AI Key Concepts (1)#

To use AI effectively, it’s important to understand the key concepts behind it. This helps you understand how AI services and tools work, and why the latest news and trends matter.

AI is still evolving at a rapid pace. Services and tools are continuously updated, and new developments emerge all the time. If you build a solid foundation, you can follow the changes more easily and keep improving your ability to apply AI in practice.

AI (Artificial Intelligence)#

AI refers to computer systems that imitate human intelligence to perform tasks such as recognizing or generating text, images, and videos. It can also refer to the technology itself, or the academic field related to implementing that technology. Discussions around AI began in the 1940s, and the term “Artificial Intelligence” was officially coined in 1956.

AI is sometimes categorized as Weak AI vs. Strong AI. Weak AI refers to AI limited to specific tasks or problem-solving. Strong AI refers to AI with human-level intelligence and consciousness. A similar classification is ANI (Artificial Narrow Intelligence) vs. AGI (Artificial General Intelligence): ANI is close to Weak AI, and AGI is close to Strong AI.

These two classification systems are used in slightly different contexts. Weak/Strong has been discussed for a long time and is often used from a philosophical perspective, while ANI/AGI is used more frequently from a technical perspective. As of December 2025, most AI services we encounter in daily life fall into the ANI category, and there are still many different opinions on when (or how) AGI will emerge.

Machine Learning and Deep Learning#

Machine learning is a field of AI where you design a model, train it on data, and use the trained model to perform tasks.

For example, suppose we have the equation (y = ax + b) and a known data point ((x_1, y_1)). If you’re asked to find (a) and (b), and then compute (y_2) for a new value (x_2), you can think of this as learning a rule from data. In machine learning, a function (or a set of functions) that represents the relationship between input (x) and output (y) becomes the model. If you train the model with enough ((x, y)) data and find appropriate parameters (e.g., (a) and (b)), you can predict (y) for a new (x).

Deep learning is a subfield of machine learning that uses artificial neural networks (ANNs), inspired by the structure of the human brain. In other words, it’s a type of machine learning that learns complex patterns from data through neural networks.

Training Methods in Machine Learning#

Machine learning training methods can be broadly categorized into four types.

Supervised Learning#

This method trains a model using pairs of input and output (labeled data). For example, to build a model that can classify whether a photo contains a cat, you can train it with photos labeled “cat” and “no cat.” In this case, the photo is the input, and the label is the output. Another example is an email spam filter, trained with spam and non-spam emails.

Unsupervised Learning#

This method trains a model using only input data without labels. While supervised learning teaches the model correct answers (“This is A, this is B”), unsupervised learning finds patterns or groups in unlabeled data (clustering). For example, it can group similar news articles together, or identify related products for recommendation systems.

Semi-Supervised Learning#

This is a combination of supervised and unsupervised learning, using both labeled and unlabeled data. In real-world datasets, labels are often incomplete. For instance, if you have many photos but only some are labeled as “cat” or “no cat,” a semi-supervised approach can use unlabeled data alongside labeled data to improve learning.

Reinforcement Learning#

This method improves a model by repeatedly evaluating its outputs and reinforcing better behavior, similar to training a dog: good actions receive rewards, bad actions do not. Reinforcement learning is not mutually exclusive with other methods. For example, you might start with supervised learning and then further improve the model with reinforcement learning.

LLM (Large Language Models)#

LLMs are deep learning models trained on massive amounts of text data, enabling them to understand and generate language. Examples include OpenAI’s GPT (Generative Pre-trained Transformer), Google’s Gemini, and Anthropic’s Claude.

Deep learning is about designing and training neural networks, and there are different architectures depending on the task. Common types include CNNs (Convolutional Neural Networks), RNNs (Recurrent Neural Networks), and Transformers. Modern LLMs are primarily based on the Transformer architecture. CNNs are often used for 2D data such as images, while RNNs and Transformers are used for natural language processing, speech recognition, and translation. In NLP, Transformer-based LLMs have become dominant, and the use of RNNs has decreased significantly.