-quality, go and watch a few first, find out what kind of rules are set up for videos to get demonetized. And I go back to my original claim - it's not up to google to decide what is low quality and what isn't.

No, we are not stuck with what google spits out for us. There is a crap ton of research happening on MTL, there are other tools available too. But this is besides the point

When did I say they don't do cutting edge? I said they don't do formal methods. The edge is cutting differently in different fields. They can be considered cutting edge in the machine learning field and some other areas. You can also say they are cutting edge in NLP since this is the trend among scientists lately, which is pissing me off all the time. Everyone just started using neural nets for anything and everything, getting degrees by running text through various NNs and see which one work better. Now if you publish anything in NLP you have to justify why you are not doing neural network shit. All this NN crap is just a minor part of computational linguistics and so many people started to treat it like some sort of messiah that came to save our world in the name of Alan Turing.

And btw, google publishes most of their research, so algorithms, networks, whatever - most are availible or known https://ai.google/research/pubs/

What they hide is some details, which can be considered corporate secrets. Like data, specific weights, etc.

And even if they didn't publish, a lot of this stuff is not that hard to figure out at least approximately. There are only so many ways you can do stuff.

 

If you don't want others to decide how content should be ranked, then you shouldn't use any search function anywhere at all. Like, they have to rank it somehow. I've not yet seen any proper proof of censorship on e.g. right-wing whatsoever on the search engine Google. And being able to configure filters is a pipe dream.

As for YT, demonetization is another matter entirely and deserves a separate discussion. It's not about ranking or quality here - in the first place, it's whether they can sell ads next to that video to their advertisers. Google and the advertisers agree on a set of terms, and they demonetize if a video doesn't conform. In the end, all sorts of videos get demonetized for the stupidest reasons, seems almost indiscriminate. Do demonetized videos get less views? From a commercial standpoint it would make sense to recommend videos that do run ads, so that's probably how it goes. It's business, indiscriminate, not targeted censorship.

So I don't know what point you want to make but it sure as hell does not have anything to do anymore with the subject of this thread.

There is a world of difference between a rough architectural sketch/theoretical concepts/general principles, and an actual efficient large-scale implementation. I'll repeat again that they probably have plenty to hide from their competitors, which would easily explain why they don't open source parts you'd like to see.

I only brought up NLP as a possible field where they might have an advantage over their competitors, where they might have practical knowledge. Rant as much as you want but you're not telling me anything new. NNs are just a tool? Sure, preaching to the choir here.

 

What I want to know is why is it that difficult to distinguish between a search and a filter. Ranking and filtering are two entirely different concepts. There is also a difference between searching for a relevant content and filtering out ides google find inappropriate for you. Why do you even think that there is something like "low" or "high" quality parameter? Quality of the content is not something that is well defined and definitely not something that google even can decide. They might be able to rank something higher or lower based on the language used in the page, but they have no idea if the information on the page is good or bad, correct or not, something you want to see or not. You do realize that many people considered Alex Jones a good and high-quality content, right? If I search for "vaccine", do I want to get pages with anti-vaccine content? Yes. Should they rank lower? Yes. I wanna know what kind of crazy shit people are thinking. If google arbitrarily decides outright delete all anti-vaccine content, that's manipulation and propaganda. They did ban Alex Jones, didn't they? I disagree with everything he says, but he has the right to say it. If someone doesn't want particular pages to appear in the search, they should refine the search - google provides a way.
Do you seriously not get the difference between relevance, quality and validity? Any search service should be just that - a search, providing relevant information, and leaving to the users to decide the quality and validity of it.

