Generative Artificial Intelligence (AI)

Big Picture

Generative artificial intelligence is a type of AI that creates new content, such as text, code, images, audio, video, or data, by learning patterns from existing examples. Generative AI is important because it changes the role of computers from systems that only follow explicit instructions to systems that can produce new, useful, and sometimes surprising outputs. Instead of only retrieving information or performing calculations, generative AI systems can write explanations, generate images, produce code, summarize documents, answer questions, create music, and simulate conversation.

The key idea is this: generative AI does not “understand” in the same way a human understands. It uses statistical and mathematical models to identify patterns in training data and then generates new outputs that are likely to match those patterns. NIST defines generative AI as a class of AI models that emulate the structure and characteristics of input data to generate synthetic content. 

What Is Generative AI?

Generative AI is artificial intelligence that produces new content rather than simply classifying, sorting, searching, or predicting from existing data.

Examples of generative AI outputs include:

• text, such as essays, summaries, explanations, emails, and stories
• programming code, such as Python functions, SQL queries, or HTML pages
• images, such as illustrations, diagrams, or photorealistic scenes
• audio, such as speech, sound effects, or music
• video, such as generated clips or edited visual sequences
• structured data, such as tables, synthetic datasets, or test data

A traditional program usually follows explicit instructions written by a programmer. A generative AI model, by contrast, is trained on large amounts of data and learns patterns from that data. It then uses those learned patterns to generate new output.

Generative AI Compared With Other AI Systems

How Generative AI Works

Most modern generative AI systems are based on machine learning. A model is trained using large amounts of data. During training, the model adjusts internal numerical values called parameters. These parameters help the model represent patterns in the training data.

A simplified process looks like this:

1. Collect data: The system is given a large training dataset, such as text, images, code, or audio.
2. Preprocess data: The data is cleaned, filtered, formatted, and converted into a structure the model can use.
3. Train the model: The model learns statistical patterns from the data.
4. Evaluate the model: The model is tested to see whether its outputs are accurate, useful, safe, and appropriate.
5. Generate output: A user provides a prompt, and the model produces new content based on learned patterns.

For text-based systems, such as large language models, text is often broken into smaller units called tokens. The model predicts likely next tokens based on the prompt and its training. Google’s machine learning materials describe LLMs as systems that use concepts such as tokens, parameters, context, and self-attention. 

Large Language Models (LLMs)

A large language model, or LLM, is a generative AI model designed to process and generate language. LLMs are commonly used for chatbots, translation, summarization, question answering, tutoring, code generation, and writing assistance.

Many LLMs are based on a neural network architecture called a transformer. Transformers are effective at processing sequences of data, such as words or tokens in a sentence. IBM describes LLMs as deep learning models trained on large amounts of data and commonly built using transformer-based architectures.

An LLM does not simply copy a sentence from memory each time it answers. Instead, it generates output by using learned statistical relationships among words, phrases, concepts, and structures. This is powerful, but it also creates risks: the model may produce fluent text that sounds correct but is actually false.

Foundation Models

Many generative AI systems are built from foundation models. A foundation model is a large model trained on broad data that can be adapted to many different tasks. IBM describes foundation models as AI models trained on very large datasets that can perform a wide range of general tasks and serve as building blocks for more specialized applications. :contentReference[oaicite:4]{index=4}

For example, a foundation model might be adapted for:

• customer support chatbots
• medical document summarization
• programming assistance
• legal document analysis
• language translation
• image generation

This flexibility is one reason generative AI has become so important. A single underlying model can support many different applications.

Important Generative AI Model Types

A Simple Example

Suppose a student asks a generative AI system:

Explain binary search to a 10th grade computer science student.

The model processes the prompt, identifies the task, and generates an explanation based on patterns it has learned from examples of computer science explanations, tutorials, textbooks, and other language data.

The generated answer might be useful, but the student should still check it. The model might make a mistake, skip an important condition, or explain the algorithm without mentioning that binary search requires sorted data.

Why Generative AI Is Powerful

Generative AI is powerful because it can reduce the time needed to create drafts, prototypes, examples, and explanations. It can help users move from a blank page to a first version quickly.

Useful applications include:

• drafting and revising text
• generating programming examples
• explaining complex topics at different levels
• summarizing long documents
• creating practice questions
• generating synthetic training data
• brainstorming design ideas
• supporting accessibility through translation and simplification

In computer science education, generative AI can be helpful when used as a tutor, debugging partner, or explanation generator. However, it becomes harmful when students use it to replace their own thinking.

