What Is Artificial Intelligence?

What You Will Learn

By the end of this lesson, you should be able to:

• Explain what artificial intelligence actually means.
• Understand why there is no single universally accepted definition of AI.
• Distinguish an AI system from an ordinary computer program.
• Understand the relationship between AI, machine learning, deep learning, and generative AI.
• Understand that AI is not a single technology.
• Explain why today's AI systems can perform tasks that appear intelligent without necessarily thinking like humans.
• Understand the difference between an AI model and an AI system.
• Identify common examples of AI in everyday life.
• Understand why today's AI boom is different from earlier periods of AI development.
• Begin thinking about AI as a technology stack rather than a magical black box.
• Understand where local AI fits into the larger AI landscape.

1.1 So, What Actually Is AI?

Let's start with the question that sounds incredibly simple but is surprisingly difficult to answer:

What is artificial intelligence?

You will find hundreds of definitions.

Some definitions focus on machines performing tasks that normally require human intelligence. Others focus on machines making predictions, recommendations, or decisions. Some definitions focus on learning from data. Others focus on reasoning, perception, language, planning, or problem-solving.

There is a reason for this disagreement.

Artificial intelligence is not one specific technology.

AI is an enormous field containing many different techniques, algorithms, models, systems, and applications. A modern large language model is AI. A computer vision system that detects objects in photographs is AI.

A system that predicts whether a bank transaction is fraudulent can be AI. A recommendation system deciding which products you might want to buy can be AI. A system controlling a robot can be AI.

A system that generates an image from a text description can be AI. These systems can be radically different internally. There is no single "AI algorithm" sitting underneath all of them.

The National Institute of Standards and Technology (NIST), for example, describes an AI system in terms of an engineered or machine-based system that can generate outputs such as predictions, recommendations, or decisions for given objectives. Google's machine-learning glossary uses a broader description of AI as a non-human program or model capable of solving sophisticated tasks.

These definitions are useful, but they still don't tell you everything.

To really understand AI, we need to stop thinking of it as a mysterious category of software and start thinking about what the software is actually doing.

1.2 The Basic Idea: Getting a Computer to Perform Tasks That Require Intelligence

Computers are extremely good at following explicit instructions. Consider a simple calculator. If you tell a calculator:

25 + 17

it performs a defined mathematical operation and gives you:

42

There is nothing particularly mysterious happening. The computer doesn't need to understand what 25 means. It doesn't need to understand what addition means in a human sense.

It simply executes instructions that implement an arithmetic operation. Traditional software works in much the same way. Imagine a very simple program:

IF temperature > 30
    display "It's hot"
ELSE
    display "It's not hot"

The programmer explicitly wrote the rule. The computer follows it. Now imagine a much more complicated task:

Look at this photograph and determine whether it contains a dog.

You could theoretically attempt to write thousands or millions of explicit rules describing what a dog looks like. You might try to tell the computer:

• look for four legs
• look for fur
• look for two ears
• look for a particular body shape
• look for a nose
• look for eyes
• look for a tail
• look for particular proportions

But this quickly becomes ridiculous. What happens if the dog is lying down? What if the dog is partially hidden?

What if the photograph is dark? What if it is a tiny dog in the distance? What if the dog is wearing clothing?

What if it is a cartoon? What if the photograph contains a wolf? What if the dog is facing away from the camera?

Writing explicit rules for every possible situation becomes extraordinarily difficult. This is one of the fundamental ideas behind modern AI.

Instead of attempting to manually specify every rule required to solve a complicated problem, we can create systems that learn patterns from data.

That distinction is enormously important.

