Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Google featured snippets are worse than fake news

theoutline.com

Post reply on HN
Print view View original

121–130 of 192 posts

Re: Google featured snippets are worse than fake news

#121

Post by Bartweiss »

jerf wrote: post #117
theseatoms wrote: post #110

Earlier quoted context omitted.

Google should report a probability distribution for each truth claim.

While an interesting idea in theory... and I mean that!... in practice that's saying they just shouldn't provide answers. Even saying for the sake of argument that Google could assemble a meaningful and correct probability distribution, itself a rather bold statement, the average Google user won't know how to correctly interpret such a thing.

And realistically, asking for probability distributions over unbounded sample spaces is trouble even for human experts. I suppose the space for "are mermaids real?" is workable (there are caveats like "yes but extinct", but yes/no is basically sufficient), but for anything subjective or political it's almost hopeless.

That said, this is a really interesting idea for behind-the-scenes work. The public might not want percentages, but I suspect experts could get some interesting results by diving into confidence levels and secondary answers.

Re: Google featured snippets are worse than fake news

#122

Post by cdubzzz »

ccozan wrote: post #99
cdubzzz wrote: post #78

Earlier quoted context omitted.

Pardon my ignorance, but would such a service need to be specific to a site? I'm wondering if there is a business (or perhaps better yet, non-profit) opportunity to provide data to drive something that can be tacked on to individual sites. Would an organization like Mozilla have an opportunity here?

It depends. Organizations have different data, so you need a lot of abstractions in order to fit into a certain search pattern (meaning, what are the users looking for). For example, current solutions implies there is a "document" with some metadata. But what is a document in wikipedia or SO context? So is quite hard to get a an answer which satisfies both cases and you end up with cramming all kind of data together…

Interesting. Thanks for your insight.

My line of thinking was on working with "what is the user asking" vs. "what is the user looking for" in a way that could be applied regardless of the information being searched (by, for example, a non-profit with some existing insight - hence the Mozilla thought). But I know almost nil about the space so I suspect that is somewhat naive.

Re: Google featured snippets are worse than fake news

#123

Post by deong »

DanielBMarkham wrote: post #118
deong wrote: post #112

Earlier quoted context omitted.

It's not even limited to this kind of question for which there's no simple answer to give. Last week, my wife was joking around with our 9-year-old, leading to him trying to convince her that mermaids aren't real. Cue, "OK Google, do mermaids exist?" To which our Google Home announces, "According to Weekly World News... ". If "do mermaids exist" can't be classified as a thing you know, I'm not even sure you're at the…

A common complaint that pg and many of us have about popular forums is that you always end up in some kind of semantics game. That's true, and that's frustrating, but that's also the basis of shared human understanding, the dialectic. I wish I could just say "There are no mermaids", but the definition of "is" and "being" here is complex. In many other cases this definition is not complex. We programmers get dissed a…

Context is important. For a general question-answering bot, you have to synthesize information and make a decision. The answer to a question posed by my uncle on Facebook isn't necessarily the same as the answer to that question posed by an audience member at a talk I'm giving at an academic conference. In the latter case, there's an enormous shared understanding that lets us skip the basic-agreed-upon facts and get right to the nuance. My uncle doesn't need nuance. He needs to know whether climate change is a Chinese conspiracy or vaccines cause autism.

Minority opinions are sometimes the thing that advances the species, but that's a job for a time and place that isn't "answering random questions for arbitrary users".

The answer to "do mermaids exist" is "no". It just is. That doesn't preclude a group of 200 experts devoting their research careers to investigating the question, and if they happen to prove otherwise, then we alter the textbooks and move on. Ideally, Google would then start answering "yes" instead. But if you want to be that world's dominant authority at answering questions, you can't just always say "well, it's complicated. People disagree. The world is a complex place and the definitions of terms and the facts are in flux", because you're no longer answering a question.

Re: Google featured snippets are worse than fake news

#124

Post by DanielBMarkham »

deong wrote: post #113
Sharlin wrote: post #102

Earlier quoted context omitted.

Plus, of course, it depends on which year you're talking about, in a very nontrivial fashion. Googling "when is easter" gives the date of the current year's Easter, which is reasonable, but might lead to an incorrect assumption that the date of Easter is fixed.

That's true, but I don't think it's a problem. That's a question that just has two separate intentions, and even the world's foremost expert on Easter could give you the wrong one. If Google were as good as the best human experts at answering any question, then I think we'd have to declare victory and live with the occasional miscommunication, just as we do today.

I think the crux of the issue is that real experts in any area, when presented with a question in their field, will answer "That depends" and then engage you in a conversation that ends with you being more knowledgeable about the specific question you should have asked, and why.

This is because most areas of human endeavor aren't solid sciences like physics, although for most day-to-day activity we treat them as if they were. I do not care about the nuances of Easter as long as I can get the flowers delivered to my grandma on the correct day.

But I do care about the nuances in a ton of other situations, and the dividing line isn't so clear. Simply BSing an answer that's good enough for most cases actually ends up hurting the rest of us more than it helps us. For lack of a better term, it's fake facts.

