# AI in Plain Language, Part 1: Definitions and Models

An introduction to artificial intelligence and models, using human cognition to explain parameters, training sets, fitting, and cost functions.

Canonical: https://samanhappy.com/en/writing/ai-in-plain-language-definitions-and-models
Language: en
Built by: Saman
Last updated: 2023-03-11

English translation of my Chinese article originally published on WeChat on 2023-03-11. The account below preserves the original context.
https://mp.weixin.qq.com/s/9wOB6wITa2u9ZIxt9cF8bw


Artificial Intelligence refers to an artificial form of intelligence manifested through machines, as distinct from the innate intelligence of humans.

Humans possess abilities that seem ordinary but are actually extraordinary, such as vision, language, and thought. AI simulates these abilities to reproduce intelligence and even surpass it. The human body is an immensely complex and intricate system, to say nothing of the structure and workings of our most important organ—the brain. Most people, we might say, underestimate their own capabilities as human beings. We often marvel at how impressive an AI is, but the truly impressive ones are ourselves. Someone who has not fully recognized and realized their own abilities is like a person holding the latest smartphone but using it only to make calls and send texts. What a waste of its potential.

One of the most important concepts in AI is the **model**, which is also the most basic processing unit of the human brain. Consider the familiar task of object recognition: the brain can identify what we see in a very short time—for example, a person across from us holding a dog on a leash. Recognition is so fast because we have already developed a series of models that can quickly produce results from visual input. Put simply, a model is a fixed **equation** that produces an output from input information, making it very fast to process things. Using models is also a necessity for humans: they reduce complexity to simplicity. Otherwise, we could not cope with a world of ever-increasing information.

Having a perfect model is the key to success, whether for an individual human or an AI program. The process of generating a model and continually improving it is generally called model training or modeling. AI trains models in the same way that humans come to understand the world. All human learning is essentially accomplished through repeated trial and error. If we continue comparing a model to an equation, training a model in AI means continually adjusting the equation's parameters until we obtain an equation that best satisfies the given training data. These equation parameters are called **model parameters**, the supplied training data is called the **training set**, and satisfying the training set is called **fitting**. Model development is a process of continuous improvement. The mathematical expression used to evaluate a model's quality is called the **cost function**.

I find the idea of a cost function thought-provoking for human behavior. As we grow, we develop countless models and use them unconsciously every day, yet we never consider whether the models themselves are correct or accurate. Although we cannot evaluate our models with an explicit cost function as AI does, we do need regular reflection and self-examination to keep improving them. That is how we can become better versions of ourselves as time and experience accumulate.