Hacker News Reader

Comment by tomweingarten | Hacker News Reader

The comments he made are not just questionable, they're outright wrong (and a great example of the problem with cherry-picking data):

https://www.vox.com/2016/6/10/11903028/hillary-clinton-googl...

https://www.politifact.com/punditfact/statements/2016/jun/23...

(Disclosure: I work at Google, but this opinion is my own)

View on HN

Google's claim that the algorithm is generic is demonstrably false. Type in "hillary clinton e" and there is no suggestion for "email", type "donald trump e" and email is the first suggestion. Given the news content that we know is out there, that can only be the result of adjusting the results for clinton specifically (if anything, we would not expect "email" to be autocompleted for trump). This is not research that tells us what exactly Google is doing, but you cannot deny the example.

This is not the most scientific test, since previous searches are generally taken into account. Was this test conducted from a system that mostly searches for / clicks on pro-trump or anti-trump content?

Well, that's kind of the point: it's not scientific, but it's relevant. I believe this was also the example that was recently used in a Project Veritas video, with the same results.

I searched from a Firefox private window over a VPN from the Netherlands. But since the results are the same (regarding presence/absence of "email" as an autocomplete term) I don't think it matters much.

This is not "research" period. Using one arbitrary search comparison to draw conclusions about the nature of a system that processes billions of queries a day is pretty weak. Additionally, I don't get the same results you do. "hillary clinton e" does not bring up emails, nor does "donald trump e" bring up emails (the first results I see are election, education, england visit, ex wife).

I'm not ruling out the possibility that google actually is manipulating search results, but this is not proof of that.

Try "hillary clinton emai". From a fresh chrome session in NYC I get nothing, not a single autocomplete result. On the other hand "donald trump emai" gets:

* donald trump email * donald trump email address * donald trump email list * donald trump email newsletter * donald trump email list signup

And just to drive the point home I tried "root_axis emai" and got "root_axis email". Try anyone else and you get similar results, 'barack obama emai', 'george bush emai', etc etc. So yes, this is proof that the results are scrubbed for Clinton email.

Seems to go both ways though as I'm not getting any auto completion results for "donald trump stormy daniels" either. I'm guessing they scrub things that are highly sensationalized in the news.

"stormy daniels" doesn't get any autocomplete results for me even without trump, my guess is this is more about adult search terms getting blacklisted rather than political. For example "donald trump e j" gets "donald trump e jean carroll", E Jean Carroll recently accused Trump of a serious sexual assault and this is autocompleting.

I got curious and tried the names of a bunch of public figures. Some of "<first and last name> e" yielded "email" as the suggestion. But these did not: elizabeth holmes, tom jones, tom cruise, brad pitt, gwyneth paltrow, roger federer, will smith, jimmy carter.

Since Hillary Clinton is not unique, then it's not proof that her results are treated differently.

They all work fine for me, e.g., 'tom jones emai' -> 'tom jones email', only Clinton didn't.

None of those other people were involved in major stories with email in the headline. If you look at the actual results, you'll see that it doesn't make any sense to not get suggestions for "Hillary Clinton email".

Compare these results:

* https://www.google.com/search?hl=en&q=hillary%20clinton%20em...

* https://www.google.com/search?hl=en&q=will%20smith%20email

Why would you expect “brad Pitt email” to be something that auto completes? You would, on the other hand expect “Hillary Clinton email” to auto complete because there was a huge controversy about it.

I’m not saying google is manipulating auto complete intentionally (though they might be), I’m just saying your counter examples are irrelevant.

It would be like “Donald trump Russia” NOT auto completing then someone saying “but neither does Taylor swift Russia, so we’re good.”

The poster claimed Hillary Clinton was unique, meaning the only person that applied to. For me, she was not. Since she's not unique, then her being unique can't be used as evidence.

Claiming that she's unusual, since you expect it to work for her based on stories written about her is a different claim.

You're misrepresenting my claim.

