Short answer: AI is the umbrella; machine learning and generative AI are two branches under it that do opposite jobs. Machine learning predicts a number or category from your history — the angel’s share, bottling maturity. Generative AI produces words — a draft, an explanation. The test for any distillery problem is simple: what shape is the answer? A number means machine learning, words mean generative AI, and “just show me what’s already there” means a dashboard, which often isn’t AI at all and is frequently the cheapest right answer.
This is part four of the foundations series. The previous three posts defined each term on its own — AI, machine learning and generative AI. This one puts them in the same room, because the expensive mistakes happen at the boundaries: buying a chatbot to do a forecasting job, or a forecasting project to do a reporting job.
The relationship in one breath
AI is the umbrella word. Underneath it, machine learning learns patterns from data to predict — a number, a category, a yes/no. Also underneath it, generative AI produces new content — words, tables, images. They are siblings, not stages: generative AI is not “more advanced” machine learning, and machine learning is not “old” AI. They do different jobs. And sitting outside the umbrella entirely are dashboards and rules — logic you write, no learning involved — which get called AI constantly and usually shouldn’t be.
The test: what shape is the answer?
Forget the vocabulary for a moment and look at the problem. The shape of the answer you need tells you the tool:
- You need a number or a yes/no predicted from history — “what loss will this cask take by year ten?”, “is this cask likely to drift off-profile?” That’s machine learning.
- You need words — a draft, an explanation, a summary, a structured note. “Write the maturing-stock narrative”, “explain why floor position matters”. That’s generative AI.
- You need to see or flag what’s already in the data, by a rule you can state — “show every cask past twelve years”, “total this month’s regauges”. That’s a dashboard, no AI required, and frequently the cheapest correct answer.
Most real distillery questions sort cleanly into one of these in a single sentence. When one doesn’t, it’s usually because it’s really two questions — predict the bottling window (machine learning), then draft the bottling plan (generative AI) — and naming the two halves is exactly the clarity that stops you overpaying for one tool to do both jobs badly.
Why the boundary mistakes are expensive
Ask generative AI to predict a cask’s evaporative loss and it will happily produce a confident figure — invented, because predicting from your specific history is not what it does. Ask a machine-learning model to write your board report and it can’t; producing language isn’t its job. Each failure is quiet: you get an answer, fluent or numeric, and only later discover it came from the wrong kind of machine. Matching the tool to the answer-shape up front is the whole defence — and it costs nothing but a moment’s thought.
Where this breaks
The honest caveats. The line blurs in real products — a modern copilot is generative AI on the surface, calling a machine-learning model underneath, reading from a dashboard’s data. You may be using all three through one chat box, which is fine; the point of the test is to know which part is answering so you know how much to trust it. “Just a dashboard” can be the hardest part — the unglamorous rules layer is often where the real value and the real work sit, and calling it un-sexy doesn’t make it optional. And the oldest caveat in this series: none of the three works on bad records. The test tells you which tool fits; it can’t conjure the measured data all three stand on.
The bottom line
AI is the umbrella; machine learning predicts numbers from your history; generative AI writes words; dashboards show what’s already there and usually aren’t AI at all. Match the tool to the shape of the answer — number, words, or “just show me” — and you’ll route most distillery problems correctly in one sentence, and stop paying buzzword prices for the wrong machine. The last post in the series brings it home: where a distillery actually starts — the order of operations that makes all three pay off.
Frequently asked questions
What is the difference between AI, machine learning and generative AI? AI is the umbrella term for software doing tasks that used to need a person. Machine learning is the branch that predicts a number or category from your historical data. Generative AI is the branch that produces new text or images in plain language. Machine learning predicts from data; generative AI writes; both sit under the AI umbrella.
Which one does my distillery problem need? Ask what shape the answer is. A number or yes/no predicted from history is machine learning. Words, a draft or an explanation is generative AI. Seeing or flagging what’s already in your data by a rule you can state is a dashboard — not AI at all, and often the cheapest right answer.
Is generative AI better than machine learning? Neither is better — they solve different problems. Generative AI can’t reliably predict a cask’s evaporative loss; machine learning can’t draft your board report. Using the wrong one is the common, expensive mistake. Match the tool to the shape of the answer you need, not to whichever term is loudest this quarter.