AI Strategy6 min read

Types of AI Explained: Predictive, Generative, Vision and More

Part 2 of a plain-English AI terms series: deterministic vs probabilistic, predictive machine learning, generative AI, chatbots, recommendation AI and computer vision.

Haojun See

Founder & Director, On The Ground

Updated 22 October 2026

Why the type of AI matters

AI is not one technology. It is a family of tools that do different jobs, and picking the wrong one is a common reason AI projects disappoint. This is part 2 of our plain-English series on AI terms. It explains the main types of AI so you can tell which one a vendor is actually selling. The terms follow the vocabulary of the Singtel AI.dea – AI bridge programme, run with SIM Academy (Programme Architect: Michelle Yeo). The explanations are our own. Part 1 covers hyperscalers, observability and AI agents.

Deterministic vs probabilistic: the idea behind every AI type

Deterministic software gives the same output for the same input, every time. Probabilistic AI gives its best estimate, which can vary and is sometimes wrong. Your accounting software is deterministic: if it adds up GST differently on a second run, something is broken. An AI model is probabilistic: ask it the same question twice and you may get two different, both plausible, answers. This changes how you judge the tool. Deterministic software is right or wrong. AI is judged on how often it is right, how wrong it is when it misses, and whether a person catches the misses. Ask: if this tool gets an answer wrong, what does it cost us, and who will notice?

Predictive (traditional) machine learning

Predictive machine learning studies past data to estimate what is likely to happen next. It can predict which invoices are likely to be paid late, how many staff a clinic needs on a Saturday morning, or which machine part is due to fail. It has been in business use for years, which is why it is sometimes called traditional machine learning. Its output is usually a number with a probability attached: a 65% chance an invoice is paid late, or about 900 orders expected next week. The risk is that people read the number as fact rather than odds, and that the model assumes the future will look like the past, which breaks when the market shifts suddenly. Ask: what data was this trained on, and how have its past predictions compared with what actually happened?

Generative AI and large language models (LLMs)

Generative AI creates new content, such as text, images, code or audio, from an instruction written in plain language. The text tools, including ChatGPT, Claude, Gemini and Microsoft Copilot, run on large language models (LLMs): models trained on huge amounts of text to predict the most likely next words. That is why they write fluently and also why they can state things that aren't true. They are strongest at first drafts, summaries, rewrites and turning rough notes into structured documents. They are weakest wherever an exact figure or a final, high-stakes decision is needed. Ask: which outputs go to customers or regulators, and who checks them before they do?

Conversational AI and chatbots

Conversational AI handles dialogue by text or voice: answering questions, booking appointments and guiding people through a process. Older chatbots follow a menu of set questions and answers. Newer ones run on LLMs and can handle questions they were never explicitly set up for. That flexibility cuts both ways. A chatbot on your website speaks for your business, and customers may rely on what it says, even when it is wrong. Ask: what does the chatbot do when it doesn't know the answer, and how quickly can a customer reach a person?

Recommendation and personalisation AI

Recommendation AI predicts what each person is most likely to want next and shows it to them. Think of a bakery's ordering site suggesting the add-ons regulars usually buy, or a hardware supplier's catalogue showing the parts that go with the one you are viewing. For SMEs it usually arrives switched on inside software they already pay for, rather than as a project of its own. It runs on customer behaviour data, so it raises questions about consent and how that data is used. Ask: what customer data does this use, and did customers agree to that use?

Computer vision

Computer vision lets software understand images and video: what is in the picture, where, and whether anything looks wrong. In business it can read a supplier's delivery order into your system, flag dented cartons as they come off a truck, check that a delivery photo shows the parcel at the right door, or read old analogue meters. Accuracy depends heavily on real conditions: lighting, camera angle and how closely your images match the ones the model learned from. Ask: has this been tested on images from our premises, under our lighting, rather than on the vendor's demo set?

Matching the type of AI to the job

The quickest way to avoid a failed pilot is to name the job first, then pick the type of AI. A simple guide: • A fixed rule or calculation, such as GST or a sales commission formula: use ordinary software. No AI needed. • A number or a likelihood, such as next month's demand or which invoices will be paid late: predictive machine learning. • A draft, a summary or an answer in words: generative AI, with a person checking. • Something in an image or video: computer vision. • A multi-step task across several systems: an AI agent, covered in part 1. Ask: what exactly is the output we need, and is AI really the simplest way to get it?

Next in the AI terms series

Part 3 covers AI risk terms: hallucination, workslop, shadow AI and the main ways AI systems are attacked. The series ends with AI strategy terms and AI architecture and governance terms. To learn these with your own team, you can register your interest on the AI.dea programme page.

Frequently asked questions

What is the difference between AI and generative AI?

AI is the broad field of software that learns from data to classify, predict or decide. Generative AI is one type of AI that creates new content, such as text, images or code, from plain-language instructions. A fraud alert uses predictive AI; a tool that drafts your proposal uses generative AI. Many business tools now combine both.

What does probabilistic mean in AI?

Probabilistic means the AI gives its best estimate based on patterns in data rather than a guaranteed answer. The same question can produce different answers, and some will be wrong. Ordinary software is deterministic: the same input always gives the same output. This is why AI outputs need checking in a way spreadsheet formulas do not.

Which type of AI should a small business start with?

Start with the job, not the technology. Most SMEs get the quickest value from generative AI for drafting, summarising and answering questions from their own documents, because it needs no data project. Predictive AI suits firms with years of clean sales or operations data. Use ordinary software wherever a fixed rule will do.

AI Strategy

AI Terms for Leaders: Hyperscalers, Agents and Observability

AI Strategy

AI Strategy Terms: Ambition, Readiness, Quick Wins and RICE

AI Strategy

AI Risk Terms: Hallucination, Workslop, Shadow AI and More

PreviousInside Our AI Readiness Audit: The Research Behind the Score (2026)Next AI Terms for Leaders: Hyperscalers, Agents and Observability