Algorithms: The Technology Deciding What You See Online

Every time you open TikTok, Instagram, or YouTube, you see something specifically chosen for you. Not randomly. Not chronologically. Specifically chosen by an algorithm that studied millions of pieces of information about you and predicted what would keep you scrolling. You didn't ask for this curation. You didn't opt in to having your behavior analyzed. It's just what happens when you use social media. The algorithm watches what you like, what you watch, how long you watch it, what you skip, what you comment on, and hundreds of other data points. Then it predicts what you'll engage with next and shows it to you.

This happens billions of times per day. On Facebook, Instagram, TikTok, YouTube, Twitter/X, Pinterest, Snapchat, and every other platform that shows you content. Each one has its own algorithm making its own predictions based on its own rules. Algorithms are one of the most important and least understood forces shaping the modern world. They decide what news you see. They determine what products get recommended to you. They influence what videos go viral and what videos nobody watches. They even help determine who gets hired for jobs, who gets approved for loans, and who gets admitted to college.

This is the story of algorithms: what they actually are, how they work, what makes them biased, who controls them, whether we can regulate them, and what your options are if you want to push back.

What Is An Algorithm? Breaking Down The Jargon

An algorithm is simply a set of step-by-step instructions designed to accomplish a specific task. The word comes from the name of a Persian mathematician named Al-Khwarizmi who lived around 800 AD and developed early mathematical procedures.

You use algorithms every day without thinking about them. If you follow a cookie recipe, you're following an algorithm.

Combine flour and butter. Mix until crumbly. Add eggs. Bake at 350 degrees for 12 minutes. These are step-by-step instructions designed to produce a specific outcome: cookies.

The same is true for computer algorithms. A computer algorithm is a sequence of mathematical operations and logical steps that process input data and produce output. The simplest algorithm might be something like: "if the temperature is below 32 degrees Fahrenheit, display a snow warning." More complex algorithms might involve millions of mathematical operations processing millions of data points.

Social media algorithms are a type of recommendation algorithm. They take data about a user (what videos they watched, what they liked, how long they watched it, what their friends liked) and feed it through mathematical equations designed to predict which content will keep that person engaged.

Here's what makes this different from traditional recommendation systems you might be familiar with. In the old days, a human editor decided what was important and what you should see. When you watched TV, a station manager decided which news stories to show and in what order. When you went to a bookstore, a manager decided which books to display in the front window.

Now, algorithms make these decisions. And they make them based on a simple mathematical principle: engagement. The algorithm's primary goal is to predict and show you content that will keep you on the platform. Not content that's most important. Not content that's most accurate. Content that's most engaging.

This is the key to understanding algorithms. They're not neutral. They're not just showing you what's popular. They're actively trying to keep your attention and predict what will make you keep scrolling, watching, and clicking.

How Do They Actually Work? The Technology Behind The Curtain

Social media algorithms work in roughly five stages, though the exact process differs between platforms.

The first stage is data collection. Every action you take on the platform generates data. When you like a post, the algorithm records it. When you watch a video for 30 seconds before scrolling, the algorithm records it. When you comment on something, search for something, or even hover your mouse over content without clicking, the algorithm records it. When you follow an account or unfollow an account, the algorithm records it. Modern algorithms pull data not just from the app itself but also from your browser history, device information, location, and other apps on your phone.

The second stage is data aggregation. All this data gets combined and stored. In 2025, the world created approximately 181 zettabytes of data. That's 181 trillion gigabytes. Each year this number grows. All of this data feeds into the algorithms.

The third stage is pattern recognition using machine learning. The algorithm looks for patterns in your behavior and the behavior of millions of people similar to you. If people who watch sports videos also tend to watch fitness videos, the algorithm learns this pattern. If people who engage with political content also engage with news content, the algorithm learns this. Machine learning algorithms are trained on massive datasets to recognize these patterns.

The fourth stage is prediction. Based on the patterns learned in stage three, the algorithm predicts what content you're likely to engage with. It might assign a score to different pieces of content representing how likely you are to engage with each one. A video about basketball might get a score of 0.8 (80 percent likely you'll watch it) while a video about cooking might get a score of 0.2 (20 percent likely).

The fifth stage is ranking and display. The algorithm ranks all available content by these scores and shows you the highest-scoring content first. This is what appears in your feed.