1.3 Traditional Programming vs Machine Learning

A simplified traditional programming model looks like this:

Human-written rules + data → output

For example:

Rules:
IF transaction > R50,000
AND country = unusual
AND account has never made an international transaction
THEN flag transaction

The programmer determines the rules. Machine learning changes the basic process. A simplified machine-learning model looks more like:

Data + desired outcomes → learned model

For example, imagine giving a machine-learning system millions of historical bank transactions. Each transaction might contain information such as:

• transaction amount
• location
• time
• merchant
• account history
• device
• transaction frequency
• whether the transaction was eventually determined to be fraudulent

The machine-learning process attempts to discover statistical patterns in the data. Instead of a programmer manually writing:

"If these 14 conditions occur simultaneously, call it fraud."

the model can learn relationships from examples. The resulting model can then examine a new transaction and produce something such as:

Fraud probability: 97%

That is fundamentally different from a simple IF/ELSE rule. The model has learned a mathematical representation of patterns from data. This is one of the foundational ideas behind modern AI.

1.4 Does AI Actually "Think"?

This is where language becomes dangerous. People frequently say things like:

"The AI thinks..."

"The AI understands..."

"The AI knows..."

"The AI wants..."

These phrases can be useful conversational shortcuts, but you should be careful with them.

An AI system can produce behaviour that looks intelligent without necessarily possessing human-like consciousness, subjective experience, emotions, desires, or understanding.

For example, a language model can generate a detailed explanation of astrophysics.

That does not automatically mean the model experiences an understanding of astrophysics in the same way a human scientist does.

A model can generate a convincing emotional response. That does not establish that the model feels the emotion it is describing. A model can explain the concept of fear.

That does not mean the model experiences fear. A model can discuss its own existence. That does not by itself demonstrate consciousness.

This distinction becomes particularly important when studying large language models.

They can produce remarkably sophisticated outputs, but the mechanism producing those outputs is fundamentally computational.

At a high level, a language model receives numerical representations of input, processes them through a large mathematical network containing learned parameters, and produces probabilities for possible next tokens.

We will eventually go extremely deep into how this works. For now, remember this:

An AI system can exhibit intelligent behaviour without being a human mind implemented inside a computer.

That distinction will prevent a huge amount of confusion later in this course.

1.5 Intelligence Is Not One Thing

Another reason the definition of AI becomes difficult is that "intelligence" itself is not one simple ability. Humans can:

• recognise objects
• understand language
• remember information
• learn from examples
• solve mathematical problems
• navigate physical environments
• plan
• reason
• make decisions
• recognise faces
• understand social situations
• generate ideas
• create art
• manipulate physical objects
• communicate
• adapt to unfamiliar situations

An AI system might be extremely capable at one of these tasks while being terrible at another.

For example, an image-classification model might be exceptionally good at distinguishing different types of plant disease while having absolutely no ability to hold a conversation.

A language model may be capable of writing software but unable to physically pick up a cup. A chess engine can defeat elite human players while knowing nothing about how to make breakfast. So when somebody says:

"Is AI intelligent?"

the question is incomplete. A better question is:

Intelligent at what?

This idea becomes increasingly important as you explore modern AI.

1.6 AI Is a Family of Technologies

Think of AI as an umbrella. Under that umbrella are many different fields and techniques. A simplified hierarchy looks something like this:

ARTIFICIAL INTELLIGENCE │ ├── Symbolic AI │ ├── Machine Learning │ │ │ ├── Supervised Learning │ ├── Unsupervised Learning │ ├── Reinforcement Learning │ │ │ └── Deep Learning │ │ │ ├── Computer Vision │ ├── Speech │ ├── Natural Language Processing │ └── Generative AI │ │ │ ├── Large Language Models │ ├── Image Generation Models │ ├── Audio Models │ └── Multimodal Models │ └── Other AI techniques and hybrid systems

This diagram is deliberately simplified. The boundaries between these fields can overlap.

For example, modern systems can combine neural networks, search, symbolic reasoning, retrieval systems, databases, and other software components.

But the hierarchy gives you a useful mental model.

AI is the broad field.

Machine learning is one major approach within AI. Deep learning is a major approach within machine learning. Generative AI is a category of systems capable of generating content.

Large language models are one type of generative AI model. We will explore each of these concepts in detail throughout the Academy.

1.7 What Did AI Look Like Before Chatbots?

If you discovered AI through modern conversational systems, it can be easy to assume that AI began with chatbots. It didn't. AI has been studied for decades.

