What Is Artificial Intelligence? – Narrow AI, General AI, and the Myth of "Thinking" Machines
You hear "artificial intelligence" everywhere – your phone's face unlock, Netflix recommendations, ChatGPT. But what does the term actually mean? Is AI truly "intelligent" like a human? In this post, we will cut through the hype, explain the formal definition of AI, and look at the difference between Narrow, General, and Superintelligence.
1. The Standard Definition
AI is a field of computer science focused on creating systems that perform tasks that normally require human intelligence: recognizing patterns, making decisions, understanding language, and learning from experience. Most modern AI focuses on rational action – achieving the best outcome given available information, regardless of whether the internal process resembles human thought.
2. The Three Levels of AI
A. Artificial Narrow Intelligence (ANI) – The Only AI That Exists
ANI is designed to perform one specific task extremely well. It cannot transfer its skills to other domains.
|
System |
Task |
Can it do something else? |
|
AlphaGo |
Play Go |
No – it cannot play chess or drive a car. |
|
ChatGPT |
Generate text based on patterns |
No – it cannot reliably calculate math or recognize faces. |
|
FaceID on iPhone |
Recognize your face |
No – it cannot listen to voice commands. |
Key insight: Even the most impressive AI today (GPT-4, Gemini, Claude) are still Narrow AI. They are incredibly good at language, but they have no consciousness, no desires, and no common sense outside their training data.
B. Artificial General Intelligence (AGI) – The Holy Grail
AGI would be an AI system that can perform any intellectual task that a human can – learning across domains, applying common sense, reasoning abstractly, adapting to new situations without retraining. No AGI exists today. Experts disagree on whether current approaches (large neural networks) will ever achieve it.
C. Artificial Superintelligence (ASI) – Science Fiction (for now)
ASI would be an intellect vastly superior to the best human minds in every field. It is purely hypothetical. You cannot "accidentally" create ASI from ANI by scaling up – the leap is qualitative, not quantitative.
3. How Today's AI Works – The Statistical Core
Most modern AI is built on machine learning. Instead of programming explicit rules, the system learns patterns from data. The three ingredients:
- Data: Thousands, millions, or billions of examples.
- Model: A mathematical function with many adjustable parameters (weights). A neural network can have billions of parameters.
- Training algorithm: An optimization process (like gradient descent) that adjusts the model's parameters to reduce error on the training data.
The model has no understanding of meaning. It only knows statistical relationships from training data.
4. Why AI Is Not "Thinking" – The Chinese Room Argument
Philosopher John Searle imagined a person in a room with a rulebook for responding to written Chinese characters with other Chinese characters. From outside, the room appears to understand Chinese. But the person inside does not understand a word – they are just following rules.
Searle argues that today's AI is like that room. It manipulates symbols based on statistical patterns but has no semantic understanding – no meaning, no consciousness. Treat AI as a tool, not as a mind.
5. Common Myths About AI
|
Myth |
Reality |
|
"AI is conscious or sentient." |
No evidence. Today's AI has no feelings or self-awareness. |
|
"AI will soon replace all human jobs." |
It will automate tasks, not entire jobs. |
|
"AI learns like a human child." |
AI needs millions of examples; humans need a handful. |
|
"AI is objective and unbiased." |
False. AI learns from human-generated data, which contains biases. |
Summary
|
Term |
Definition |
|
ANI (Artificial Narrow Intelligence) |
AI designed for one specific task. All existing AI is ANI. |
|
AGI (Artificial General Intelligence) |
Hypothetical AI that can perform any intellectual task a human can. Does not exist. |
|
Machine Learning |
AI that learns patterns from data instead of being explicitly programmed. |
|
Neural Network |
A machine learning model with layers of artificial neurons. |
|
Training |
The process of adjusting a model's parameters to reduce error on examples. |
Review Questions
- Is ChatGPT an example of Artificial General Intelligence? Why or why not?
- What are the three ingredients needed for machine learning?
- The Chinese Room argument questions whether AI can truly "understand." What is the difference between following rules and genuine understanding?