AI, ML, DL and GenAI explained

A simple story from someone trying to make sense of AI without the tech details.

I’m Eglė, and I’m a CFO. I used to work in traditional businesses, from manufacturing to airlines, but now I’ve been swept into the world of generative AI at Moterra. I like to understand things from the ground up: numbers must add up, and structure matters. So I don’t just want to know how to prompt or navigate the AI landscape, I want to understand how it actually works and why.

The trouble is, AI discussions often jump straight to tools, mixing AI with GenAI and confusing models with apps. It gets messy, fast. Most content is either too technical or too surface-level. When that happens, I open Excel, fix the vocabulary, and map the logic. I’ve done exactly that for AI, and I’m sharing it here as a short series. If you want the foundations - clear, useful, and without drowning in detail - join me.

The AI map in one line

Artificial Intelligence (AI) → Machine Learning (ML) → Deep Learning (DL) → Generative AI (GenAI)

Goal → Learn → Learn deeply → Create

When the terminology gets muddled, I open Excel, clarify the vocabulary, and map the logic. Here is exactly what that looks like:

Table mapping the AI layers: foundation model providers and their LLMs, infrastructure and orchestration platforms, and business applications

This table is my reference point. Now let me walk you through what each layer actually means in practice.

How we reached GenAI: the short story

Early AI was mostly rules: if X, then Y. Useful, but brittle. Machine learning arrived and let systems learn from examples, spotting patterns we didn’t hard-code. With more data and computing power, deep learning took off: multi-layer neural networks that capture rich structure in language, images, and audio. From there, generative AI emerged - models that don’t just label things, they produce new content.

What each layer does: the strategic view

Artificial intelligence: the strategic layer

AI represents the broadest ambition: software handling tasks that typically require human judgment. While the table shows it solves problems, what matters for business is how it solves them. AI systems can adapt to new scenarios and improve over time, making them valuable for high-level decisions that involve multiple variables and uncertainty.

Machine learning: the pattern engine

The table shows ML learns from historical data, but here is what that means practically: instead of programming every possible rule, you show the system examples and let it develop its own logic. This is why ML excels at forecasting and classification. It spots patterns humans might miss or cannot process at scale.

Deep learning: the complexity handler

While the table notes DL processes unstructured data through neural networks, the real breakthrough is what it can handle. Text documents, images, audio files, video - all the messy, real-world data that traditional systems struggle with. This is why DL revolutionised fields like medical imaging and natural language processing, which means understanding everyday human language, not code.

Generative AI: the content creator

The table shows GenAI creates new content from prompts, but the transformation goes deeper. Unlike previous AI that analysed existing information, GenAI produces original outputs. This shifts the conversation from what does this data tell us, to what can we create with this understanding.

How the layers work together: the flow I actually use

Here is where it gets practical. I treat this as one AI-powered accounts payable system that bundles multiple capabilities. When I upload a scanned invoice, the system orchestrates:

AI foundations Basic validation rules check things like VAT format and totals.

Machine learning Classifies the supplier and routes to the right cost centre and workflow.

Deep learning OCR and data extraction pull names, dates, line items and amounts.

Generative AI Drafts a concise payable note and, if needed, a polite vendor email.

All of it finished with a quick human review before approval.

How generative AI works, in plain English

When I paste text in, it is split into small pieces called tokens. The model looks at the tokens I have provided - the context - then predicts the most likely next token, again and again, until it forms a paragraph, a summary, or a reply.

The mind-blowing part? It is not actually thinking or understanding. It is getting incredibly good at guessing what word comes next, based on patterns from massive amounts of text.

GenAI today: what it is great at, what it is not, and why

Great at, and how I use it

First drafts, fast Emails, summaries, outlines, short code. I turn messy meeting notes into clear minutes, then fact-check.

Transformations Rephrase tone, translate, tidy, extract key fields. I adapt a proposal intro for a CFO audience in seconds.

Explaining and structuring Shape scattered notes into a plan. I produce a one-screen brief before a stakeholder call.

Not great on its own

Guaranteed facts It predicts plausible text; it does not verify truth by default. I always check numbers against our sheets.

Forecasting and scoring Classic machine learning often outperforms on numeric tasks. If I need a forecast, I use ML, not GenAI prose.

Replacing guardrails Privacy, approvals and audit still matter. I keep sensitive data inside a private setup and review outputs.

Why the limits exist. Generative AI is predictive generation at scale. It excels at form and fluency. Accuracy improves when paired with the right information and sensible checks.

My takeaway

Artificial intelligence is the foundation. Machine learning is the method. Deep learning is the engine. Generative AI is what we’ve built.

Next step

Keep sensitive data inside a private setup.

Moterra deploys Claude inside your own cloud, so your teams get generative AI without your security team having to say no.

AI map simplified: understanding AI, ML, DL and GenAI