What's remarkable is that this entire process happens in real time. When you open TikTok, the algorithm doesn't just sort through your previous watch history. It's processing millions of data points in milliseconds. It's considering what you just did five minutes ago. It's considering what's trending globally. It's considering what your friends are watching. It's considering what similar users are watching. All in milliseconds.

Each platform has slightly different algorithms with slightly different goals. TikTok's algorithm is famously powerful at predicting engagement, which is why people describe TikTok as incredibly addictive. YouTube's algorithm is optimized for watch time. Instagram's algorithm considers multiple factors including likes, comments, saves, and shares. Twitter/X's algorithm considers retweets, replies, and quote tweets.

The platforms don't tell us exactly how their algorithms work because they treat the algorithms as proprietary secrets. Companies guard their algorithms fiercely because the algorithm is literally the engine of their business. A better algorithm means more user engagement, which means more advertising revenue.

The Problem: Algorithmic Bias And Why This Matters

Here's the problem with algorithms: they're only as good as the data they're trained on. And if that data is biased, the algorithm becomes biased.

Consider Google Translate. In 2017, researchers discovered that when translating from Turkish, a language that doesn't have gendered pronouns, into English, Google Translate exhibited gender bias. When translating a Turkish sentence that could apply to any gender, Google would pair he with hardworking and she with lazy. He was paired with strength and she with weakness. The algorithm learned these biases from the text it was trained on, which reflected biases in the broader culture.

The same problem occurs with social media algorithms. If the training data includes biases against certain groups, the algorithm will perpetuate and even amplify those biases. Research on algorithmic glass ceilings has shown how recommendation systems can systematically disadvantage certain groups. If women's content gets shown less frequently in early iterations, it accumulates fewer views, likes, and engagement metrics. The algorithm then learns that women's content performs worse and shows it even less, creating a vicious cycle.

This problem extends far beyond social media. Algorithms are used in criminal justice systems to determine bail amounts and sentencing recommendations. Studies have shown these algorithms exhibit racial bias, recommending harsher sentences for people of color. Algorithms are used in hiring to screen resumes and even conduct interviews. These algorithms can discriminate against women, people of color, and other groups based on biases in training data.

Algorithms are used in healthcare to determine insurance coverage and resource allocation. Algorithms are used in banking to determine credit scores and loan approvals. Algorithms are used in college admissions. In every domain, algorithms have been found to perpetuate historical biases and inequalities.

The insidious part is that algorithms feel objective. They feel like pure mathematics, free from human bias. But algorithms are created by humans, trained on human-generated data, and optimized for human-defined goals. All of these factors introduce bias.

Can We Control Algorithms? The Question Everyone's Asking

The short answer is yes, but it's complicated. Companies that own algorithms can control them. Facebook can change how its algorithm works. TikTok can adjust its recommendation system. YouTube can modify what videos it prioritizes. They do this constantly. When public backlash gets loud enough, companies make changes. But individuals using these platforms have almost no control. You can't opt out of the algorithm. You can't see how it works or understand why you were shown a particular post. You can't really push back against it.

There are a few limited ways to influence what an algorithm shows you. You can deliberately unlike content you don't want to see more of. You can block accounts or mute keywords. You can try to tell YouTube not to recommend a particular channel. But these are minor adjustments. The algorithm is still in control.

Some people argue that we can control algorithms through regulation. The European Union has taken this approach, passing the Digital Services Act which requires transparency and some algorithmic control. Some U.S. states have started regulating algorithms. Colorado's Senate Bill 24-205 covers algorithmic bias in high-risk AI systems. California's Assembly Bill 2013 requires disclosure of data sources for generative AI systems.

But enforcing these regulations is difficult. Companies don't want to reveal how their algorithms work because it's proprietary technology. Government agencies don't have the technical expertise to audit algorithms effectively. There's an inherent conflict between companies wanting to maximize engagement and regulators wanting to prevent harm.

Can We Change Algorithms? What Would That Look Like?

Yes, algorithms can be changed. But changing them in ways that actually help people faces significant obstacles.

One approach is transparency. Require companies to disclose how their algorithms work, what data they use, and what outcomes they produce. The theory is that if people understand how algorithms work, they can understand bias and hold companies accountable. But transparency has limitations. Even if you understand exactly how an algorithm works, you can't necessarily change it. And some researchers have found that transparency alone doesn't solve bias. In fact, sometimes knowing how an algorithm makes decisions allows people to game the system rather than improve it.

