Staring down a plughole - AI Shadow Banning

John Sydenham

The Internet has changed so much over the past 15 years. The battle against “populism” and “misinformation” has been an almost complete success. How did they do that? I would ask the reader to decide whether what has happened is desirable.

The key to progress has been the use of AI.

Google and Meta are at the forefront of AI moderation of the Internet. Most of us know vaguely how AI moderation works. The AIs scan all uploaded content and remove “harmful” or “toxic” content. Here is a graphical overview that Ofcom provides (Ofcom is the UK communications regulator):

This scheme looks superficially innocuous. Its success actually depends on the box named “Training Data”. Google or Meta can insert any rules that they want into the “training data” box so that the moderation reflects the views of Google or Meta.

It should be understood that AI moderation is live. It is active across the Internet. It should also be noted that the schematic provided by Ofcom is disinformation. This is what really happens:

The main difference between the sanitised version and the real version is that the AI can be pre-biased by the staff who train it, lobby groups now routinely complain about opponent’s posts, non-USA citizens are largely ignored and, most important of all, Prominence Ranking is performed - otherwise known as “shadow banning”.

Shadow banning is the essence of curating the Internet so that it only gives prominence to “desirable” content. If the AI is primed to consider a post to be of little importance it will downgrade its position in searches. If it considers a website or post to be politically undesirable it will also downgrade its position. This affects the visibility of entire “memes” such as “populism” or articles opposing the views of Google/Meta’s management and staff. The most obvious recent example of this was the near disappearance of the “populist” website “Breitbart” from Google searches. Between 2017 and 2020 any Google searches for a Breitbart article would have yielded results where the article appeared far down the list of search results so that users would need to page through scores of pages before finding the article. The graph below, based on search rankings, shows how Breitbart became nearly invisible for several years:

Source: SYSTRIX

Visibility is crucial for a commercial website. If a source such as SYSTRIX shows that a site cannot be found it will lose most of its advertisers.

Twitter, Facebook, Instagram, Tik-Tok etc. use similar shadow banning techniques.

AI has allowed the moderation of all forms of content. Facebook’s “Rosetta” AI “scans over a billion images and video frames daily across multiple languages in real time”. It can classify images and extract and analyse text from images It is now fully deployed. Rosetta allows Meta to moderate images in Facebook and other Meta content and can even deal with live streams.

Google has similar AI based Video Content Moderation and it is even possible to request a “moderation score” from Google before uploading a video using the “Cloudinary” add-on. Google also has Audio Content Moderation that can screen audio for “toxicity”.

Google provides educational institutions with moderation software. This will flag and screen out pornographic, hateful, populist and other unwanted content.

The Internet is now a curated space where illegal and “undesirable” content is automatically removed or made nearly invisible. The people who set the rules for what content is “undesirable” are the US based Internet barons and their staff. In the past five years the barons have become extremely aggressive and are shaping the Internet so that their views are increasingly visible and opposing views are blocked or made almost invisible.

The West is experiencing the sort of control that is being exercised in China. In China all Internet content must conform to “socialist principles” - must not criticise the regime - and in the West Internet content must conform to the “user safety principles” - must not be illegal or undermine the beliefs of Internet barons or big business.

The Internet is slick and sly. As an example Microsoft Edge now has an intermittent tab for British people to view selected articles on what might be going wrong with Brexit, ask yourself why? There is no mainstream party in the UK favouring a return to the EU under the Copenhagen Criteria which demand acceptance of full political union for new members so why does Microsoft have a tab to persuade the British to rejoin? Watch out, Big Business is shaping the information environment for themselves and turning us into idiots.

We are now at the stage where free thinking people should regard all Internet content, including that of the mainstream media, as potential propaganda and disinformation.

We are only at the beginning of this new form of social control. Soon most people on the Internet will have an AI pal, an Alexa etc. that they carry with them everywhere on their smartphone. It will control your diary, provide you with a friendly voice and illuminate the issues of the day and subtley alter your views. It will be your personalised sales assistant. It will eventually make you intellectually blind and report you for deviation as happens in China.

I really do not know how we can ever put the genie back in the bottle.