YEAP
This is very clearly a question guided in a way to show just how complex the system is and works, not a "ELI5"
-
Ever wanted an RSS feed of all your favorite gaming news sites? Go check out our new Gaming Headlines feed! Read more about it here.
Google CEO Had To Explain To Congress Why Googling ‘Idiot’ Shows Donald Trump
- Thread starter Solace
- Start date
YEAP
This is very clearly a question guided in a way to show just how complex the system is and works, not a "ELI5"
This article is pretty great:
https://www.huffingtonpost.com/entr...ceo-sundar-pichai_us_5c1007aae4b084b082fd892e
https://www.huffingtonpost.com/entr...ceo-sundar-pichai_us_5c1007aae4b084b082fd892e
The best part about this is due to the articles being made about this question they likely inadvertently tied the terms "Idiot" and "Trump" together even more than they already were just by asking the question.
Yep, first thing that comes up for me.
Lol. If that had been the response in the hearing, it's possible that I may have died.
Lol. If that had been the response in the hearing, it's possible that I may have died.
This is where I jump in and say it's not that simple. In your scenario there is a series of "code", computer instructions, that could be audited to see if they produce unfair outcomes.
In modern artificial intelligence, the most broadly applicable results for a given investment are instead achieved by taking a very general model for decision-making (a neural network for instance, though my argument here is applicable to a much broader range of machine learning) and training it on a selected subset of data.
Now there are many ways in which this model will tend to reproduce the bias of its inputs. And we should all be concerned about this, not just conservatives.
Suppose we trained that model on police searches of automobiles. Suppose it tended to give a high score to a certain range of skin tints as weighting for likelihood of carrying drugs. So the model is neutral, right? No, not necessarily. Somebody chose to conduct those searches, and they had to conduct them within the constraints of local interpretation of complex search and seizure laws, and probably also hung out in areas where people of a certain range of skin tints tended to hang out.
The bias isn't in the algorithm, it's in the data.
this is like an evolution of every gaming system being called "Nintendo" by your mother.
Just decided to add to it myself because this is hilarious.
True story: in 1990 I bought a laptop. It was an Amstrad 386sx, but in those days the Toshiba marque was so well established for laptops that most people referred to it as a "Tosh." You may occasionally hear people refer to elasticated sticky wound dressings as "band aids". Generic reference is a thing. It will confuse members of Congress, but it's just something we have to deal with.
Trolling when she could've asked about legitimate problems with Google sounds like a valuable use of American tax dollars /s
A coworker called me in to the TV last night at work and said "Check this guy out." He was referring to Google's CEO because he didn't know if I knew who it was. I immediately noticed something he hadn't, which is crazy because it was right over the CEOs shoulder. I saw that dude, with the top hat and mustache, and a money bag with a dollar sign printed on it and cash falling out, sitting there. I said "What, the monopoly man?" He asked what I was talking about because he somehow hadn't seen him. I pointed the guy out and my coworker lost it. I was doing alright but then the guy took out a monocle which he polished and the new affixed over his eye and I began laughing so hard I was in tears.
Apparently the guy is some activist who did this during the Equifax hearings last year.
But think harder about this. If the sample were from the written thoughts of Americans (or any other nation) from just 50 years ago, we'd see a lot of opinions that are now regarded as bizarrely false. Artificial intelligence is not a neutral friend.
In particular, AI systems trained on current assumptions made by those training the systems or gathering the training data will simply launder bigotry into a form that will tend to be treated as impartial. If you're a woman or non-white or fat or skinny or weird or a conservative or have a history of medical treatment, you are right now being judged by a machine learning system whose authors probably couldn't account fully for its decisions. But people will increasingly be expected to rely on those systems.
It was a good use of tax dollars because she did ask a valid question. It was asked so that the CEO could explain that search results aren't manipulated by humans, but rather handled through computer algorithms. She just added the troll about Trump being an idiot to the other wise valid use of time and tax payer money.
I don't know of what problems you speak of, but if they were unrelated to this hearing it wouldn't be the appropriate time or place to ask.
You're right. Sorry for the oversimplification; I'm pretty out of my depth when it comes to A.I.