Researchers have attempted to build machines capable of reasoning, playing games, understanding language, recognising images, planning actions, and solving problems since long before today's generative AI explosion.

Earlier AI systems often relied heavily on explicitly represented knowledge, logical rules, search algorithms, or specialised statistical techniques.

One famous approach was expert systems. An expert system attempted to encode knowledge from human experts into a computer-readable set of rules. A simplified example might look like:

IF engine temperature is high AND coolant level is low

THEN possible cause = coolant leak

This approach could work surprisingly well in narrow domains. But it also had serious limitations. Real-world knowledge is enormous.

Situations can be ambiguous. Rules can conflict. People don't always know how to explicitly describe how they perform a task.

And manually encoding knowledge becomes extremely difficult as the number of possible situations increases.

Machine learning changed the equation by allowing computers to learn patterns from examples rather than requiring humans to manually encode every possible rule.

That shift eventually contributed to the modern AI revolution.

1.8 Why Did AI Suddenly Become So Powerful?

This is one of the most important questions in modern technology. AI has existed for decades. So why did the public suddenly experience an explosion of capable AI systems?

There isn't one single reason. Several technological developments came together.

More Data

Modern digital society generates enormous quantities of data. Text. Images.

Video. Audio. Code.

Scientific measurements. Web pages. Documents.

User interactions. This provided training material for increasingly sophisticated machine-learning systems.

More Computing Power

Modern processors can perform enormous numbers of mathematical operations.

Graphics processing units, or GPUs, became particularly important because their architecture is well suited to the highly parallel mathematical operations used by many neural networks.

We will spend an entire lesson studying CPUs, GPUs, RAM, and VRAM.

Better Neural Network Architectures

Researchers developed increasingly effective neural network architectures.

One of the most important developments was the Transformer architecture, introduced in the 2017 research paper "Attention Is All You Need."

Transformers became foundational to many modern language models. You don't need to understand Transformers yet. We will get there.

For now, understand that modern AI didn't simply appear from nowhere. It is the result of decades of research combined with improvements in:

• algorithms
• data
• hardware
• software
• training techniques
• model architectures
• engineering infrastructure

1.9 The Modern AI Explosion

The current AI era is dominated by increasingly capable foundation models and generative AI systems. The scale of this change is substantial.

Stanford's 2026 AI Index reports that industry produced more than 90% of notable frontier models in 2025, while AI adoption and model capabilities continued to increase rapidly. The report also notes that evaluation itself is becoming more difficult as models reach and exceed the capabilities of established benchmarks.

That last point is particularly important. AI progress isn't simply:

"Computers are getting better."

We are increasingly dealing with systems that can perform a broad range of tasks involving language, images, reasoning, coding, scientific problems, and other forms of information processing.

At the same time, this does not mean that AI has solved intelligence. Modern AI systems still have limitations. They can make factual errors.

They can generate convincing nonsense. They can misunderstand instructions. They can fail unexpectedly.

They can behave differently depending on context. They can struggle with tasks that appear trivial to humans. They can also perform extraordinarily well on tasks that humans consider difficult.

That combination is one of the defining characteristics of modern AI.

1.10 What Is Generative AI?

You will hear this term constantly.

Generative AI refers to AI systems that can generate new content.

The generated content might include:

• text
• images
• audio
• music
• video
• computer code
• structured data

A traditional classification system might receive:

Photograph → "Dog"

A generative system might receive:

"Write a description of a golden retriever"

and produce: A golden retriever is a friendly... The difference is that the system is not merely assigning a predefined label.

It is generating an output. Large language models are a particularly important category of generative AI.

Given a sequence of tokens, a language model can calculate probabilities over possible continuations and generate an output token by token.

This is something we will examine in extraordinary detail later.

1.11 What Is an AI Model?

This distinction is absolutely fundamental for understanding Loco.

An AI model is not the same thing as an AI application.

An AI model is, broadly speaking, a learned computational representation that can be used to perform a task. For a neural network, that representation is largely contained in its learned parameters or weights. Think of a model as the learned mathematical component.

