Artificial Intelligence: Part 5

Capability

In AI, capability means what a system can do. One model may be capable of translation, another of image generation, another of coding or reasoning across documents. Capability is not the same as reliability. An AI system may be capable of doing a task sometimes, but still make mistakes. For practical use, we need to ask: can it do the task consistently, safely and usefully?

Catastrophic Forgetting

Catastrophic forgetting happens when an AI model learns new information but loses performance on what it previously knew. This is a problem in systems that are updated over time. Humans can also forget, but AI models can sometimes forget in abrupt and surprising ways. Researchers study ways to help models learn continuously without wiping out earlier knowledge.

Causal AI

Causal AI focuses on cause and effect, not just patterns. Many AI systems can find correlations: for example, people who buy product A often buy product B. But correlation does not prove causation. Causal AI asks deeper questions: did A actually cause B? This matters in medicine, economics, public policy and business decisions, where acting on false patterns can be harmful.

Chain-of-Thought

Chain-of-thought refers to the step-by-step reasoning that may help an AI model solve a problem. For example, a maths question may be easier if the model breaks it into stages. However, users should be careful: an AI’s displayed reasoning may sound convincing but still be wrong. The important thing is not whether the explanation is elegant, but whether the final answer is correct and verifiable.

Chatbot

A chatbot is software that communicates through conversation. Early chatbots followed simple scripts. Modern AI chatbots can answer questions, write text, summarise documents, translate languages, help with coding and simulate dialogue. A chatbot can feel intelligent because it uses natural language, but it may still misunderstand the user, invent facts or give poor advice if not checked.

ChatGPT

ChatGPT is an AI assistant that uses large language models to understand and generate text. People use it for writing, brainstorming, learning, coding, research support, summarising and many other tasks. It is not a search engine in the traditional sense, and it does not automatically know everything that is true or current unless connected to appropriate tools or sources. Its output should be reviewed, especially for factual or professional use.