1. I said "only for Clinton" in the contect of Trump vs Clinton. Then I compared to other U.S. politicians, where the example still holds. The intended meaning is perfectly clear.

2. Obviously the fact that Clinton was the topic of a scandal involving email is the assumed context here. That's why I said "if anything, we would not expect "email" to be autocompleted for trump" (implied: but for Clinton, email is a more relevant search term based on published news, etc.).

When I type "Donald Trump R" (or Ru or Russ etc) no autocomplete results contain the word Russia despite plenty of news coverage. When I type "Donald Trump Epstein" no autocomplete results. Must be a conspiracy by google to protect the president... or drawing conclusions based on individually cherry-picked autocomplete results is like drawing conclusions from numerology.

The same thing happens with Russia for "barack obama rus" "hillary clinton rus", "george bush rus", etc. - none autocomplete for me despite the fact that there was lots of news items relating those politicians and russia during their careers. However Clinton is the only one that doesn't appear to autocomplete for email, suggesting that it is specific to her. When I type "donald trump epstei" I get "theo epstein donald trump" (for some reason the word order is flipped), which would suggest to me that epstein is not blocked in the way you suggest, and it's just that not enough people are searching that term, or that the autocomplete algorithm hasn't caught up to the latest news on epstein and trump yet. However "donald trump e j" does autocomplete to "donald trump e jean carroll", which is relating to a serious scandal for the president. This isn't cherry picked I'm afraid, it really does look like intentional blocking of "hillary clinton email" from autocomplete.

None of those people had a gigantic Russia scandal though, Trump did, so you must still account for this unexplained aberration. If anything, the Trump/Russia scandal had more coverage than the Hillary e-mail scandal, so it's an even more difficult aberration to explain.

Also, if I type "hillary e" I get "Hillary Emails PDF" as an autocomplete suggestion. If I type "clinton e" I get several email suggestions "clinton email PDF" ,"clinton email film", "clinton email FOIA", "clinton email download".

Please include these confounding results in your analysis

Actually both Obama and Clinton had major stories relating to Russia including the failed 'reset', the annexation of Crimea during the Obama administration, obama's "red line" of using chemical weapons in syria which was circumvented by russia, and Obama telling Mitt Romney that Russia was not a threat in the 2012 presidential debates. So yes, they have huge stories relating to russia. And your autocomplete results just bolster the evidence that 'hillary clinton email' has been made a special case and is not organic!

Those stories barely saw a fraction of media coverage compared to Trump/Russia. The Obama/redline thing was primarily reported as a story about Syria, not Russia. If you type "obama red line" you get plenty of Syria suggestions which makes sense.

This is the problem with search query anecdotes, it ultimately produces a subjective and pointless debate about how one should interpret search suggestions for arbitrarily selected one-off queries. There is no methodology here, and we don't even know how widespread any particular suggestion results are. Any person with an agenda will be able to cherry-pick search queries that confirm their narrative.

I can believe Google is attempting to de-emphasize scandals on both sides. As an aside, I do get "russia investigation" as autosuggestion for trump.

I think the bigger problem is that these adjustment appear to be done manually. We know that's what they're doing to avoid racist autocomplete phrases (and reasonably so, I think most people would agree). Having such judgments made on specific political topics/events/scandals will inevitably result in political bias, especially in an organization whose workforce is so politically skewed to one side.

My only point is that autocomplete results for arbitrary one-off queries isn't instructive, especially because many people see different autocomplete suggestions. As I've already pointed out, its possible to find strange results for anything if you look hard enough.

If you type "Donald Trump Ru" into an incognito search you will not get the Russian Investigation. To GP's point, it's numerology to keep focusing on this and drawing conclusions of political bias.

For me "donald trump ru" autosuggests "russia investigation" (4th place). Compared with absence of "email" even for "hillary clinton emai".

Try news.google.com. It works there.

