I think they are being genuine when they say their plan is to combat fake news with AI; their business is built around AI. It is one of their core competencies.
I think they are going to fail, and quietly increase the human involvement once they realize that their AI isn't good enough. Then they will slowly scale back the humans as their AI improves, until the next controversy when they realize that they still need humans.
> I think they are going to fail, and quietly increase the human involvement once they realize that their AI isn't good enough. Then they will slowly scale back the humans as their AI improves, until the next controversy when they realize that they still need humans.
Exactly. AI is not good at subjective decisions of qualitative data. For example, nobody knows any political candidate's net worth apart from IRS, until they make those records public. And say political candidates make statements that they are way less or way more than their net worth they cannot detect it is true or false.
That's why I was surprised. Pichai is smart enough that AI can't combat fake news. Hence he was just saying to save face.
"Fake news" has lost all meaning in the intervening years, but at the time, it is hard to remember, we were concerned about essentially phishing websites that made themselves out to be (say) CNN, but were not.
As others have said, evaluating trust on the internet using computers (call it AI or not) is literally Google's core competency.
Realistically there doesn't exist a good solution for this. That's why every single site out there with user content that is large enough is struggling with moderation.
There is no solution that scales up to billions of users, and while it's true that AI most likely won't work, it is the best they've got right now. Do you have a better solution? Because Google has hired some of the smartest people and even then they still are having issues with Youtube every other week, so I'm sure you'd be paid a pretty hefty sum if you could solve this.
Let the receiver evaluate the information? Sure, some will have false believes. Cannot change that and you shouldn't even attempt to do so. Because authority has been wrong to an equal degree.
The whole premise falls flat in my opinion.
There won't be any viable solution, if the problem cannot even be defined. Currently it is based on a feeling that there are mean and false statements on social media.
And this fact is emphasized by parties who like more control about content.
Anyone aware of the current capabilities of AI should know, that it is no solution, at least in its current state. Sure, big tech likes to signal otherwise. But these statements just underline their business interest in that sector. Nothing wrong with that.
This ignores one of Google's primary missions. As jlebar posted in a different comment
> evaluating trust on the internet using computers (call it AI or not) is literally Google's core competency.
This idea that we should pretend as if there is no problem to solve seems incredibly strange to me. Particularly when we know unreliable information is being weaponized by bad actors to influence entire populations of people. This is a very real problem which needs solving.
Humans have limited time for research and we value reliability. Consider all of the hype around the recent severe decline in trust of authenticity and reliability of products purchased through Amazon. We tend to prefer shopping at places where we know the products we purchase will be of a certain standard. We go to trusted sources (as amazon was previously) because we don't have the time to carefully evaluate each of the many many items we purchase every week. We choose a store we trust will have already done that quality check.
It seems as if Google believes it is important for it's business that the information it prioritizes has a certain quality or reliability level. Google prefers their information be more like a Target store than a back alley tent. And if the amount of people who shop at Target stores over back alley tents is any indicator, I'm guessing Google is probably on to something.
I'm certainly skeptical about who should be able to decide what is "true" on subjective issues, but I don't see how we can pretend as if there isn't a very real problem of actual provably false information being weaponized.
I for one am fairly busy and just as I don't have time to do a deep dive of research into whether or not the shampoo, deodorant, milk, cereal, orange juice, whiskey, dish soap, cold medicine etc etc etc are fake; I also don't have time to deep dive research if the 30+ articles I skim everyday are provable outright lies. Not only don't I have the time, I don't necessarily have the inclination.
> We tend to prefer shopping at places where we know the products we purchase will be of a certain standard.
we don't just prefer; we pay to shop at places where we trust that the products will be of high quality (or at least authentic).
no one i know around my age pays for any news whatsoever (including myself, admittedly). this should tell you something about how much we value quality news.
>AI is not good at subjective decisions of qualitative data
That's exactly the core strength of AI. It is what differentiates AI from hard coded solutions. Estimations(subjective decisions) based on correlations in fuzzy (qualitative) data.
