Google says AI generated content is against guidelines
searchenginejournal.com
searchenginejournal.com
Fast forward 10+ years and for knowledge-related queries search is going to be more about generated results personalized to our level of understanding that at best quote pages, and more likely just reference them in footnotes as primary inputs.
These knowledge-related queries are where most content farms, low quality blogs, and even many news sites get traffic from today. If the balance of power between offense (generating AI content) and defense (detecting AI content) continues to favor offense, there will be a strong incentive to just throw the whole thing out and go all-in on generated results.
Big question is how incentives play out for the people gathering the knowledge about the world, which is the basis for generated results. Right now many/most make money with advertising, but so do content farms, and more generated results means more starving of that revenue source. For a portion of info that people want to know (e.g. factual info, not opinions, guidance, etc), Wikipedia is an alternative fact- and context-gathering model, but if search relies on it more, it will strain Wikipedia's governance model and become more of a single point of failure.
Really interesting stuff ahead.
What I'm trying to get at is that ML is currently terrible at contextualizing information, but in the future the successful knowledge-query engine will be the one that do the best job of wading through the explosion of digital content and pulling out the parts that are coherent and more connected to reality.
This program will need to be able to contextualize information to form a coherent model where knowledge is interconnected into a larger model of reality.
Easier said than done, we'll see if that is even possible. This is getting towards AGI.
I think it's important to distinguish between ML-assisted and whole cloth ML generation too. I imagine many high quality content sources are going to be using ML-assisted writing tools; for instance, at my old news company (edsurge) the journalists spent most of their time gathering info and constructing the story, which was the real meat of the work. The writing and wordsmithing was the not operating at the top of their skillset, it was pretty inefficient manual labor, especially the drafts. So from content production perspective, ML could automate a lot of that away while letting the journalists tune the narrative and edit the output for better nuance - in other words, editing vs producing.
This also assumes a single output, but if you take the edtech lens and couple it with the power of ML and you could easily have any given news article have nearly infinite varieties based on reading level, language, and even past knowledge (by including or removing context). As someone interested in new mediums and innovating news, this is absolutely exciting to me. A really interesting question here is if there's an opportunity for collaboration between expert sources and search engines, such that for a given set of domain-specific queries the expert sources are in charge of fine-tuning the generated outputs.
Completely agree about contextualization. The automated detection of ML (the defense part) needs to be tuned toward trying to extract and analyze claims that are being made in the content, and a lot of ML content is nonsensical, or introduces wild, novel claims that lack evidence and alignment with other sources. Some big questions for the analysis are whether those claims are corroborated by other higher quality sources, whether relevant context is included, and whether its bringing anything new to the table.
In this kind of setup redundant sites will be ruthlessly demonetized by search. As you say, this also drives us in a direction toward expert systems with some probabilistic measurement of the truth of claims and completeness of context, which may also have actual experts at the controls of some of this (like above). This of course raises a lot of questions where unorthodox ideas may fit in this model.
I'm not an ML nor search expert though, my experience is more on production side.
Oh. Oh dear.
In the arms race between ML-copy generation and its detection, eventually the edge becomes "does this make sense in a coherent world-view", so there should be a natural pressure towards that outcome.
But maybe we're just not good enough at building these tools, and the incentives will align with an endless onslaught of digital sludge.
There's also the problem of: how does legitimate fiction, poetry, love stories, and art fit into a www where only the most logical is allowed to be seen? It could be a logically puritanical nightmare.
Why can't the Wikipedia model be adapted to a federated, directly community run approach? This works well enough for services such as email, matrix, and the fediverse. There's gravitation towards centralized hosting services but that's largely behavioral - the model itself works perfectly fine with lost of small players.
Heavyweight multimedia can be a challenge but text content itself is quite easy to serve up from very small devices.
EDIT: Here's an example from the EU, which has their complete budget in a triple store: https://query.linkedopendata.eu/#SELECT%20%3Fproject%20%3Fpr...
(you have to click the blue "play" button to run the query.)
I'm not sure if they are federating with wikidata or just importing the data, but the result is similar either way.
(For budgets, always go for the treemap: https://query.linkedopendata.eu/#%23defaultView%3ATreeMap%0A...)