The application around it is everything required to actually make that model useful to a person. A simplified AI system might look like this:

AI APPLICATION │ ┌──────────────┼──────────────┐ │ │ │ Interface Memory Tools │ │ │ └──────────────┼──────────────┘ │ Inference Engine │ AI Model │ Computer Hardware

The model is only one part. This distinction becomes critical when we start discussing local AI. A model file by itself doesn't magically become a conversational AI application.

Something has to:

1. load the model
2. allocate memory
3. process the user's input
4. convert text into the appropriate representation
5. execute the model
6. generate output
7. convert that output back into something readable
8. provide an interface for the user

Loco exists at the application/platform layer around local AI. That is why understanding the underlying technology matters.

1.12 AI Is a System, Not Just a Model

One of the biggest misconceptions among newcomers is:

"The AI is the model."

The model is important, but the complete AI experience can involve many other components. Imagine asking Loco:

"Read these five company documents and tell me which one contains our refund policy."

The language model itself may not contain the information in those documents. The complete system could involve:

The model is only one component. Later in the Academy, you will learn about:

• inference
• context windows
• embeddings
• retrieval
• RAG
• vector databases
• agents
• tool calling
• workflows
• APIs
• local servers
• model files
• inference runtimes

All of these pieces can combine to create an AI system.

1.13 What Makes an AI System Different From Normal Software?

This is a subtle question. Normal software can also produce sophisticated results. A spreadsheet is software.

A web browser is software. A video editor is software. A database is software.

So what makes AI different? One important distinction is how behaviour is produced. Traditional software often contains explicitly programmed logic.

For example:

IF customer_age < 18:
    reject application

The developer wrote the rule. A machine-learning model may instead contain millions, billions, or more learned numerical parameters. The developer doesn't necessarily write an explicit rule corresponding to every individual behaviour.

Instead, the model's behaviour emerges from patterns learned during training. That does not mean the model is mysterious magic. The mathematics can be studied.

The architecture can be studied. The training process can be studied. The parameters can be measured.

The outputs can be evaluated. But the behaviour is represented differently from a traditional collection of manually authored rules.

1.14 The Importance of Data

Data is one of the foundations of modern machine learning. Imagine trying to teach a child what a dog is. You might show the child:

• a Labrador
• a German Shepherd
• a Chihuahua
• a Great Dane
• a puppy
• an old dog
• a dog lying down
• a dog running
• a dog from different angles

Eventually, the child develops a flexible concept of "dog."

Machine learning does not learn in exactly the same way a child does, but the analogy helps illustrate the importance of examples.

A machine-learning model can be exposed to huge quantities of training data and adjusted repeatedly so that its outputs become more useful according to the training objective.

For language models, training involves enormous quantities of text and sophisticated mathematical optimisation. The details are much more complicated than:

"The AI reads the internet."

That phrase is an oversimplification.

Models are trained through carefully engineered datasets, objectives, optimisation procedures, filtering processes, architectures, and training infrastructure.

Later lessons will unpack these pieces.

1.15 Training vs Using an AI Model

Another foundational distinction:

Training is not the same as inference.

Training is the process through which a model's parameters are adjusted so that it learns useful patterns. Inference is what happens when the trained model is used to generate a prediction or output. Imagine a student preparing for an exam.

Studying for months is analogous, very loosely, to training. Sitting the exam is analogous, very loosely, to inference. Once a model has been trained, you don't normally retrain the entire model every time you ask it a question.

Instead, the trained parameters are loaded and used to process new input. This distinction is absolutely central to local AI. When you download a model and run it on your computer, you are generally performing inference.

You are not rebuilding the model from scratch.

We will dedicate an entire lesson to training versus inference because misunderstanding this concept causes enormous confusion.

1.16 Where Does Loco Fit?

Now we can start connecting the theory to the thing you actually care about.

Loco is a local AI platform.

That means Loco is concerned with making AI models useful on a user's own computer or local environment. The model is one component. The computer is another.

The inference technology is another. The user interface is another.

The surrounding features, configuration, model management, knowledge capabilities, APIs, workflows, and other functionality form additional parts of the overall system.

