The question is, “How do AI chatbots work?”
An even better question is, “What is generative AI generating, exactly?”
Generative AI does not know anything. And, contrary to popular opinion, it does not have the ability to think. It is not sentient, nor can it evaluate sentiment. It is a neural network modeled on the human brain’s network of neurons. It learns based on a “large language model” that requires feeding it billions of data points from digital text, including books, news articles, online chat logs, Wikipedia content and more.
From this massive and endless data resource, it pinpoints billions of distinct data points in the way people connect words, letters and symbols so that, when prompted, it learns to generate its own content. For example, to identify a fire hydrant, it would have to pinpoint patterns in thousands of images of fire hydrants to learn to recognize them.
Given that the internet is its teacher, generative AI can, and does, regurgitate offensive content or content with blatant factual errors, with absolute confidence. It also cannot discern between fact or fiction, nor legally protected material, such as copyrights, trademarks or sourcing rights, or the origins of any source material, whether true or false. This leads to accusations of legal violations and plagiarism. For example, when it writes code, it can introduce inappropriate steps or blatant errors that can cost the commissioning company dearly in recalls and damage to its reputation.