I could see several nonprofits and news brands along with Wikipedia shift to becoming a set of sources of truth for different & likely overlapping topics. That shift could happen gradually, as part of a mix of monetization incentive changes and more explicit 'here's how you participate' coersion (medium is message stuff; see FB's pivot to video, or youtube algo changing how content creators create). The generated result that Google spits out could reference those and note them as inputs, including noting where they disagree or choose to include or exclude certain context.
I still don't see how these ideas get funded without Google directly funding them, where algorithm transparency comes in play, etc.
Maybe Linux development is a good example, with some centralisation but other power centers of varying size connected, such as distributions or non-kernel software projects.
But then again Wikipedia already is federated into hundreds of local communities and horizontal projects like commons or wikidata, and it works not quite as terrible as one would think.
this does not seem to be a model that will help make people smarter over time.
As an edtech person that started a news company to try to solve this, one of my pet peeves about news is that it doesn't adopt an education mindset / approach despite it's mission being an educational one at its core. Part of that is due to search being what drives traffic, and that means there's a lack of relationship to the reader. But the root is internal as news and journalism is still too stuck in newspaper-mind of bespoke, reverse pyramid one-off articles that act as islands of information, delivered to an audience they have no ongoing relationship with and who have no means of getting follow-up. It's a cultural holdover that's bound up in relationship to it's old medium, not what optimally teaches.
Generated results need not be in the form of articles, that's one of the constraints that ML lifts. You could give people some generated text but also give people drill-down tools, letting people expand on a simple summary, or read in dialectic or debate style in areas where facts are not settled or multiple viewpoints exist. Just gotta expand your POV a bit and ML generated results represent an opportunity for incredible leverage educating the world.
Anyway if someone in ML at OpenAI/Google/etc wants to get deeper into this, please reach out!
Nowadays I'm almost happy if I find at least one sentence of useful information in an entire "article" recommended by Google.
I think the above game example is an example of erosion of both sides: articles have gone to shit, and the amount of care, attention and depth to most games have gone down a lot.
If you run an older internet reference, please add https. The search engines also have an obvious 'recent' bias, but everything we can do to prevent the loss of accessibility of older reference material helps.
true. but then again, in the context of accessibility and compatibility, ssl is a hindrence.
next thing google decides that sites without ads are to be downranked ... fuck this
Probably already happened, they just didn't make a public announcement.
Generative models will asymptotically approach the same statistical distribution of real data.
Eventually, content farms will make actual websites... and nobody will know the difference.
The rise of social media has moved most human generated content onto a few platforms, most of whom don't let Google index them (facebook being the main one).
They probably need to figure out how to pull information out of youtube (which there seem to be some initial efforts at).
However they might already be in trouble there, as TikTok is growing a huge archive of 20 second answers to common queries.
However, they seem to be leaving open the possibility of AI-assisted writing, where a human comes up with the information and guides the AI as it puts that information into words.
> From our recommendation we still see it as automatically generated content. I think over time maybe this is something that will evolve in that it will become more of a tool for people. Kind of like you would use machine translation as a basis for creating a translated version of a website, but you still work through it manually.
> And maybe over time these AI tools will evolve in that direction that you use them to be more efficient in your writing or to make sure that you’re writing in a proper way like the spelling and the grammar checking tools, which are also based on machine learning. But I don’t know what the future brings there.
In my opinion GPT-3 has already reached the point of being useful for this purpose - there are several GPT-3 based apps that do exactly what he's describing.
I guessed it was used for filler content for SEO sites.
A question is whether that company would save money by using "AI" text generation, when they were paying real humans so little, for arguably higher quality.
You can power a GPU for a few cents of electricity per day, so I think the answer is a resounding yes.
https://you.com/search?q=how%20to%20write%20well&fromSearchB...
Also, it would have worked perfectly for my girl's homework assignment, but she has morals and decided she couldn't ethically submit it. She did use the output for ideas on what to write, though.
It didn’t seem to know any musical artists or songs. It was weird. When I asked for song suggestions the best it could do is write bad poetry.
Note that, to win, the machines just need to produce approved content at a faster rate than the humans. If some AI can spew text at a measly 1GB/s, and Google blocks 99% of it, 10MB/s will still getting through. In 2016, there were ~4.6B pages on the internet. If each page contains 10KB of text, then the AI in my example produces 1000 pages per second. There are 31 million seconds per year, which means it would churn out 31B pages per year, or the entire 2016 internet every few months.
My conclusion? Internet search is screwed. Maybe people will start paying for curation.