A simplified conceptual view is:

LOCO │ ┌────────────┼────────────┐ │ │ │ Models Interface Features │ │ │ ├─────── Inference ───────┤ │ │ └──────── Hardware ───────┘

The exact architecture can become considerably more sophisticated. But this mental model is enough for now. The important point is:

Loco is not the model. Loco is the platform that helps turn local AI models into something a person can actually use.

This distinction will become more important with every lesson.

1.17 What Does "Local" Actually Mean?

You will hear the phrase local AI throughout this course.

"Local" generally means that important AI computation is performed on hardware controlled by the user or organisation rather than requiring every inference request to be processed by a remote cloud service.

For example:

A local architecture can instead look more like:

This does not mean that every part of every local AI product is necessarily offline. A system may still use the network for things such as:

• downloading models
• checking updates
• accessing websites
• sending email
• calling external APIs
• synchronising information
• connecting to external services

The important distinction is where the AI inference is taking place. That is one of the core ideas behind Loco.

1.18 Why Does Local AI Matter?

There are several reasons an individual or organisation might want AI running locally.

Privacy

If sensitive information is being processed locally, the organisation can potentially keep that information within its own environment rather than sending the inference workload to an external AI provider.

This does not automatically make a system perfectly private. Security still matters. Operating-system security matters.

Network configuration matters. Application security matters. User permissions matter.

Data storage matters. But local processing can change the data-flow architecture significantly.

Control

With local AI, the user can have substantially more control over:

• which model is used
• when it runs
• where it runs
• what hardware performs inference
• what data is provided to it
• which software surrounds it
• whether external services are involved

Availability

A local system can potentially continue operating when an internet connection is unavailable, provided that the required models and software are already present locally and the particular workflow does not depend on external services.

Cost Structure

Cloud AI typically involves remote infrastructure. Local AI shifts some of the infrastructure requirement onto the user's hardware. That creates a different economic model.

Instead of every inference request necessarily consuming remote compute, the user can use hardware they already own. However, local AI is not automatically free. Hardware costs money.

Electricity costs money. Storage costs money. Maintenance costs money.

And sufficiently capable local hardware can be expensive. The point is not that local AI eliminates costs. It changes where those costs occur.

1.19 Does Local AI Mean You Need a Giant Computer?

No. This is one of the biggest misconceptions newcomers encounter. AI models come in many different sizes.

Some models require enormous amounts of memory and computational resources. Others are designed to operate within much more modest hardware constraints.

The size of a model, the way it is stored, its numerical precision, its architecture, its context length, the workload being performed, and the hardware being used can all affect the resources required.

This is why you will eventually learn about:

• parameters
• model weights
• quantization
• RAM
• VRAM
• CPU inference
• GPU inference
• context length
• memory requirements
• tokens per second

A computer does not need to be a data centre simply because you want to run AI. The practical question is:

What model are you trying to run, and what does that model require?

That is a much better question than:

"Is my computer powerful enough for AI?"

1.20 AI Does Not Automatically Mean Human-Level Intelligence

This deserves its own section because science fiction has influenced public understanding of AI enormously. An AI system does not have to be equivalent to a human mind to be useful. A calculator is not intelligent in the human sense, but it performs arithmetic extremely well.

A navigation system does not understand geography like a human traveller, but it can calculate routes extremely effectively.

A chess engine does not need human consciousness to play chess at an extraordinary level.

Similarly, modern AI models can perform specific classes of tasks at remarkable levels without being human beings in digital form.

The phrase artificial general intelligence, or AGI, refers to a much broader concept involving systems capable of general-purpose intellectual abilities across many domains.

There is substantial debate about what AGI should mean, how it should be measured, and whether current systems qualify. For the purposes of Loco Academy, you do not need to resolve that debate yet. The important distinction is:

Today's AI systems can be extremely capable without necessarily representing a complete artificial human intelligence.

1.21 Why AI Can Feel Like Magic

A useful way to understand the modern AI experience is to consider the difference between the complexity of the mechanism and the simplicity of the interface.

You type:

"Explain quantum mechanics like I'm twelve."

