We write algorithms. But are we ready to be responsible for their consequences?
Each of us at Codeforces is used to thinking about how to optimize code, how to pass all the tests, how to get a rating. But the reality is that the "tests" of algorithms in life are different: it is what children watch on YouTube, what news the society reads and what decisions companies make. An error in these algorithms costs more than a Wrong Answer.
Let's take Spotify as an example. Its recommendation algorithms are built on collaborative filtering, audio analysis, and language models. The goal is simple: to keep the user in the app longer. In practice, this leads to a narrow feed — we hear the same thing, and young artists are drowned in noise. Unfair? Yes. But the risk is limited to music and cultural diversity.
Now YouTube. The algorithm is the same principle: maximize "engagement". But here the price of error is higher. Imagine a child watching a scientific video. The algorithm quickly finds similar "scientific" videos, only about UFOs or "secret experiments". An hour later, the child is sure that this is true. The result is not just a narrowing of tastes, but an undermining of trust in science, cognitive distortions, and the mass dissemination of misinformation.
Same algorithm. Two platforms. Two realities. And if we're losing new artists on Spotify, we're risking losing an entire generation of critical thinkers on YouTube.
Codeforces is not just a rating or virtual competition. It is a place where thousands of people gather who will work in Google, OpenAI, DeepMind, NVIDIA, Yandex and dozens of startups tomorrow.
That is why the conversation about responsible AI should start here. If we only discuss graph and dynamics problems, but ignore the question of how these algorithms are applied, then we miss the chance to influence the future. Today we practice solving optimization problems for the sake of points, and tomorrow these same optimizations will manage news feeds, recommend content to children or filter data for medicine.
Key Challenges
Technical risk: Algorithms can unintentionally generate misinformation or discrimination. What appears to be “neutral code” in practice affects people’s worldviews, opportunities, and even behavior.
Social risk: Children and other vulnerable groups are exposed to algorithms without any protection. On YouTube, this manifests itself in the fact that a recommendation system can convince a child of false information, and on health platforms, an algorithmic error can cost health.
Ethical challenge: Regulation and accountability must take into account the rights of people, not just the efficiency of algorithms. We need to think about who creates the data, who uses the product, and who may be harmed by the system’s errors.
How my view has changed
Was: “AI is a tool, and regulation is a boring formality.”
Now: “Regulation is part of system design: it makes algorithms safe and socially useful.”
Now I see that rules and frameworks do not hinder innovation, but help create algorithms that take into account people, culture, and consequences. Algorithms are no longer just “code,” but part of a system where fairness, transparency, and accountability matter.
Question for the CF community
We are designing systems that will manage the attention of millions tomorrow. What kind of "insurance" should be built into these algorithms? Each of us can write algorithms that affect society tomorrow. Let's think: what principles will help make them safe and useful?
P.S. AI without rules is like a car without brakes. We can make it safe or dangerous.
