Chatbot

A chatbot is a program you can have a conversation with. You type a question into an input field, and shortly afterward an answer appears in full sentences. To the user, this looks like a chat with a human. But there is no human behind it, only software running on a remote computer. The first programs of this kind existed as early as the 1960s, though they followed rigid rules. Only in the last few years have the responses become so fluid and flexible that millions of people use such programs every day.

From niche tool to the most widely used AI interface

Today, the chatbot is the most important way ordinary people come into contact with artificial intelligence. You need no programming knowledge and no manual. Anyone who can write can use the tool. This low barrier to entry explains why ChatGPT reached around 100 million users within two months of its launch at the end of 2022. Hardly any product has ever grown that fast.

For companies, the economic appeal is obvious. Customer service staffed by humans costs money per conversation. A chatbot answers standard questions around the clock at a fraction of that cost. That is why chatbots are now showing up at banks, insurance companies, online shops, and government agencies. Critics point out that customers often end up stuck in loops of meaningless answers.

On the stock market, the term is also a signal. When a corporation announces it is launching its own chatbot, it usually means more than just a help window on the website. What is meant is entry into a business field that is currently attracting enormous amounts of investment.

What happens between question and answer

At the core of modern chatbots is what is known as a language model. This is a program that has learned from enormous amounts of text which word typically follows a given sequence of words. It builds its answer piece by piece. Each new word fragment is calculated based on everything written so far. The model does not understand the content the way a human does.

For this word predictor to become a polite conversation partner, a second step is needed. Humans rate many example answers and mark which ones are helpful and which are inappropriate. The model is then fine-tuned based on these ratings. Only through this does it learn to answer questions instead of simply continuing them, and to refuse sensitive requests.

An important point is memory. The chatbot does not remember anything on its own; instead, the previous conversation history is sent along again with every new message. This history is limited. Once it is full, the oldest parts drop out, and the bot forgets the beginning. A well-known problem is called hallucination: the chatbot invents facts, sources, or figures and phrases them as convincingly as genuine information.

Recognizing chatbots in everyday life

The best known are the major assistants: ChatGPT from OpenAI, Gemini from Google, Claude from Anthropic, and Copilot from Microsoft. They help with writing texts, explain school material, translate, and write program code. Many schools are currently debating how to handle them. Beyond that, chatbots have long been built into products, into search engines, office programs, and development environments.

The second, less conspicuous group are the service bots. The small window in the bottom right corner of a shop page belongs to this category. Such systems are often built much more simply than the major assistants. Some merely follow a fixed decision tree with pre-written answers. If you notice the same sentences coming up again and again, you are probably dealing with such a rule-based system.

A common misconception is equating chatbot with artificial intelligence. The chatbot is only the interface, that is, the way you talk to the system. The same technology also powers search functions, image generators, or coding assistants without any chat window at all. Conversely, behind a chat window there can also be simple technology with no language model whatsoever.