Another approach is algorithmic accountability. Hold companies responsible when their algorithms cause harm. If an algorithm discriminates against a group, the company should face penalties. This creates financial incentive to design better algorithms.

But accountability requires proving that an algorithm caused harm, which is difficult. If someone is denied a loan, is it because of the algorithm or because their credit score actually is low? If someone isn't shown a job posting, was it algorithmic bias or just bad luck?

A third approach is regulation. Governments could mandate that algorithms meet certain standards. They could require companies to test algorithms for bias. They could demand that high-risk algorithms have human oversight. They could ban certain types of algorithmic decision-making in critical domains like criminal justice and healthcare. But regulation is slow and companies are powerful. The tech industry lobbies aggressively against regulation. Some argue that regulating algorithms will stifle innovation. Others argue that regulating algorithms written in code is nearly impossible because code is constantly changing.

A fourth approach is algorithmic literacy. Teach people to understand algorithms, recognize bias, and question algorithmic decisions. If more people understood how algorithms work and why they're biased, they might demand change. But this is the slowest approach and puts the burden on individuals rather than on the systems that need to change.

What's Being Done Now? The Current Landscape In 2026

As of 2026, governments around the world are starting to act. The European Union passed the Digital Services Act and the AI Act, which include requirements for algorithmic transparency and bias mitigation. Companies operating in the EU must disclose information about how their algorithms work and must conduct impact assessments. The United States has not passed comprehensive AI regulation at the federal level, but individual states have started. Colorado, California, Illinois, and New York have all passed laws regulating algorithmic decision-making systems, particularly in areas like employment, insurance, and lending. China has taken a different approach, implementing strict government control over algorithms. Companies must register their algorithms. Content must be reviewed by government censors. The goal is less about preventing bias and more about preventing content the government dislikes.

At the international level, there's debate about whether algorithms should be regulated at all or whether self-regulation by industry is sufficient. The IEEE has developed ethical standards for AI systems. Various professional organizations are developing best practices. However, 62 percent of Americans say they have little or no confidence in federal agencies to regulate AI effectively. Even 53 percent of AI experts doubt that government can effectively oversee algorithmic systems. This points to a fundamental challenge: technology moves faster than regulation can keep up, and government agencies often lack technical expertise.

What Are Your Options? What Can You Actually Do?

If you want to push back against algorithms, here are realistic options.

1. Understand that algorithms exist and have goals. Just knowing that you're not seeing a chronological feed but rather a mathematically optimized one changes how you should think about social media. Every post shown to you was chosen because the algorithm thinks it will keep you engaged. This knowledge alone is valuable.

2. Diversify your sources of information. Don't rely entirely on social media algorithms for news. Read newspapers, listen to podcasts, watch different news sources. Actively seek out information rather than passively consuming what the algorithm serves you.

3. Engage in algorithmic resistance. Deliberately like and engage with content the algorithm wouldn't predict you'd like. Follow accounts that don't fit your profile. Watch videos on topics outside your normal interests. By doing this, you're training the algorithm with data that doesn't fit your normal pattern, making predictions harder.

4. Use privacy tools. Some people use VPNs or privacy browsers to limit the data companies can collect about them. This won't stop algorithms but it will give them less data to work with.

5. Support regulation. Vote for politicians who support algorithmic transparency and accountability. Contact your representatives. Demand that social media companies disclose information about how their algorithms work.

6. Consider using platforms differently. Use YouTube in incognito mode so it can't track your history. Turn off algorithmic recommendations and browse chronologically when possible. Some platforms offer settings to limit algorithmic personalization.

7. Understand that opting out entirely might be your best option. Not all algorithms are necessary. You don't need social media. You could choose to live without Facebook, Instagram, TikTok, or other algorithmically driven platforms. Some people have found that eliminating algorithmic feeds improves their mental health and information diet.

The Bottom Line

An algorithm is a set of step-by-step instructions designed to accomplish a specific task. Computer algorithms process input data through mathematical operations and produce output. Recommendation algorithms on social media use machine learning to predict what content will keep you engaged based on your behavior and billions of data points from similar users.