I'll be taking a front-seat row watching that show, if it happens. Perhaps in the future, we'll have to deal with discriminators that approximate some originality-index. Will be fun fighting with those algorithms, to interact with the internet as a normal user (to some extent we already do - proving that you're human is becoming more and more tedious.)
Google isn't yet effective at detecting blogspam generated by naive scripts that simply swap words in the source material for other words in a thesaurus. I don't think they're going to start picking up continuity errors, factual errors or "weirdness" in GPT-3 - which often satisfies human readers - any time soon.
The other reason is I started using a search engine that cracked the code to let users rank and block sites (kagi) so I don't have to worry about it.
The reason those results remain in your searches is because sending people to those sites makes Google money, and by definition that is of greater value to Google than the quality and relevance of your search results as a non-paying user.
I recently did this, and ended up in a warren of sites like "10 best Widget repair in Smallsville, MO", where all the addresses were for nonexistent businesses in other states, with sort of plausible names tangentially related to widgets. (High plains frobulator sales, Mesa, AZ, or whatever).
I'm pretty sure a deep learning algorithm produced that corner of the internet. It definitely contained lots of referral links and phone numbers. Never managed to reach a human or figure out what the scam was.
I put that in quotes as I've entered numbers I know and find that what's being presented is apparently randomly-generated IDs.
The overt use is checking when unknown numbers call (or going through call history) and trying to get a sense of who the number is associated with.
The general location is usually plausible (so the generators are relying on local exchange numbers), but the specific street addresses and subscriber names appear generated, and specific street addresses may not actually exist.
By the same criterion we may need to consider the Wattage cost of intelligence when granting rights to AI. Their kernels needs to be aligned with evolutionary wisdom encoded in our highest motivations.
When the percentage of human generated content approaches 0.0000%, what does that internet look like?
And the result of this? Genuine conversation progressively retreats from the internet at large as it becomes overrun with the intellectual flatus of the bot wars, and people move further and further into silos in which bot behavior can be spotted and eliminated.
Welcome to the future. How do you like it, gentlemen? All your posts are belong to us.
I have seen a couple of times on some other even smaller technical forums what appeared to be sequences of comments generated by an AI, from a fake account (until the accounts were banned).
There could not have been any kind of financial gain from those actions, so the conclusion was that someone was testing the capability of their AI to carry a conversation on such subjects.
The messages posted were remarkably coherent, but in the end it was obvious that their source could have been only either an AI or a human with mental problems.
What made those messages stand-out was that even if they contained mostly correct information, the same that might be found e.g in technical magazines, there was always some weird disconnect between their answer and the message to which they replied.
Apparently, in most cases the AI failed to identify which exactly was the point-of-view of the previous poster. The reply referred to various items mentioned in the parent message, but instead of referring to the essential points it referred to some non-relevant things mentioned more or less accidentally, or it tried to argue for or against some points as if trying to contradict the previous poster, when they actually had argued in the same direction.
I could definitely see someone already having trained a bot to write HN comments and posting them.
What's anyone going to do about it? It's super hard to write a discriminator that works well enough to not destroy the site for everyone.
Yes, which will produce things that are stylistically similar to HN comments, but without any connection to external reality beyond the training data and prompt.
That might provide believable comments, but not things likely to be treated as high-quality ones, and not virtual posters that respond well to things like moderation warnings from dang.
Replies to comments are difficult, so I guess the bot must just write somewhat related top level comments.
It will accumulate a lot of "Did you RTFA?" and also polite request for clarifications, and it will be suspicious that the bot never replies or replies with nonsense.
If every commit has a link to http://www.example.com then people will notice and start flagging.
I guess it can fly under the radar posting a few top level bland comments per week in the megathreads with a few hundred comments, but I don't imagine how this can be used to gain anything.
One could argue it's already happening. How many of the people we talk to everyday get their facts from SEO-spam websites and Google instant answers (which often sources its content from such websites)? Even if we avoid AI-generated content, we might be gettting fed it by proxy.
I think the world will be a better place if everybody follows a similar pattern. The only reason to use deepfakes is if the victim who's identity is being stolen is not cooperating with you. - It's a new way to violate a person's integrity and their right to agency in our already fading grasp of reality. You could probably gaslight your girlfriend with it, if you are incomprehensibly evil.
I'm honestly a bit more concerned with people gaslighting courts.
