Artificial intelligence has moved from research labs into boardrooms, classrooms and dinner-table debates. But the vocabulary that surrounds it can be dense, inconsistent and, at times, deliberately vague. Here is a plain-language guide to the terms you are most likely to encounter.
The basics
Artificial intelligence (AI) is a broad umbrella term for computer systems designed to perform tasks that normally require human intelligence common examples include recognizing images, translating languages and making recommendations.
Machine learning is the subset of AI in which systems improve at a task by processing data rather than by following rules written line-by-line by a programmer. Most of what people call “AI” today is machine learning.
Neural network refers to a computing structure loosely inspired by the brain, built from layers of interconnected nodes that pass signals to one another. Deep learning describes neural networks with many such layers, an approach that powers most modern AI breakthroughs.
The generative era
Generative AI is the category responsible for the current wave of attention. Rather than simply classifying or predicting, these systems produce new content el text, images, audio, video or code.
Large language model (LLM) is the engine behind chatbots and writing assistants. Trained on enormous quantities of text, an LLM works by predicting the next piece of text in a sequence, which allows it to answer questions, summarize documents and draft prose.
Prompt is the instruction or question a user gives a model. Prompt engineering is the practice of phrasing those instructions to get better results.
Token is the unit a model actually processes a chunk of a word, a whole word or a punctuation mark. Pricing and usage limits for many AI services are measured in tokens.
Training is the costly, computation-heavy process of building a model from data. Fine-tuning adjusts an already-trained model for a narrower purpose, such as legal research or customer service.
Parameters are the internal values a model adjusts during training. Models are often described by their parameter counts, though bigger does not automatically mean better.
The problems
Hallucination describes the tendency of AI systems to state false information confidently. Because models generate plausible-sounding text rather than retrieve verified facts, errors can look identical to correct answers.
Bias refers to skewed or unfair outputs that reflect patterns in training data or design choices, with consequences in hiring, lending, policing and health care.
Deepfake means synthetic audio, images or video depicting real people doing or saying things they never did n a central concern for elections and fraud.
Black box captures the difficulty of explaining why a complex model produced a particular output, even for its own developers.
The frontier
AI agent describes a system that takes multi-step actions on a user’s behalf booking travel, writing and running code, or navigating websites ather than simply responding to a single prompt.
Multimodal means a model can handle more than one type of input or output, such as reading an image and describing it in text.
Artificial general intelligence (AGI) refers to a hypothetical system matching or surpassing human ability across virtually all cognitive tasks. There is no agreed-upon definition or test, and experts disagree sharply about whether and when it might arrive.
Alignment is the effort to ensure AI systems pursue goals that match human intentions and values a field that has grown alongside the technology’s capabilities.
Understanding the vocabulary will not settle the debates. But it does make it easier to tell substance from salesmanship. Read More
Leave a Reply