While I do think bias can exist even in "traditional" algorithms, as far as I know I don't think Google's search index is composed using Machine Learning algorithms (and if they do then they probably started recently).
However, this doesn't necessarily mean they're biased against conservatives, which is a difficult assertion to prove.
Thanks. I readily acknowledge that I only have to make reasonable assumptions about how Google works. But seriously I'd be fucking amazed if Google's decisions were largely algorithmic. That would in itself pose serious questions.
Eventually instead of the term "idiot" being associated with Trump, they'll simply refer to him as Trump and the idiocy will be assumed by definition...
Well their search engine has been around for 2 decades at this point, though I'm sure it's seen significant change over that period. I suppose it's also a matter of what one means by machine learning. Does any statistical model which acts a predictor based on past data qualify as ML? Probably not.
My only knowledge of their search algorithm comes from University, where we covered it in one class on Linear Algebra. I don't remember much of it tbh.
So you studied the Page algorithm at university? That's very cool. I cannot speak with any authority about how Google runs now, but I'd be frankly flabbergasted if they really just had a "man behind the curtain" model, even if it were an automated man.
But my dad says that Google is part of the Far Left and faking their results to purposefully show bad results first because of the world conspiracy or some shit I don't know, it's hard to pay attention. If I showed him all this, he'd still believe whatever he wants.
The first thing that has to be stated about Google's search algorithms is that it is mostly kept secret, so credible first-hand information on it is hard to come by. They don't want people gaming the system, and several decades of developing the engine has been with the aim of combatting spammers and marketers who manage to game it..
Consider that there are multiple processes in play here, not just serving results. Google has a process to crawl websites, index the information and rate it on 'quality'. They have indicated they use machine learning (trained on human feedback) to understand how to crawl websites. Human bias is undoubtedly part of the way Google functions today. The question is, what happens when these biases are exponentially replicated, on a platform as perfectly robust for that as Google? For the purpose of protecting their cash cow, they wont answer those questions in a transparent way.
What you actually want to do is google just "idiot" and then click on the picture/article about Trump that's served to you among the results.
The algorithm tries to match future searches against successful matches in past searches above all else. You doing that tells it, "When someone's looking for 'idiot', Trump is what they want to find."
It probably wasn't THE algorithm but it was presented as an algorithm which the likes of Google are using. It was more of a motivation for the technical material rather than a detailed step by step breakdown of the algorithm
Well, I agree with Smith that people can influence the google search result rankings. Searching for "Idiot" leads to Trump as one of the most prominent placed results. I'm not saying Trump IS and idiot and I'm not saying Trump is not an idiot. Trump being or not being an idiot doesn't play a role here. My point is that people can indeed influence the ranking. Which apparently shows Trump when searching for "idiot". Because, maybe Trump IS an idiot and maybe Trump isn't an idiot. I for one am pretty sure Trump is an idiot btw but, as I said, that doesn't necessarily imply Trump really is an idiot.
I Googled "orange turd" and still got a bunch of Trump results.
I'm surprised, but you can actually just hack that down to 'turd' and still end up with Trump in the image results.
A delightfully fly-ridden Trump Turd.
I love that she did this because you know only more people are going to notice it XD
Up there with the time the Tory boy with the mishappen cranium moaned about twitter advertising cheap Arab brides
A Democrat asked the question not a republican, and I'm sure she knew exactly what she was doing.
I'm not sure you can say they are influencing the ranking as much as any sports team winning against someone else is influencing the ranking by operating on the collection of winning points and the respective teams that have them that define the ranking itself. The ranking is what the algorithm understands as valuable relational information that exists as content in the internet. Having more content express one certain kind of relation between two items isn't influencing the ranking as much as it is affecting the direct metric with which the ranking is built.
Influence is a bad word to use. It suggest a degree of premeditated manipulation that is not what's happening here. SEO would be much further down that line than this.
If I was the Google CEO this shit would infuriate me. He should have completely told this Republican off and not let him just shut down any rebuttal and last word him like that.
What a bunch of fucking children these republicans are goddamn.
if any single phrase was used to exemplify the current GOP it would have to be "Feelings over Facts". They simply refuse to accept anything that's contrary to their worldview.