The technology isn't going away, unfortunately. Society will have to adapt to these new invasive norms, as they already have time and again in the past.
In the process of remembering that little tidibt it reminded me of a short story I read once that was circulated without attribution for a few decades, i.e. Terry Bissom's "They're Made Out of Meat"[1] published in OMNI in 1990. "They're meat all the way through."
[1] https://web.archive.org/web/20190501130711/http://www.terryb...
Much of the internet is currently full of not-even-approximately true (and often maliciously false) information, so I’m not particular worried about that.
> When the percentage of human generated content approaches 0.0000%, what does that internet look like?
...approximately the same, but less false :)
This is a lie, i sw
<NUL>
<NUL>
Segmentation fault
(Another risk might be governments-enforced identification, no more pseudonymity !)
Google shouldn't be the arbiter of what's ok or not on the internet. They use AI to take away all human recourse with the company but want to tell others not to use it? It's a pretty laughable position. Good luck trying to detect GPT-3 and the like when you compare with non native speakers of languages. Are you suddenly going to just block them too?
If an AI can generate high quality content, why is that any less than human generated content. Human generate a ton of trash content, it's not inherently better.
Those same models on copilot are generating useful and often good code for me on a daily basis. If someone told me it's not OK to use any copilot generated code as unethical/wrong I'd laugh in their face. It's basically saving me a google search to find snippets/examples of things I wasn't sure of.
Maybe that's the threat? If we have access to AI directly (ala copilot) then I am googling less.
I'm not concerned with ethics but both Google-search and co-pilot are some version of processed human generated content and I think problem both have is how to get the content for cheap and still sell processed version. Google pays for it's stuff by advertising and . Co-pilot just munged the contents of github and didn't pay anything. But if AI-generators continue the path of scanning free content and selling the processed result, someone will start working on how to spam them as well - there's a project for you, upload a whole bunch of general purpose code to github with spam in the comments.
I can't wait to see how they will manage false positives.
Now not only you can't contact a human, you can't prove you're one.
Realistically, I don't think I will ever start regularly failing it. Go ahead, call it arrogance or hubris or whatever. I've looked at the originality and predictability metrics on a large corpus of my online comments. Let's just say, you won't be replacing me with gpt-3 or similar.
Not to mention that in cases where it results in a direct ban, you better hope that odds are better than 50/50 ! (New users are particularly susceptible.)
Of course, this suggests a corollary: that Google shouldn’t be indexing the output of content farms where the writing and derivative-ness is strictly worse than what you’d get from an AI article generator. :)
... Let's start with some demonstrations of good AI-generated content being down-sampled by this policy.
Mission. Fucking. Accomplished.
The fact that Google uses AI is irrelevant.
Like all those countless examples where someone got a picture of a criminal mashed up with their own Wikipedia page because they happen to have the same family name. [1] Disregarding the shame factor of this, it’s showcasing low quality auto generated content created only to put the google name above others in search results.
Same applies to those auto generated opening hours of restaurants etc, which are incorrect all of the time. Again low quality ai content put above others and bypassing page rank.
Pseudo biblical language to provide an illusion of authority is stomach turning.
> And people have been automatically generating content in lots of different ways. And for us, if you’re using machine learning tools to generate your content, it’s essentially the same as if you’re just shuffling words around, or looking up synonyms, or doing the translation tricks that people used to do. Those kind of things.
> My suspicion is maybe the quality of content is a little bit better than the really old school tools, but for us it’s still automatically generated content, and that means for us it’s still against the Webmaster Guidelines. So we would consider that to be spam."
So are people reacting thinking this is a new policy or...?
Both are spam trash I never want to waste my time reading.
Google should aim for higher standards.
In the arms race between Google and spammers, honest websites sometimes get caught in the crossfire. For some reason lately lots of people want to blame google for this and not the spammers.
Why couldn't it be that reasonably both sides are to blame. Google enables the commodification of search results. Yes they claim that they want to show the best result. But best for whom? I don't believe that Google is a neutral party as they earn more if they promote sites that lead to additional advertising clicks. How should users know that there are often better results on pages 3 or 4 and below.
And than there are the ones trying to make a living with minimal effort. No need to create great content. Good enough is sufficient. As long as the content is optimized and the page receives relevant traffic from search the revenue via affiliate links is secured.
In this arm's race the other sites loose. They are the collateral between two fighting parties in this arms race. And also people loose. Loose great content and the diversity of the net.
