Algorithmic Amplification

When you open Instagram, YouTube or Facebook, you do not see posts in the order they were made. A computer program picks and ranks posts for you, based on what it predicts you will watch, like or share. This program is called a recommendation system. When it pushes some content to far more people than it would otherwise reach, we call it algorithmic amplification.

Why do platforms use recommendation systems?

Every minute, people upload huge amounts of content. No one can see all of it. So platforms need a filter that chooses a small number of posts for each person. Platforms earn money mainly from advertisements. The longer you stay on the app, the more ads you see. So most recommendation systems are built to maximise engagement: watch time, likes, comments and shares. A senior YouTube official said in 2018 that more than 70% of watch time on YouTube came from its recommendations.

How does a recommendation system work?

Think of a shopkeeper who remembers everything you have ever bought, and then puts similar items at the front of the shop every time you walk in. A recommendation system works in a similar way, in a few steps:

1. It collects signals. What you watched, how long you watched, what you liked, skipped, shared or searched, and what people similar to you liked.
2. It finds candidates. From millions of posts, it picks a few thousand that might interest you.
3. It scores and ranks them. A machine learning model (a program that learns patterns from past data) predicts how likely you are to engage with each post, and sorts them.
4. It shows the top posts and learns again. Your reaction to these posts becomes new data, and the cycle repeats.

Where did the idea come from?

Early websites showed content in time order. Online shops like Amazon began suggesting "people who bought this also bought" in the late 1990s. Social media feeds moved from time order to ranked feeds in the late 2000s and 2010s. In 2011, the American writer Eli Pariser wrote a book called The Filter Bubble. He argued that personalisation quietly builds a separate, invisible world of information around each person.

Concern grew after events like the 2016 United States election and the Cambridge Analytica data scandal (2018), and after cases where false rumours spread on social media led to violence in several countries, including India.

What problems can amplification cause?

The program does not "know" whether a post is true. It only knows whether people react to it. Content that makes people angry, scared or shocked often gets more reactions. So such content can get extra push. The main problems are:

• Filter bubble: You keep seeing only one kind of view because the system learned that you like it. Other views slowly disappear from your feed.
• Echo chamber: A group of like-minded people keep repeating the same ideas to each other, so those ideas start to feel like the only truth.
• Misinformation and disinformation spread faster. Old videos shown as new events, edited clips and false claims can reach millions before anyone checks them.
• Polarisation: Society splits into groups that see completely different "facts" and distrust each other.
• Mental health and addiction: Endless feeds and autoplay are designed to keep you scrolling, which affects children and teenagers the most.

How does India regulate platforms today?

India has no separate law only for algorithms. Platforms are governed mainly by the Information Technology Act, 2000 and the IT (Intermediary Guidelines and Digital Media Ethics Code) Rules, 2021:

• The IT Rules, 2021 were notified on 25 February 2021. They list the kinds of content a platform must tell users not to post (Rule 3(1)(b)), such as content that threatens public order or is knowingly false.
• A significant social media intermediary (SSMI) is a platform with at least 50 lakh (5 million) registered users in India. It must appoint a Chief Compliance Officer, a Nodal Contact Person and a Resident Grievance Officer, all living in India, and publish a monthly compliance report.
• Grievance Appellate Committees were added by a 2022 amendment, so users can appeal against a platform's decision on their complaint.
• Amendments notified on 10 February 2026 (in force from 20 February 2026) made platforms label synthetically generated information (AI-made or AI-altered content that looks real), and cut the time to remove unlawful content after a court order or authorised government notice to three hours.
• If a platform does not follow these due diligence rules, it can lose its safe harbour protection under Section 79 of the IT Act. That means it can be held legally responsible for what its users post.

What is an "advisory", legally?

An advisory is a letter from a ministry explaining what it expects platforms to do under existing law. It is not a new law passed by Parliament, and it is not a formal rule notified in the Gazette. For example, MeitY issued an advisory on artificial intelligence models in March 2024 and revised it soon after.

Advisories often carry weight because they remind platforms of duties that, if ignored, could cost them safe harbour. But critics argue that they can be used to give orders without the checks that a law or rule must pass.

How do other countries handle algorithms?

The European Union's Digital Services Act (DSA) is the leading example:

• Article 27: All online platforms must explain, in plain language, the main factors their recommendation systems use and any options users have to change them.
• Article 38: Very large online platforms (those with 45 million or more monthly users in the EU) must offer at least one feed option that is not based on profiling (not based on your personal data), such as a simple time-order feed.
• Very large platforms must also assess "systemic risks" every year, such as effects on elections and public safety, and allow independent audits and researcher access to data.

Commonly confused concepts

• Content moderation vs algorithmic amplification: Moderation is about removing or keeping a post. Amplification is about how widely a post is pushed. A post can be legal and still be amplified in a harmful way.
• Misinformation vs disinformation: Misinformation is false information shared without bad intention. Disinformation is false information spread on purpose to mislead.
• Filter bubble vs echo chamber: A filter bubble is created mainly by the algorithm choosing for you. An echo chamber is created mainly by people choosing to talk only with like-minded people.
• Takedown vs down-ranking: A takedown removes a post completely. Down-ranking (also called demotion) keeps the post online but shows it to fewer people. Down-ranking is harder to see and harder to challenge.
• Section 69A vs Section 79: Section 69A of the IT Act lets the government order blocking of specific content on listed grounds, with reasons recorded in writing. Section 79 gives platforms protection from liability for user content, if they follow due diligence.

Issues, criticism and the way forward

• Free speech concerns: Under Article 19(1)(a), citizens have freedom of speech, and limits must fit the grounds in Article 19(2). Critics say vague demands like "limited set of narratives" can push platforms to quietly down-rank lawful political speech, which is hard to challenge in court.
• Who decides what is "credible"? If the state tells platforms which views count as credible, it may affect the free flow of ideas. If platforms decide alone, private companies hold great power over public debate.
• Lack of transparency: Platforms rarely explain how their ranking works, so neither users nor regulators can easily check for bias.
• Real harm from viral falsehoods: The government's position is that rapidly spreading false or edited content can trigger riots and violence, so platforms must act fast and cooperate with law enforcement.
• Way forward suggested by experts: clear rules in a law passed by Parliament (the proposed Digital India Act is meant to replace the IT Act, 2000), published reasons for every government request, independent audits of algorithms, a non-personalised feed option for users, regular transparency reports, and better digital literacy so that users can spot fake content themselves.

Concepts to Know

• Algorithm: A fixed set of steps a computer follows to solve a problem, like a recipe. In social media, the algorithm decides which posts you see.
• Engagement: How much users interact with a post: views, watch time, likes, comments and shares.
• Intermediary: A company that carries or hosts other people's content without creating it, such as social media sites, search engines and messaging apps.
• Due diligence: The basic care and checks a platform must follow under the IT Rules to keep its legal protection.
• Profiling: Using a person's data (age, location, likes, history) to predict and decide what they want.
• Systemic risk: A danger that affects society as a whole, such as harm to elections, public health or public safety, not just one user.