What is artificial intelligence?
Artificial intelligence (AI) is software that performs tasks which normally require human judgment: recognizing what is in an image, reading and writing language, forecasting demand, or deciding the next step in a process. The term covers a family of techniques rather than a single technology. Some systems follow rules written by people. Most modern systems learn their behavior from data instead, which is the branch called machine learning.
In practice, AI is a model (a mathematical function with learned parameters) wrapped in ordinary software. Data goes in, a prediction, classification, or generated text comes out, and surrounding code decides what to do with it. A spam filter, a credit scoring model, a warehouse vision camera, and a large language model answering questions about a contract are all AI by this definition. They differ in scale and technique, but each one maps inputs to outputs using patterns learned from examples.
Types of artificial intelligence
Business readers meet four broad approaches. They are often combined in one product.
| Approach | How it decides | Typical business use |
|---|---|---|
| Rule-based (symbolic AI) | Follows logic written by experts | Eligibility checks, tax rules, configuration validation |
| Classic machine learning | Learns statistical patterns from structured data | Credit risk, churn prediction, demand forecasting |
| Deep learning | Learns features from raw data with neural networks | Visual inspection, speech recognition, document reading |
| Generative AI | Produces text, code, or images from learned language patterns | Drafting, summarizing, answering questions over documents |
A second distinction matters when reading vendor material. Narrow AI performs one task or a related set of tasks, and every system in production is narrow. Artificial general intelligence, a system that could learn any intellectual task a person can, remains a research goal.
Why artificial intelligence matters
AI changes which work can be automated. Traditional software automates tasks with fixed rules, such as posting an invoice. AI extends automation to tasks with variation and ambiguity: reading a supplier invoice in any layout, deciding which collections accounts to call first, or answering a customer in plain language. For a company with 50 to 5,000 employees, that usually means removing hours of repetitive reading, matching, and routing from skilled staff.
It also changes risk. A model’s output is a probability, and it will sometimes be wrong in ways a rule never is. Every business deployment therefore needs to decide how much authority the system has, who checks its output, and how errors are caught. The glossary entries on AI governance, human-in-the-loop, and AI system cover those decisions.
Artificial intelligence in practice
A regional distributor receives several thousand purchase orders a month by email, as PDFs, spreadsheets, and free text. A document model extracts line items, a matching model links them to catalog SKUs, and an agent enters clean orders into the ERP while routing anything with low confidence to a person. A hospital network uses a forecasting model to predict admissions by unit a week ahead so that staffing can be adjusted. A law firm runs a language model on its own servers to find relevant clauses across thousands of pages of discovery material. None of these systems is intelligent in a general sense, and each one is valuable because it is narrow and measured.
Artificial intelligence and private AI
Where AI runs is a business decision. Models can be rented through an API, where every prompt and document leaves the company, or operated on infrastructure the company controls, using open-weight models. Companies with sensitive data or steady volume often find that owning the capability gives them control over cost, data residency, and auditability. The entries on private AI and open-weight model explain the trade-offs.
What to watch for
- Ask what kind of AI a product actually uses, because “AI-powered” can describe a lookup table or a frontier language model.
- Insist on measured accuracy on your own data, since results on a vendor’s demo set rarely transfer directly.
- Identify the systems of record the AI can change and require a human approval step wherever the cost of an error is high.
- Plan for maintenance from the start, because models degrade as the business and its data change.
- Check where data is processed and stored, and whether that location satisfies your contracts and regulators.