"hillary clinton emails" is a topic that is widely published around the web. Autocomplete should pick up on this and recommend it. I even went to the trouble of typing "hillary clinton emai" and no autocomplete suggestions were brought up.

It is a rather suspicious result. Suspicious enough that it is hard not to imagine a deliberate act is behind it. I admit that I don't have any proof.

That's the point: this is not research, but whatever is going on at Google, the explanation has to account for examples like these. It's simply one observation that you cannot discount.

I just tried searching again a few times with new private windows, and "email" alternates between first and fourth suggestion for trump. But the more important point is the absence of the suggestion for clinton: we know it's been in the news extensively, we know people searched for this phrase a lot, and now "email" has been removed from the suggestions only for Clinton. I tried searching a few more U.S. politicians, and for all of them "e" autosuggests "email" somewhere between first and fourth place. So the complete absence for Clinton does not look like a generic algorithm change.

When I type "Donald Trump R" I don't see any autocomplete for "Donald Trump Russia" despite plenty of news coverage on this topic. So what? This isn't proof of anything. I can indeed discount "one observation" because it is literally a single search query used to draw a conclusion about an insanely complex system that processes billions of queries a day. I am open to the possibility that google is manipulating search results, but to demonstrate this you need to account for many other possible search queries that produce seemingly unexpected results. Dissecting one politically charged query and claiming it is proof of google's malfeasance doesn't make sense. Anyone can string together a couple strange query results to support their own subjective narrative about what should appear and the supposed sinister machinations behind the query results.

Right, and further observations show that it's not unique to Hillary Clinton. This means you can discount the claim since it uses cherry picked data.

But we still haven't answered the actual question: Why doesn't "email" come up in autocomplete for Hillary Clinton, when it clearly should?

Yes we have. Further observation shows Google removes autocomplete for controversial items, like "Russian Investigation" for Donald Trump. If that example doesn't answer your question then you have confirmation bias.

The question doesn't need answering any more than any other arbitrarily selected individual query needs an answer. Why does "trump helsinki" have zero suggestions? Why does "bill oreilly sex" have zero suggestions? Why does "alex jones sandyhook" have zero suggestions? I used right-wing examples because I presume any celebrity or left-wing examples will be considered evidence in favor of your position, but there are plenty of examples all over the place.

How is autocomplete a sign of bias...Take an average voter...not very plugged in..he types in "hillary clinton e" and gets no autocomplete suggestions...he thinks Hillary Clinton didn't do anything wrong with her e-mail server? Do you seriously believe this?

Curious....Did you get any results for "hillary clinton e-mails"?

Speaking of cherry picking, I wouldn't use vox as reliable source for bipartisan data.

Or politifact...

tomweingarten can't see past the tip of his ideological nose. It's gonna be such a shocker to him when his megacorp gets shattered into a million little pieces.

https://www.politifactbias.com/

I've seen plenty of politifact and snopes fact-checks that go against the conspiracies that you guys seem to think are underway in those types of organizations.

Do you have an example of Vox being dishonest?

https://mediabiasfactcheck.com/vox/

I'm not sure this answers the question. Clearly Vox has a progressive (maybe originally neo-liberal?) editorial bias, but that may or may not mean that they have dishonestly distorted facts. There is a big difference between editorial bias and dishonest reporting!

>Overall, we rate Vox Left Biased due to wording and story selection that favors the left and High for factual reporting based on only one failed fact check and appropriately issuing a correction to a second. (5/15/2016) Updated (M. Huitsing 5/30/2019)

speaking of cherry picking, I wouldn't use autocomplete suggestions as a source for bias. Has anyone on this thread claimed bias in the search results?

Here is the trend data where google autocompletes two results while not autocompleting other two. The trend data groups the results pretty clearly, however the autocomplete engine does not agree.

If the trend data is not being used for the autocomplete results, what is? And why?

https://trends.google.com/trends/explore?geo=US&q=hillary%20...

do an image search for "european people art"

what's up with that

Reply on news.ycombinator.com