Social media algorithms work through five stages: data collection (recording everything you do), data aggregation (combining all this data), pattern recognition (learning patterns in behavior), prediction (scoring content by engagement likelihood), and ranking and display (showing highest-scoring content first). This entire process happens in milliseconds and happens billions of times per day.

Algorithms are not neutral. They're optimized for engagement, meaning they show you content designed to keep you on platforms, not necessarily content that's accurate, important, or healthy. Each platform has slightly different algorithms with slightly different goals.

Algorithms exhibit bias because they're trained on biased data. Gender bias, racial bias, and other forms of discrimination are perpetuated and amplified by algorithms. Algorithms are used in hiring, lending, criminal justice, and college admissions, where they have been documented to discriminate against marginalized groups.

Individuals have almost no control over algorithms. Companies can change them, but users cannot opt out. Some regulation is emerging through government action, particularly in the European Union and certain U.S. states, but enforcement is difficult, and technology moves faster than regulation. Algorithms can be changed through transparency requirements, accountability measures, regulation, and algorithmic literacy. However, each approach has limitations and faces resistance from industry.

As of 2026, governments are increasingly regulating algorithms, but 62 percent of Americans lack confidence that government can effectively oversee algorithmic systems. The gap between how fast technology evolves and how fast regulation responds remains a fundamental challenge. Your options include understanding how algorithms work, diversifying information sources, engaging in algorithmic resistance, using privacy tools, supporting regulation, using platforms differently, and potentially choosing to opt out entirely.

Algorithms are increasingly central to how information is distributed and decisions are made in modern society. Understanding them, recognizing their biases, and demanding accountability is crucial for individuals and for society.

Sources

Akpinar, G. G., Buolamwini, J., Katz, J., Riedl, M. O., & Selbst, A. D. (2022a). Technology, Policy, and Society: Regulating Algorithmic Bias. MIT Media Lab. Retrieved from https://www.media.mit.edu

Celis, L. E., Huang, S., Keswani, V., & Vishnoi, N. K. (2018). Fair Algorithms: A Survey. arXiv. Retrieved from https://arxiv.org/pdf/1808.10521.pdf

Cornell Brooks Public Policy. (2025, November 12). What Americans Really Think About AI Algorithms: Public Confidence and Transparency in Government. Retrieved from https://publicpolicy.cornell.edu/masters-blog/what-americans-really-think-about-ai-algorithms-public-confidence-and-transparency-in-government/

Ekstrand, M. D., Das, A., & Burke, R. (2018). Fairness and Discrimination in Recommendation and Ranking Systems using Machine Learning. arXiv. Retrieved from https://arxiv.org/pdf/1811.07867.pdf

Gunderson Dettmer. (2026, February 5). 2026 AI Laws Update: Key Regulations and Practical Guidance. Retrieved from https://www.gunder.com/en/news-insights/insights/2026-ai-laws-update-key-regulations-and-practical-guidance

Hootsuite. (2026, May 7). Social media algorithms in 2026: How they rank content. Retrieved from https://blog.hootsuite.com/social-media-algorithm/

QuickFrame. (2026, March 4). How Do Social Media Algorithms Work in 2026? Retrieved from https://quickframe.mountain.com/blog/how-do-social-media-algorithms-work/

Social Champ. (2026, February 13). How The Social Media Algorithm Works In 2026. Retrieved from https://www.socialchamp.com/blog/social-media-algorithm/

Sprinklr. (2025, July 3). Social Media Algorithm and How They Work in 2025. Retrieved from https://www.sprinklr.com/blog/social-media-algorithm/

TechTimes. (2025, December 25). How Recommendation Algorithms Shape What You Watch and Share: Streaming and Social Media Explained. Retrieved from https://www.techtimes.com/articles/313585/20251225/how-recommendation-algorithms-shape-what-you-watch-share-streaming-social-media-explained.htm

The Regulatory Review. (2025, December 15). The Political Limits of Algorithmic Governance. Retrieved from https://www.theregreview.org/2025/12/15/livermore-the-political-limits-of-algorithmic-governance/

Unified AI Hub. (2025, December 19). Current State of AI Regulation in 2026: Global Trends, Policies, and Challenges. Retrieved from https://www.unifiedaihub.com/blog/current-state-ai-regulation-2026

Wikipedia. (2026, February 27). Regulation of algorithms. Retrieved from https://en.wikipedia.org/wiki/Regulation_of_algorithms