Re: Google featured snippets are worse than fake news

#125

Post by jasonthevillain »

TuringTest wrote: post #51

Why is it that hard for engineers to rely on good old attribution? If every Google's featured snippet started its reply with "Breitbart says..." or "Trent Online, the leading Internet Newspaper in Nigeria said...", it wouldn't matter so much for those inevitable cases when the reply is taken straight from a white-supremacist or radical anarchist forum. The problem comes when the same reply is provided as "Google's tr…

Well, for example, "According to The Vaccine Research Institution..."

Now I just made up that organization. But how can an algorithm tell if it's a legitimate medical institute that develops vaccines or anti-vaxxer propaganda?

If the algorithm actually had a way to distinguish the two, it could blacklist bullshit just as easily as it could disclose it.

Re: Google featured snippets are worse than fake news

#126

Post by marcosdumay »

Vinnl wrote: post #57
TheGRS wrote: post #17

While hiring a ton of people to fact-check would be one solution, that obviously wouldn't be very scalable. I think the problem lies in google's algorithm. They seem to be pulling answers from well-visited sites that may purport to have an answer to these questions. Quantity of visits does not equate to truthfulness. What Google should at the very least do is whitelist certain publications for answers to pretty simpl…

A whitelist is sure to get them anti-competitiveness suits, and rightly so.

What is the difference between a whitelist and an algorithm with unknown but discoverable biases?

Re: Google featured snippets are worse than fake news

#127

Post by dmboyd »

ronack wrote: post #120

I've also wondered why Google isn't held responsible for publishing libelous claims and hoaxes as facts. Examples: Is Hillary Clinton a pedophile? https://www.google.com/search?q=is+hillary+clinton+a+pedophi... Is John Travolta gay? https://www.google.com/search?q=is+john+travolta+gay Does Lil Wayne have HIV? https://www.google.com/search?q=does+lil+wayne+have+hiv Perhaps worse, this is what drives Google Home's ques…

Wow, I've mostly converted to DuckDuckGo for the past two or so years, so are these "google assistant" style placements now part of their core product? or is this a setting that you've enabled?

If it's part of the core... Wow, what a useless cess pool google has become.

Re: Google featured snippets are worse than fake news

#128

Post by hobofan »

noxToken wrote: post #103
bigbugbag wrote: post #53

Earlier quoted context omitted.

Google stopped providing relevant search results more than 10 years ago. It used to be several pages of relevant search results, now you are lucky if you have more than 2 or 3 somewhat related results in the first page and that's about it. You can tell there was a paradigm shift at some point and the search engine became a second (third? fourth?) class citizen.

I think there are two things at play here. This first is that you were probably a power user back when good search results required you to be good at search syntax. Back when Google wasn't the only search engine on the web, you could ask Google questions. Your results would be limited in relevance. If you tailored your searches by ensuring specific words or phrases were included, excluded irrelevant phrases, and adde…

Your first point definately sounds plausible.

Another theory I have is that my perception of good results changed over time. It could be that for the first time in my life I know enough about the domains I'm looking up that I can differentiate the good from the bad. On the other hand even with that, when I enter a new domain, e.g. AI it still feels "less discoverable", even though there are definately good introductions out there that you can find when going beyond Google.

Re: Google featured snippets are worse than fake news

#129

Post by gregmac »

abandonliberty wrote: post #25
TheGRS wrote: post #17

While hiring a ton of people to fact-check would be one solution, that obviously wouldn't be very scalable. I think the problem lies in google's algorithm. They seem to be pulling answers from well-visited sites that may purport to have an answer to these questions. Quantity of visits does not equate to truthfulness. What Google should at the very least do is whitelist certain publications for answers to pretty simpl…

Metrics are destroying the web. Once upon a time it was possible to identify "good" pages by how often they were linked - effectively crowd-sourcing the problem. Natural selection: sites have learned to manipulate the game and the crowd. Under pressure to optimize metrics that lead to SEO and better valuations, the internet is getting less useful from a user perspective. I don't want to watch a video/slideshow, downl…

What's strange is this is not at all a new problem, and Google has historically been quite good at fighting it.

The original search engines ranked pages by their content. Naturally this led to gaming by including keywords (remember huge invisible sections of pages that just repeated keywords thousands of times?).

Google's original PageRank algorithm was a complete breakthrough for this, almost completely disregarding content and instead ranking results based on the text other pages had used to link to the page. This was so good, in fact, that most of the other search engines from the time didn't survive.

This once again led to large-scale gaming with techniques like link farms, and a small arms race as Google came up with ways to squash these new techniques. The quasi-legitimate "SEO" industry spun out of this.

I think now we're in the same cycle again, and at this moment scam sites are winning. What's to be seen is if there's a looming breakthrough, or the arms race will continue.

Re: Google featured snippets are worse than fake news

#130

Post by skocznymroczny »

DaUR wrote: post #77

Google wants to avoid editorializing, and be able to throw their hands up and say "not our fault!" due to the plausible deniability of "algorithms". They can't pull that off anymore. There needs to be a shift towards a Web-of-Trust-like system, where some sources are recognized as authoritative. Government websites, like the CDC, for example. Big media outlets. Scientific associations (IPCC, APA, etc). Not all inform…

In other words, censorship. Would CNN be considered more authoritative than Fox?
Post reply on HN