Limitations of Generative AI

Generative AI systems have significant limitations. Students must understand these limitations before trusting generated output.

• Hallucination: The model may generate false or unsupported information.
• Bias: The model may reproduce biases found in training data.
• Lack of true understanding: The model may produce correct-looking output without real comprehension.
• Outdated information: The model may not know recent facts unless connected to current sources.
• Prompt sensitivity: Small changes in wording may produce different answers.
• Privacy risk: Users may accidentally enter personal, confidential, or sensitive information.
• Academic integrity risk: Students may submit AI-generated work as if it were their own.

Ethical Issues

Generative AI raises ethical questions because it can produce content at scale and because its outputs may affect real people. NIST’s Generative AI Profile identifies risks such as fraudulent content, impersonation, disinformation, and the need for trustworthy characteristics such as accountability, transparency, safety, reliability, interpretability, and explainability. 

Important ethical issues include:

• Bias and fairness: A model may treat people or groups unfairly because of biased training data.
• Privacy: Training data or user prompts may include sensitive information.
• Consent: People may not have agreed for their work or personal data to be used in training.
• Transparency: Users may not know when content was generated by AI.
• Accountability: It may be unclear who is responsible when AI-generated output causes harm.
• Environmental impact: Training and running large models can require significant energy and computing resources.
• Misinformation: AI-generated text, images, audio, or video can be used to deceive people.
• Academic integrity: Students must distinguish between using AI for support and submitting AI-generated work as their own.

Generative AI and Computational Thinking

Generative AI can be understood using computational thinking:

Generative AI and Programming

Generative AI can assist programmers by generating code, explaining errors, writing comments, suggesting tests, and proposing alternative algorithms.

For example, a student might ask:

Write a Python function that performs a linear search on a list.

The model may generate useful code:

def linear_search(items, target):
    for index in range(len(items)):
        if items[index] == target:
            return index
    return -1

However, the student still needs to understand the code. They should be able to trace it, explain it, test it, and modify it. AI-generated code can contain bugs, security flaws, inefficient algorithms, or syntax that does not match the requirements of the task.

Good and Poor Uses of Generative AI by Students

Common Misconceptions

IB Command Term Focus

Generative AI is a good topic for several IB command terms.

Good command term understanding

For evaluate, a strong answer weighs both benefits and limitations, then reaches a justified judgement. For example, a student might argue that generative AI is useful for debugging and explanation, but dangerous if it prevents students from developing independent problem-solving skills.

Weak command term understanding

A weak evaluate answer only lists advantages. For example, “Generative AI is good because it helps students write code faster” is not enough. It does not weigh limitations or reach a supported judgement.

Exam-Style Questions

1. Define generative artificial intelligence. [2]
2. Describe two types of content that can be produced by generative AI systems. [4]
3. Explain why generative AI systems may produce inaccurate information. [4]
4. Discuss the ethical implications of using generative AI to create school assignments. [6]
5. Evaluate the use of generative AI as a programming support tool for beginner programmers. [8]

Key Terms

Artificial intelligence

A field of computer science concerned with systems that perform tasks normally associated with intelligent behaviour.

Generative AI

AI that creates new content based on patterns learned from training data.

Machine learning

A method of developing AI systems where models learn patterns from data rather than being explicitly programmed for every case.

Large language model (LLM)

A machine learning model designed to process and generate human language.

Foundation model

A large model trained on broad data that can be adapted to many different tasks.

Prompt

The input given to a generative AI system by a user.

Token

A unit of text processed by a language model. A token may be a word, part of a word, punctuation, or another text unit.

Training data

The data used to teach a machine learning model patterns.

Parameter

An internal numerical value adjusted during training that helps the model represent patterns.

Hallucination

An incorrect or unsupported output generated by an AI system that may appear convincing.

Bias

A systematic unfairness or distortion in data, model behaviour, or output.

Summary

Generative AI is a major development in computer science because it allows computers to create new content from learned patterns. It is closely connected to machine learning, neural networks, data preprocessing, hardware acceleration, model evaluation, and ethics. These systems can be extremely useful, but they must be used critically. A responsible computer scientist does not simply accept generated output. They verify it, test it, question it, and consider its social and ethical consequences.