In an era where Artificial Intelligence (AI) is permeating every aspect of life—from learning, working to content creation—understanding the foundational terminology of AI is essential. This article will help you see the big picture of AI by connecting key concepts.

What is AI and why is it important?

Artificial Intelligence (AI): A field of research in computer science aimed at developing machines capable of thinking like humans.

AI technology enables machines to reason and make decisions. An important branch of it is Machine Learning (ML), a part of computer science that focuses on developing systems capable of performing complex tasks without human intervention.

How AI learns and works

To function effectively, AI needs training:

• Training Data: A dataset used to teach a machine learning model, including input data and corresponding correct answers.
• Machine Learning Model: A mathematical representation of a learning algorithm that can make predictions or decisions based on data.

Types of Machine Learning:

• Supervised Learning: Models learn from labeled data, based on input-output pairs.
• Unsupervised Learning: Models learn patterns and structures from data without clear labels.
• Reinforcement Learning: Models learn through trial and error using feedback from their actions.

From Pretraining to Fine-tuning

• Pretraining: The process of feeding a large amount of language data such as books, websites… to recognize patterns and establish word relationships.
• Fine-tuning: The phase after pretraining, where the model is adjusted to perform a specific task using a smaller and more specialized dataset.
• Transfer Learning: A machine learning technique where knowledge from a pretrained model is used as a starting point for a new and specific task.

This leads to the creation of LLMs (Large Language Models).

The technologies behind modern AI

• Artificial Neural Network (ANN): A computational model inspired by the human brain, consisting of interconnected nodes (neurons) to perform tasks like image recognition or language processing.
• Deep Learning: A branch of machine learning using multi-layer artificial neural networks, allowing models to learn complex patterns.
• Transformer: A type of neural network architecture used in training generative AI.
• GAN (Generative Adversarial Network): An artificial neural network with two components: a generator that creates content and a discriminator that detects errors. Both improve each other to enhance output quality.

Generative AI and how you interact with it

• Generative AI: AI technology capable of generating various types of content such as text, images, audio, and synthetic data from human input.
• ChatGPT: A generative AI chatbot developed by OpenAI, using a pretrained large language model.
• Prompt: The input that a user enters into an AI model to get the desired output, for example, “plan a 10-day anniversary trip in Europe.”
• Prompt Engineering: The process of crafting and optimizing instructions so that generative AI can understand, process, and deliver accurate results.

Modern AI can also process multiple data formats:

• Multimodal AI: AI models capable of processing and generating various types of data such as text, images, video, or voice.
• Diffusion: A machine learning method that involves adding random “noise” to data (typically images) and training the AI to restore the original image or generate a new version.

The dark side of AI

AI is not perfect. Some common issues include:

• Hallucination: When a generative AI model confidently produces answers that sound plausible but are entirely incorrect.
• Jailbreaking: Manipulating an AI model using prompts to bypass its established restrictions, even eliciting harmful responses.
• Model Collapse: A hypothetical scenario where newer AI models are trained mostly on AI-generated content, lowering output quality due to repetitive and impoverished data.
• Bias: Unfair or inaccurate AI predictions or decisions caused by imbalances in training data.
• Algorithmic Bias: When an AI algorithm consistently produces biased or unfair outcomes, often due to skewed training data.

Ethics and control

• Adversarial Machine Learning: A research area focused on attacks against machine learning algorithms, as well as methods to defend against them.
• AI Ethics: The study and practice of ensuring AI systems are developed and used in alignment with ethical values and standards.
• Alignment: A research area that ensures AI models stick to intended goals, remain safe and trustworthy, and uphold human values and priorities.
• Guardrails: Policies and regulations embedded into AI models to prevent harmful behavior, such as avoiding disturbing content or misinformation.
• Transparency: The practice of making AI operations and decision-making clear to users, enhancing understanding and trust in the system.

Supporting concepts to know

• Autocomplete: A feature that predicts the rest of a word or the next word you’re typing—commonly found in messaging apps or search engines.
• DeepFake: AI-generated content (usually images or videos) that convincingly mimics a human’s face or voice.
• Labelling: A step in supervised machine learning that assigns labels to raw data so models can learn context.
• Stochastic Parrot: A critical term referring to AI models that merely “repeat” learned language patterns without real understanding—like a parrot mimicking speech.
• Turing Test: A test of a machine’s ability to exhibit intelligent behavior indistinguishable from that of a human.
• AGI (Artificial General Intelligence): A hypothetical concept of an AI that surpasses human intelligence.

Conclusion

Understanding the concepts in the AI ecosystem is essential to harness it effectively, safely, and responsibly. From basic machine learning models to risks like bias and hallucination, a solid foundation will help you not only use AI but also control and collaborate with it wisely.