I don't buy into the position that all this great content and diversity of content is lost because of Googles algorithm. It isn't as discoverable as the mainstream content that Google search returns but that's no different than being in a world where Google didn't exist. Except for perhaps in such a world people would have different content discovery habits.
Complaining that Google doesn't surface your preferred set of "great content" is like complaining that prime time cable TV only shows lame sitcom reruns. It's true, but it doesn't prevent you from buying HBO, or Prime Video.
They can't look for what they don't even know exists.
I know how to surface these sites and enjoy them. But I probably belong to a tiny minority of users within the internet regarding this knowledge.
I know that basically always there have been the types of people who were better equipped for knowledge discovery. And also that knowledge was not free. To learn was time consuming and expensive. And only in recent decades this has changed.
Still the tools as well as the fact of additional stuff outside one's current knowledge needs imho to be distributed more broadly.
> Thankfully Google operates in a free market where it is possible to compete by building a better search algorithm.
That is a sentiment I don't share. With market shares > 90 percent and reighn of Android the network effects are very strong in my opinion. Making it at least a few orders of magnitude more difficult than when Google entered the game. So while it might be possible, it is economically probably not a great bet to make.
Even then... A lot of financial news is written algorithmically. A peer in my Comp Ling masters program started a business that does just that, and sold it or consulted on it's use with Bloomberg, or maybe Forbes? It was more than a decade ago. That stuff was mostly mixed methods of rules & stochastic, so I'm sure it's come much further. But it's a valuable service because so much of this news really is very formulaic: "Stock $X came in at $Y earnings for quarter $Z, beating analyst target earnings of $A. This is in line with the general rally in the $B sector, following similar estimate-beating earning in stocks $C, $D, and $E."
Dozens of these a day (or more) could certainly look like a content farm.
Might be funny to see the dissonance when people begin to make decent art or memes with AIs, and Google will have to either drop the site, or ignore its own rules.
That presents quite a threat profile. It's far more pernicious than SEO script kiddles doing whatever passes for keyword stuffing in 2022.
I hope the search crawler folks at Google are working hard to detect that sort of thing and prevent it from getting into indexes. Let's hope Neal Stephenson isn't as right about that threat as Arthur C. Clarke was about geosynchronous communications satellites.
otherwise some dumb people will want ai generated content to be allowed everywhere
i'm pretty sure they are doing their move now because it causes them ton of issues with YouTube
allowing ai generated content, would mean allowing ai generated comments on youtube, wich is already happening and causes lot of issues
if you can't tell what is AI generated and use comments/discussions/like/dislike in your algorithms for ranking videos, then it'll be very easy for 3rd parties to push and play the game, including ad revenue
the inevitable will come sooner rather than later, get ready for your online passport!
And infoboxes are usually the first result in any Google search, before the ads even.
I might be wrong, but I think this might be a more difficult task than generating convincing dialogue: its very easy to generate text that statistically resembles human writing, and generators trained on certain topics (i.e. some science niche) might be impossible to flag.
Sounds pretty evil to me.
So by their own definition they are producing spam?
The appeal of open-to-all search as a way of navigating the web was that there was a huge long tail of interesting stuff that would be hard to manually index and categorize.
If the long tail of interesting stuff has fully drowned in a sea of spam and crap, I'm not sure that it still makes that much sense over something smaller but human-curated.
Perhaps the trick would be human curation with extensive and always-evolving AI tools to speed up the curation. You have to get past the filter to get in, versus being in by default unless you're blatant enough to get banned. There is a layer of human judgement in addition to the algorithm's score of the content, and additionally that gives you an extra scoring factor on the algorithms yourself - the humans should be able to help direct the development of the algorithm to fight spam more preemptively.
Would a mix of that give us a bigger internet than the entirely-manual "web directory" days of 1997 or so, but a less-shit-filled one than today's?
And if I can peddle my blog’s ‘content’: here’s Trump announcing he’s (not?) trans.
> I have been so concerned about this issue, I've held back from telling you that I'm transgender. I'm not transgender, but I'm so proud of the transgender community and their rights.
> The Democrats have been so horrible to the transgender community. They've made them live in these little closets, ya' know? And they've tried to force them to use certain restrooms.
Love messin’ with gpt-3.
It's certainly good idea to penalize google with intelligent AI-generating content