You receive a detailed explanation in seconds. The interface looks simple. But behind that request can be an enormous computational process involving:

• tokenisation
• numerical representations
• neural-network layers
• matrix operations
• attention mechanisms
• model parameters
• probability distributions
• sampling
• context processing
• memory management
• hardware acceleration
• inference software

The interface hides almost all of this. That's good for usability. But it can create a dangerous misconception.

People start believing:

"AI is basically magic."

It isn't. It is extraordinarily sophisticated engineering and mathematics. The purpose of Loco Academy is to pull back the curtain.

By the end of this course, you should be able to look at a local model and understand what is actually happening inside the computer when you press Generate.

1.22 AI Has Real Limitations

A serious AI education cannot only discuss capabilities. AI systems have limitations. These can include:

Hallucination

A generative model may produce information that sounds plausible but is incorrect.

Bias

Models can reproduce patterns and biases present in their training data or introduced through their development and deployment.

Distribution Problems

A model may perform well on familiar types of inputs but behave poorly when given unusual inputs.

Context Limitations

Models cannot necessarily consider unlimited amounts of information at once. Their context capacity is constrained by the architecture and implementation.

Hardware Constraints

Local models are constrained by available computational resources.

Evaluation Problems

A model scoring well on one benchmark does not guarantee that it will perform well for your particular task.

Stanford's 2026 AI Index specifically highlights growing concerns about benchmark reliability and reports that some widely used evaluations contain substantial rates of invalid questions.

This is an important lesson:

Never confuse a benchmark score with universal intelligence.

A benchmark measures something. It does not measure everything.

1.23 AI Is a Tool, Not an Oracle

This is perhaps the most important mindset you can develop as an AI user. Do not think:

"The AI knows."

Think:

"The AI generated this output based on its model, input, context, training, system configuration, and available information."

That sounds like a small linguistic difference. It isn't. It changes how you use the technology.

If an AI system gives you a medical claim, you verify it. If it gives you a legal interpretation, you verify it. If it gives you code, you test it.

If it gives you financial information, you verify the underlying facts. If it gives you a historical claim, you check the source. AI can dramatically accelerate knowledge work.

But acceleration does not eliminate the need for judgement.

1.24 The AI Stack You Are About to Learn

Before moving to the next lesson, I want you to understand the conceptual journey of the entire Local AI 101 course. You are going to move through the stack from the bottom upward.

Level 1: Hardware

You will learn about:

• CPUs
• GPUs
• RAM
• VRAM
• storage
• memory bandwidth
• compute

Level 2: Model Files

You will learn about:

• parameters
• weights
• precision
• quantization
• GGUF
• model sizes

Level 3: Model Architecture

You will learn about:

• neural networks
• Transformers
• attention
• embeddings
• layers
• context

Level 4: Inference

You will learn:

• what inference actually means
• how tokens are processed
• how output is generated
• why generation speed varies
• why hardware matters

Level 5: Applications

You will learn how software such as Loco turns the underlying technology into something a human can actually use.

Level 6: Knowledge

You will learn:

• RAG
• embeddings
• document processing
• retrieval
• knowledge bases

Level 7: Agents

You will learn:

• tools
• tool calling
• planning
• workflows
• APIs
• external actions

Level 8: Deployment

You will learn how local AI can become useful for:

• individuals
• developers
• businesses
• teams
• private environments

By the end, "running an AI model locally" should no longer sound mysterious. It should sound like an engineering problem. And engineering problems can be understood.

1.25 The Most Important Mental Model

If you remember only one thing from this lesson, remember this:

AI │ ├── Models │ └── Learned mathematical representations │ ├── Software │ └── Loads and executes those models │ ├── Hardware │ └── Performs the required computation │ ├── Data │ └── Provides information for training or use │ └── Applications └── Turn the technology into useful experiences

Loco sits primarily at the application and local AI platform layer, while relying on the broader ecosystem of models, inference technologies, operating systems, hardware, and open technical standards underneath it.