You actually have no idea how scientific research works, right? And you didn't even look at a single paper listed under google publications, right? The only "abstract and rough" type of papers you get are: papers on theory, and papers when students or researchers just start their research. Even with that, there is not a single reputable venue that will publish a paper that does have practical results (theory papers excluded). If they do publish an algorithm or a new network, they publish everything. The structure of NN, the formulas, the functions, the layers, and they even have to provide the code for most reputable venues. Any decent developer who doesn't have shit for brains can read the paper and implement the algorithm. If the algorithm cannot be implemented or the idea cannot be used after reading the paper - this paper just won't pass peer review. They are huge assholes in the peer review committee.
Do you actually believe that developers at google have some sacred development knowledge that they copyrighted and don't share? It's all the same. All of the good practices and ways to do design/architecture/optimizations are well documented and known. Go pick up a book on algorithms and optimization, and it will all be the same. The actual implementation might be more complex if you work not on the numbers, like in a textbook, but on some multi-leveled high level of abstraction data structures, but the basic concepts all stay the same. Google does select good people, so developers at google are just generally smarter and better than average. But they don't have some sort of secret optimization knowledge on how to code. And in cases, like AngularJS, I would question their sanity...
What google hides are their data, some specifics in exactly how they configure their stuff and exact things they use for rankings. These kinds of things, that belong to them.
They did come up with quite a few useful things that are now widely used everywhere by the public. They are in a position to do research that is not available to a lot of other people, and they have to solve some unique problems. But because they have to solve unique problems doesn't mean that they are the only people who know how to solve these problems.

 

Anyway, I've seen no proper proof for targeted censorship. E.g. Alex Jones can cry out false information all he wants, but that doesn't mean the company Google has to feed false information at high rank to their users. How do they construct and train the models detecting high probability low quality information? That's something that in my opinion should be checked, but only because search engines are basically a commodity right now. From the start I never disagreed on that point. As far as we know though, right now they're doing it "correctly" as in not targeting anything specifically. Last time they meddled with it (with an actual motive: money) they were caught.

Well yes, I do know how scientific research works. You also misinterpreted the statement I made.
The academic environment and way of working is not similar to the private sector at all. Though there is overlap, the problems they look at are often posed differently, and the scale is larger. As for structure and architecture, I mean the bigger picture not the design of a single NN.
It's clear Google thinks there's ideas or at least information to be stolen, they're not making critical parts open source and e.g. where I work, they don't either. Yes, companies sometimes have a neat little trick somewhere their competitors don't use.

 

you keep repeating it, but it seems to me that you haven't actually looked. What you are saying is essentially is everyone who gets demonetized/censored/banned by google deserves it without actually going and checking. And again, where did you get the idea that google actually checks for quality of information? It's almost impossible to check with algorithms and is arbitrary. Not in general case, like what google does.

I will use this thread as an example. The video was released on YouTube where a google employee claimed censorship and that google manipulates the elections. She quoted Sergey Brin about how he doesn't like Trump and some other things. Within like, an hour or two this video was deleted from pretty much all social media platforms. Why? You cannot claim that because it's false. If this is the case, why only this video? There are a crap ton of videos with provably false information that still exists. There are also other recent cases where in the exact same situation, one video that exposed a person (whistleblower or the like) was deleted, while another video that exposed a different person in the same situation stayed. You honestly don't have to look too hard to see that at least some sort of censorship is happening. As for the election manipulation - I won't claim it's true, but I also won't put it past them. This is nothing new.

I wouldn't say the problems researches try to solve in academia and in the industry are different. It's a symbiotic thing. They even often work together. The ones in the industry might get better results in some cases due to the company they work at and the data they have, but it's not always the case. Of course in a case like google, when they needed unbelievably huge data centers and distributed systems - no one envisioned this kind of scale before, so it was unique to them.
Also, this is not what you said:

Implying that google publishes rough sketches and later implements them using some secret methods that weren't in the paper. When they published their file system paper, no one cared how they implemented it - it was a well-designed distributed file system that got adopted and improved by the community.

 

I don't say they check for quality manually at all. They train a model to check and let it do the job, and this is where methodology comes into play. That's the whole point. If the methodology is in order, then there's no targeted censoring.

The article this thread is about, and the video, are as far from investigative journalism as can be. Project Veritas is James O'Keefe's project right now, this person got sued for a video like this in 2010, had to settle and pay 100k. This is Alex Jones levels of hilariously bad, and he's actually been caught for it in the past, that's the best part. At least be a bit critical of the source. The reason for removal of the video on YT is apparently privacy, due to footage taken without consent. Thus, breach of terms, a perfectly valid reason to take it down manually. If I got featured in a video like that about where I work, in such a misleading way, I'd be damned fast in getting that video pulled out of the ether as well. So even the "controversy" around it is baseless. Counter for proper proof still at 0.