In your example, you could take a set of data containing the net worths and other characteristics, e.g. birth zip code, spending habits, affiliations, and if there are any trends related to net worth, a properly architected neural network trained on the appropriate data could easily estimate subjectively on what amounts to qualitative data.
Ah, well, not to put too fine a point on it, but search engines don't even come close to rating every new page on the internet in real time.
There are like days of lag, in search engine results, unless you eagerly, eagerly, eagerly force your way into strategic points of the existing index for each independent search engine, separately.
Nevermind DNS replication, which is it's own beast. So, from domain registration, to DNS replication, to getting listed, to becoming relevant (which requires all kinds of meta tags and dom restructuring for crawlability, plus roll-your-own-site-maps, and so on), to being recognized as a reliable source of information, such as press releases, before we even get to news possibly going viral?
Well, news actually doesn't even go viral without humans in the loop. I mean consider how viral a high karma rank gets you on hacker news? It's just enough to get 10K eyes on a site in unison, to overwhelm a non-load-balanced server. And that's the voting of users doing all the work.
And then to include heavier social sites like reddit, twitter and facebook in the mix? That's almost purely human opinion performing the ranking factor. The attention of the users closes the feedback loop, and the chain reaction can strap a booster rocket to notoriety.
So, I'd almost say the really real time (like viral real time) stuff is almost only humans doing the work, and the robots are just there for the chatter threshold tripwires.
> Nevermind DNS replication, which is it's own beast.
There’s a lot wrong with your post, but let’s start here. How do you think DNS works? Your statements about it indicate a complete lack of knowledge on the topic. I can register a new domain and have it resolveable anywhere in the world in minutes.
How did we get to the state where the "truth" is such an elusive concept?
Is it so hard to determine whether basic statements are true or false? And to build larger, higher constructs out of those building blocks? That's basically what science has been and is.
It seems comically easy to identify fake news in most cases. Was this inauguration crowd larger than that one? That's a simple question to answer.
We got to that state when we started talking about so much information that it is impracticable to have a human read all of it, let alone provide a truth judgement for all of it. Even getting computers to read all of it is a significant engineering effort.
Simply getting computers to understand the statements is one of the holy grails of AI research; let alone determining if they are true.
> How did we get to the state where the "truth" is such an elusive concept?
Look into the fields of epistemology and the philosophy of science, in particular the works of Karl Popper such as The Logic of Scientific Discovery. The "truth" has always been an elusive concept. Outside the realm of mathematics, it's rather difficult to objectively prove most things "true".
Two days ago, partisan advocacy site ThinkProgress ran an article with the headline "Brett Kavanaught said he would kill Roe v. Wade last week and almost no-one noticed". The reason no-one noticed is because he said nothing of the sort and ThinkProgess knew as much. One of the fact checkers partnered with Facebook labelled it as false, they attached the usual warning, and this made ThinkProgress and a chunk of the left-leaning press furious. They accused Facebook of "defaming" and "censoring" them to "appease the right wing", claiming that this was actually some kind of partisan attack on the truth.
That, in essence, is how we got here. There are plenty of loud, vocal defenders of "truth" out there, it's just that the loudest and most vocal of them define "truth" to mean "what our side believes".
Why would you expect the truth of basic statements to be easy to establish? Science is hard and many basic questions are debated by experts. Political consensus is difficult.
Science is not a public endeavor. Science is conducted by trained professionals. Besides the training, there exist filters (self selection, academic tests, etc) which ensure the scientific community has a higher than average connection with "the truth".
The public at large has never been particularly well connected to truth. One difference now is that, in the past, the public at least respected and trusted scientists, academics, etc. Nowadays they're scorned.
I think humility of knowing how little any single individual knows, and how much "details" matter, would be much better attitude for everyone to take, rather than thinking the truth is easy, even for seemingly basic stuff.
I think they are going to fail, and quietly increase the human involvement once they realize that their AI isn't good enough. Then they will slowly scale back the humans as their AI improves, until the next controversy when they realize that they still need humans.