AI · LESSON 03 / 25 · English
Types of AI: Narrow, General & Super
Describe capabilities before using big labels.
Content reviewed: 22 October 2026 · தமிழில் படிக்க ↗
Understand it in plain language
Task-specific AI is built or evaluated for defined work. General-purpose models can serve many tasks, but that does not make them reliable at every task. AGI and superintelligence are debated concepts with varying definitions; they should not be treated as certified product categories or assigned a guaranteed arrival date. Another useful distinction is predictive systems versus generative systems: one estimates an outcome, another creates a candidate output.
How the AI approach works
Different AI taxonomies answer different questions. Capability labels ask how broadly a system can perform: task-specific systems focus on bounded work, while general-purpose models can support many tasks. AGI and superintelligence concern much broader capabilities, but definitions and proposed tests vary. A vendor’s label is not a scientific certification of general intelligence.
Approach labels ask how the system works. Symbolic methods manipulate explicit representations; probabilistic methods reason with uncertainty; learning methods estimate patterns from experience. A system may combine these. “Reactive” and “memory-based” labels describe behavior or information use, not an inevitable ladder that every model climbs toward consciousness.
Learning paradigms describe the feedback used during training. Supervised learning uses target labels; unsupervised learning looks for structure without those supplied targets; self-supervised learning constructs prediction targets from the data itself; reinforcement learning learns action selection from reward over interaction. Deep learning describes a model family and can be used across several of these paradigms.
Output labels form another axis. A classifier predicts a category, a regressor predicts a number, a generator creates a candidate sequence or image, and a planner selects actions. Do not force all systems into one mutually exclusive list: an application can classify a request, retrieve evidence, generate an explanation, and execute a plan.
Break the idea into steps
- Write down the task and the conditions under which it was tested.
- Check which input and output types are supported: text, images, audio, or other data.
- Judge performance using your own representative examples, not a broad label.
Worked example: connect the method to the task
A Tamil customer-support assistant may use a pretrained neural language model. Its pretraining may be self-supervised, its support adaptation may use supervised examples, and its application may add retrieval and tool calls. “Deep learning,” “self-supervised,” “generative,” and “agentic” describe different aspects; none alone explains the whole system.
Test the claimed breadth with separate tasks and realistic constraints. Good Tamil conversation does not prove reliable arithmetic, current policy knowledge, or safe account changes.
Apply the idea: check your reasoning
Think before revealing the answer
Can a deep neural network be trained with supervised learning?
Reveal the answer and reasoning
Yes. “Deep” describes the architecture; “supervised” describes the training signal. An image classifier can be both a deep-learning model and a supervised-learning model.
A common mistake: A model supporting many tasks does not establish human-level competence or consciousness.
Before you move on
- Explain the main idea without repeating the technical label.
- Identify the approach, its assumptions, and the evidence needed to trust the result.
- Complete the challenge, then compare your reasoning with the answer.
Sources & a deeper next step
The refresh uses primary papers and official documentation. These optional links are in English; you can understand this lesson without reading them first. Product versions and availability can change.