Election manipulation has a large economical disincentive. If they got caught doing that, it'd be a monumental shitstorm. Again not saying it's not possible, but I don't see why. Russian influence though has been proven, and Russians had motive, unlike what some in this thread seemed to think.

Like yes there's overlap and symbiotic existence, there's doctorates at companies, yada yada yada. You're not telling me anything new. But how the problems are posed, the approach can actually be different. There's things the industry doesn't give a fuck about or doesn't notice, and there's things that go by unnoticed in academic research until only after practical application encounters problems. Of course a company like Google has some trade secrets.

What they let out in the open, publicize about or open up the source from, it's what they choose to show. What they don't show, they don't show for a reason. They don't want competitors to know how exactly their search engine, arguably the best there is, works. They don't want to show how they combat e.g. link farms and the like. It's about edges over their competitors.

Again, you saying nothing special comes out of Google means literally nothing.

 

To be expected

 

I am saying that the term "quality" is not defined and you can't train a model for it. Not in the same way you are implying it. You can train it within a specific domain, like checking the validity of product reviews. But in a general case, where you have to do it for everything - that's impossible. In addition, you can easily trick such checks by using some key phrases yet writing crap.

It has no bearing here if it was journalism or not. Privacy complaint is also bs because there are videos that expose other people in and are not removed. Only this one, that related to google is suddenly got removed. And it got removed not only from YouTube, but from other web sites too, even competitors of YouTube, and it happened all at once. Magic. You know, there is a difference between critical thinking and denial. At the very least there is something suspicious happening.

So... some companies research only things specific to their business, other companies just do normal general research, and academics don't always realize problems people face in the industry.... Next you going to tell me that many developers can't code and that advisers use students to complete labor-intensive tasks under the guise of research? If this was your point from the start, why state the obvious?

You seriously need to get your comprehension skills up... I just spent like 3 paragraphs writing the opposite. Time after time you keep either misquoting me, or putting words in my mouth. srsly.

 

Of course Google is butchering my search results. I used to about a month ago use only Google, and I've been dissatisfied withe the results for years now. Didn't like the interface of Yahoo, Binge, Yandex, (Baidu is in Japanese and is harder for me to use as such) and a few others so I've stuck to Google and a little less than a month ago saw on YT a video about QuackQuackGo and since than I use only QqG. (My Firefox is set to google and safari to qQg and each time I compare them, which I do a lot, I'm left unsatisfied with Google. The search results are constantly unreliable, Google leaves me s*** ton of search results that I don't seek! And the ones I Do need are left outright t....nvm became a rant about Google which I didn't intended to do...f***... )

I'm sorry to correct you but it's yesterdecade ago. Lol.

Sorry, but what?! Not one of them is as reliable?! Other search engines may be not as popular but to call them less reliable? It's like saying windows is the best operating system. I mean come on, almost everyone uses it so it's much better then the alternatives, right? It's like saying back when windows xp and 98 were in heavy use that Internet Explorer was the best internet browser.

-_-"

 

They already said it in a way, a little while ago in this thread lol

 

a like just because of this line.

I do think in the early 2000 Google was the best search engine. But almost a decade now I find the need to double, sometime triple search in other sites. By "improving" they're algorithm (which most definitely hides relevant search results, I really do recommend you to check it yourself against QQG. Don't rely on my word or someone else's. Just check it yourself and you'll immediately notice all that filtering that you don't really need and of the hiding information, you don't have to even check it to far...just search stuff related to novels that are relate to NU and you'll find even there a huge filtering by Google. Which is absolutely not needed.) They got worst.

 

Both of you are a bit confused, it isn't your fault though, it is the media's and politicians fault because in trying to push their narratives they tend tend to go "all or nothing". But leave out a lot of details because it does not go with their narrative.

The whole Trump/Muller thing aside(yes, completely unrelated), Russia has interfered in US elections (and Europe too). This has been confirmed by both sides of the isle, just not talked about much on the media because it has no political benefits on either side.


PS Google as a US entity can legally interfere in US elections, Russia as a foreign entity legally can not.

And if you want to hold google "accountable" for what they put on their platform, you might be thinking you are hurting google cause you don't like them, but in reality you are opening up a door to wide internet censorship. Pick and choose your battles carefully and don't put the cart before the horse.