Crash Course: AI 101

Artificial Intelligence (AI)

Computer systems that perform tasks requiring human-like intelligence. AI has been defined in law as “a machine-based system that can, for a given set of human-defined objectives, make predictions, recommendations or decisions influencing real or virtual environments.” 15 USC 9401 (3)

• AI model: A software component of an information system that implements AI technology and uses computational, statistical, or machine-learning techniques to produce outputs from a defined set of inputs.  

• AI system: Any data system, hardware, tool, or utility that operates, in whole or in part, using AI. 

• Automated Decision System: Any computational process derived from machine learning, statistical modeling, data analytics, or AI that issues a simplified output, including a score, classification, or recommendation, to materially influence or replace human decision making.

Deep Learning

Deep learning kicked off the modern AI revolution in 2012, building on simpler techniques generally referred to as machine learning (ML). Deep learning uses large amounts of computing power (“compute”) to process large amounts of data to create an AI model that can predict, identify, or generate information from new examples. Deep learning relies on artificial neural networks with multiple layers – algorithms inspired by the architecture of the brain – to analyze data.

Why It Matters

Instead of programmers writing specific rules, deep learning systems learn patterns from massive amounts of data. This is why modern AI is so good at tasks like recognizing images, understanding speech, and generating text. This is also why data access is so important for AI development – more training data typically results in better deep learning models. Deep learning powers the vast majority of modern AI applications, from ChatGPT to autonomous vehicles.

Generative AI (GenAI)

AI models that can create new content, such as text, images, audio, and video, based on patterns learned from existing data. Generative AI models rely on generative architectures, such as autoencoders, transformers, and generative adversarial networks. Multimodal LLMs, diffusion models, generative adversarial networks, and others are all examples of generative AI. 

Why it matters: Instead of just analyzing or classifying information, generative AI can produce entirely new content that didn’t exist before.

Large Language Models (LLMs) 

AI models trained on vast amounts of text data to understand and work with human language. LLMs process and manipulate language by learning patterns from text, producing new text by predicting what words should come next, LLMs can also work with non-human "languages" such as code or protein sequences. The training data used to develop LLMs can influence their performance at different tasks. A common way to tailor LLMs for specific tasks is by training them further on domain-specific data, known as “fine-tuning.” 

Examples: OpenAI’s GPT-4.5, Meta’s Llama 4, Google’s Gemma 3

Use Cases: Drafting documents, code generation, language translation, summarization 

Many advancements in AI rely on augmenting generative AI or LLMs to make them more useful in different contexts.

Retrieval-Augmented Generation (RAG)

RAG is a technique that modifies LLMs to enable them to search through information from external sources such as specific documents or databases to inform their output. This can allow LLMs to provide more accurate or reliable responses in certain contexts.

Examples: Customer service chatbots that can use information from a company’s internal documentation and policies to inform responses. 

Use Cases: Document analysis, policy research, legal brief preparation, enterprise knowledge management

Reasoning Models

AI models that can work through complex problems by breaking them down into smaller tasks that it can work through step-by-step. Reasoning models are typically LLMs and can be highly effective at solving difficult or multi-faceted tasks.

Examples: OpenAI's o3, Claude Opus 4, Google’s Gemini 2.5 series, DeepSeek R1

Use Cases: Scientific research, mathematical problem-solving, complex analysis, strategic planning