In plain English
AI categories are useful only when they clarify what a system actually does. The most important distinction is between today’s specialized systems, which are built for particular tasks, and hypothetical human-level general intelligence, which has not been demonstrated.
The main ideas
Narrow or specialized AI
Nearly all deployed AI is narrow: it recognizes speech, recommends products, forecasts demand, generates text, or performs another defined family of tasks.
Predictive AI
Predictive systems estimate a label, score, event, or future value from existing data. Examples include fraud alerts, demand forecasts, and risk scores.
Generative AI
Generative systems create new text, images, audio, code, or other media by learning patterns in training material.
General and superintelligent AI
Artificial general intelligence and superintelligence are theoretical labels for capabilities beyond today’s systems. Claims about them should be separated from evidence about existing tools.
How it works
- 1
Start with the outcome
Ask whether the system predicts, classifies, ranks, recommends, generates, controls, or optimizes something.
- 2
Examine the operating boundary
Determine which inputs, environments, users, and decisions the system was designed to handle.
- 3
Separate capability from autonomy
A system may produce sophisticated output while still requiring people to initiate tasks, approve actions, and handle exceptions.
- 4
Avoid marketing categories
Product labels such as intelligent, cognitive, or agentic do not by themselves explain the system’s limits or reliability.
Where you may see it
Classification systems
Email filtering, document sorting, image recognition, and medical-image support classify inputs into categories.
Recommendation systems
Streaming, shopping, social, and news services rank items based on learned patterns and business rules.
Generative assistants
Text and media generators help create drafts, options, explanations, and simulations.
Control and planning systems
Robotics, traffic control, scheduling, and industrial systems may combine prediction with rules and feedback loops.
Important limits
- One system can fit several categories at once.
- A category name does not establish accuracy, safety, or intelligence.
- Human-like conversation can make a narrow system appear more general than it is.
- Future-oriented labels are often used more confidently than the evidence supports.
A practical reader checklist
- Describe the system by its actual task.
- Ask what happens outside its tested conditions.
- Distinguish generated content from verified information.
- Check whether a person approves consequential actions.
Key takeaway
AI categories are useful only when they clarify what a system actually does. The most important distinction is between today’s specialized systems, which are built for particular tasks, and hypothetical human-level general intelligence, which has not been demonstrated.