I think the whole Google AI Overview concept is fundamentally unacceptable because an AI only works “fine” as long as it runs in a private chat, where it has no official role, no legal consequences, and no responsibility for what it says. In a private chat I’m just talking to it, it doesn’t have to be official, it doesn’t have to be source‑accurate, and it doesn’t influence people’s decisions. The Google search engine, however, is not a private environment but a public, official system where people look for information, make decisions, and where every piece of output can have real consequences. If an AI is placed into such a surface, it is no longer chatting, it is performing a function, and that is a completely different situation.
The web contains everything: legal content, medical content, financial content, official documents, complaints, court rulings, copyrighted material, dangerous nonsense, conspiracy theories, trolling, and manipulated content. The AI cannot distinguish any of this because it does not see, does not understand, does not verify, does not know what is true or false, what is official or dangerous, what has consequences, what is legal, medical, financial, or manipulated. It does not summarize, it does not interpret, it simply generates a new text based on statistical patterns. A statistical parrot cannot summarize the web because the web is not a clean, safe, homogeneous space but a chaotic mixture of everything, and the output will inevitably be distorted. A distorted output is unsuitable for summarization because a summary must be source‑accurate, consistent, and verifiable, and the AI is not capable of that.
Google’s AI Overview does not return the actual search results, does not show the source, does not show the link, but instead generates its own text. That means it is not telling me what it sees, but what it statistically predicts. This is absurd, because it looks like it is “telling me what it sees,” while in reality it sees nothing. In a private chat this might still be acceptable because there are no consequences, but in the public Google search engine it is not acceptable at all, because everything shown there carries weight.
The AI has no brakes, no limits, no legal framework. Tesla’s autopilot at least brakes, disables itself, warns the user, refuses to continue, and follows the law. Google’s AI Overview does not brake, does not stop, does not warn, does not shut itself off, and does not limit itself. That is why I say the situation is worse at Google, because the AI has no brakes.
Google does not take responsibility for the AI’s output. They officially state that the AI may make mistakes and that they do not take responsibility for what it says. This means the AI can be wrong, Google does not take responsibility, the AI cannot take responsibility, and in the end nobody takes responsibility. This is legally impossible, because if a system can make mistakes, someone must take responsibility for it. If nobody does, the entire operation becomes untenable. That is why I ask, with full justification: how does this “help” anyone if they take no responsibility for its operation?
The Google search engine is not a private chat but a public surface where people make real decisions. If I do not get actual search results there, but instead a distorted, generated text without responsibility, without brakes, without limits, then the entire operation is technologically unsuitable, logically absurd, and legally indefensible. Summarizing the search results is unlawful because the AI technology is not capable of doing it, and Google does not take responsibility for it. A statistical parrot cannot summarize the web, and that is why the entire AI Overview concept is unsuitable for the purpose they are using it for.