Understanding these layers is going to make everything else easier. When you later hear somebody say:

"This is a 14-billion-parameter model running in Q4 quantization with 8K context on CPU inference."

you should eventually be able to understand every part of that sentence. You won't need to nod and pretend you know what they are talking about. You will actually know.

1.26 Key Takeaways

Let's consolidate the lesson.

1. AI is a field, not a single technology.

Artificial intelligence encompasses many different approaches to building systems capable of performing tasks associated with intelligence.

2. AI does not necessarily mean human-like thinking.

A system can perform sophisticated tasks without possessing human consciousness or subjective experience.

3. Machine learning is one major approach within AI.

Machine learning allows systems to learn patterns from data rather than requiring every rule to be manually programmed.

4. Modern AI is built from multiple technologies.

Data, algorithms, neural networks, models, hardware, inference software, and applications all contribute to the final system.

5. A model is not the same thing as an application.

The model is one component of an AI system.

6. Training and inference are different.

Training produces or modifies the learned model. Inference uses that trained model to generate outputs.

7. Generative AI produces new content.

This can include text, images, audio, video, code, and other forms of data.

8. Local AI means important AI computation can happen on local hardware.

The exact architecture varies, and local systems can still use network connections for other functions.

9. Local AI is not synonymous with massive hardware.

Different models have radically different resource requirements.

10. AI has limitations.

Models can make mistakes, hallucinate, reproduce biases, misunderstand instructions, and behave unpredictably in some situations.

11. Loco exists within this larger technology stack.

Understanding the underlying technology helps you understand what Loco is doing when it loads models, performs inference, manages knowledge, and provides AI capabilities locally.

1.27 Knowledge Check

Before moving on, try answering these questions without looking back.

Question 1

What is artificial intelligence?

Answer:
AI is a broad field involving engineered systems capable of performing tasks such as prediction, recommendation, decision-making, perception, language processing, generation, or other forms of problem-solving that can be associated with intelligence.

Question 2

Is AI one specific technology?

Answer:
No. AI encompasses many different techniques, architectures, algorithms, models, and systems.

Question 3

What is the difference between traditional programming and machine learning?

Answer:
Traditional programming generally relies on explicitly written rules and instructions. Machine learning allows a system to learn patterns from data through a training process.

Question 4

Is an AI model the same thing as an AI application?

Answer:
No. A model is one component of an AI system. An application can provide interfaces, model management, memory, retrieval, tools, workflows, APIs, and other functionality around the model.

Question 5

What is inference?

Answer:
Inference is the process of using a trained model to produce an output from new input.

Question 6

Does local AI necessarily mean the entire application never connects to the internet?

Answer:
No. A local AI system can perform inference locally while still using network connectivity for things such as downloads, updates, external APIs, web access, synchronisation, or other configured services.

Question 7

Does a powerful AI model necessarily understand the world exactly like a human?

Answer:
No. Capability and human-like understanding are not the same thing.

1.28 A Preview of the Next Lesson

Now that you understand what AI is, we can start taking apart the terminology that gets thrown around constantly. You've probably heard:

AI

Machine Learning

Deep Learning

Neural Networks

Generative AI

Large Language Models

These terms are related, but they are not interchangeable. In the next lesson, we'll build the hierarchy properly. You'll learn exactly how:

Artificial Intelligence → Machine Learning → Deep Learning → Generative AI → Large Language Models

fit together, where the boundaries become blurry, and why people often use these terms incorrectly. Once that foundation is established, we can start descending into the actual machinery that makes modern AI possible.

Sources and Further Reading

• National Institute of Standards and Technology, Artificial Intelligence Risk Management Framework (AI RMF 1.0): https://www.nist.gov/itl/ai-risk-management-framework
• Google for Developers, Machine Learning Glossary: https://developers.google.com/machine-learning/glossary
• Stanford HAI, 2026 AI Index Report: https://hai.stanford.edu/ai-index/2026-ai-index-report
• Vaswani et al., Attention Is All You Need: https://arxiv.org/abs/1706.03762

Loco Academy: Lesson Status

Completed: What Is Artificial Intelligence?

Next: AI vs Machine Learning vs Deep Learning

Course: Local AI 101