How to moderate a global megaphone? Why the EU’s new internet safety law will be a challenge
Social media has allowed us to connect with each other like never before. But it came at a price – he handed out a megaphone to everyone, including terrorists, pedophiles and hate groups. The institutions of the European Union recently reached an agreement on the Digital Services Act, which intended to “make sure that what is illegal offline is treated as illegal online”.
The UK government also has an Online Safety Bill in the works, aimed at strengthening requirements for digital platforms to remove illegal material.
The scale at which major social media platforms operate – they can have billions of users around the world – presents a major challenge in monitoring illegal content. What is illegal in one country may be legal and protected expression in another. For example, rules regarding criticism of the government or members of a royal family.
This gets complicated when a user posts from one country and the post is shared and viewed in other countries. In the UK there have even been instances where it was legal to print something on the front page of a newspaper in Scotland, but not in England.
The Digital Services Act leaves it up to member states of the European Union to define illegal content in their own laws.
Illegal Content Tracking
Even where the law is clear, such as someone selling controlled drugs or recruiting for banned terrorist groups, content moderation on social media platforms faces challenges of scale.
Users make hundreds of millions of posts a day. Automation can detect known illegal content based on a fuzzy fingerprint of file content. But it doesn’t work without a database and content needs to be reviewed before being added.
In 2021, the Internet Watch Foundation investigated more reports than in its first 15 years of existence, including 252,000 containing cases of child abuse: a 64% year-over-year increase from 2020.
However, new videos and images will not be captured by a database. While artificial intelligence may try to find new content, it won’t always get it right.
At the start of 2020, Facebook had around 15,000 content moderators in the US, up from 4,500 in 2017. TikTok claimed to have 10,000 people working on “trust and safety” (which is a bit broader than content moderation ), from the end of 2020.
A 2020 NYU Stern School of Business report suggested that Twitter had around 1,500 moderators.
Facebook claims that in 2021, 97% of the content they flagged as hate speech was removed by artificial intelligence, but we don’t know what was missed, unreported or unremoved.
The Digital Services Act will force the biggest social networks to open their data and information to independent researchers, which should increase transparency.
Need human review
Reviewing violent, disturbing, racist and hateful content can be traumatic for moderators and has led to a $52 million legal settlement. Some social media moderators report having to review up to 8,000 reported pieces of content per day.
While there are emerging AI-based techniques that attempt to detect specific types of content, AI-based tools struggle to distinguish between illegal and distasteful or potentially harmful content. (but otherwise legal).
Artificial intelligence can falsely flag harmless content, miss harmful content, and increase the need for human review.
Facebook’s own internal studies reportedly found instances where the wrong action was taken against posts up to “90% of the time”. Users expect consistency, but it’s hard to deliver on a large scale, and moderators’ decisions are subjective. Gray area cases will thwart even the most specific and prescriptive guidelines.
Freedom of expression, fake content
The challenge also extends to misinformation. There is a fine line between protecting free speech and freedom of the press and preventing the deliberate dissemination of false content. The same facts can often be phrased differently, which is well known to anyone familiar with the long history of “spin” in politics.
Social networks often rely on users reporting harmful or illegal content, and the Digital Services Act aims to reinforce this. But an overly automated approach to moderation can flag or even hide content that reaches a set number of reports. This means that groups of users who wish to remove content or views can weaponize mass reporting of content.
Social media companies focus on user growth and time spent on the platform. As long as the abuse isn’t holding them back, they’ll likely make more money. That’s why it’s important for platforms to take strategic (but potentially polarizing) action – like the impeachment of former US President Donald Trump. from Twitter.
Most of the demands made by the Digital Services Act are reasonable on their own, but will be difficult to implement on a large scale. Increased monitoring of content will lead to increased use of automation, which cannot make subjective assessments of context. Calls may be too slow to provide meaningful recourse if a user mistakenly receives an automatic ban.
If the legal penalties for content moderation errors are high enough for social networks, they may have no short-term choice but to more carefully limit what users get. TikTok’s approach to hand-picked content has been widely criticized.
Platform biases and “filter bubbles” are a real concern. Filter bubbles are created where the content presented to you is automatically selected by an algorithm, which attempts to guess what you want to see next, based on data such as what you have already viewed. Users sometimes accuse social media companies of platform bias or unfair moderation.
Is there a way to moderate a global megaphone? I would say the evidence says no, at least not on a large scale. We will likely see the answer play out through the enforcement of the Digital Services Act in court.
Greig Paul is a Senior Mobile Network and Security Engineer at the University of Strathclyde, Scotland.
This article first appeared on The Conversation.