What does NLP stand for?

NLP stands for natural language processing. It’s an area of computer science that draws on machine learning to train AI models. NLP allows computers to create text and speech. Chatbots are an example of NLP.

What are the seven domains of AI?

The seven domains of AI are machine learning, natural language processing, computer vision, robotics, expert systems, and autonomous systems. These domains are fields of study that allow AI to function and advance.

Is AI a mind or a machine?

AI is a machine that is capable of learning. Sometimes called “machine intelligence,” AI uses algorithms, data processing abilities, and large language models to make predictions and create generative text, like with ChatGPT.

What is the most common type of AI used today?

The most common type of AI used today is called narrow AI. This type of AI can perform tasks quickly and efficiently, but narrow AI lacks the general intelligence of humans. Examples of narrow AI include chatbots and virtual assistants.

Types of AI compared

Table of Contents

Stages of AI defined
Stages of AI compared
FAQ

By: Genevieve Carlton, Edited by: Rebecca Munday

Published: April 21, 2025

Closeup of two men data science specialists working at office together, analyzing big data on screen.

Artificial intelligence (AI) is already revolutionizing how we work and live. But AI comes in many forms. As technology advances, new types of AI can take on more complex tasks.

Discover four types of AI, including stages that scientists are currently developing.

The stages of AI

Reactive

Reactive AI is the most basic type of artificial intelligence. This type of AI reacts to situations without drawing on stored data or prior experience. Unlike limited memory AI, reactive AI does not learn or store memory.

Many of the earliest examples of reactive AI used machine processing to play games. Some examples include:

IBM Deep Blue: In 1997, IBM developed an AI capable of beating chess champions. While Deep Blue could compete in chess at the highest levels, it could not learn from prior games and had no memory.
Customer service chatbots: Simple chatbots provide customer service functions in response to input without needing memory. These chatbots respond to keywords but do not learn or adapt as customers interact with them.
Smart home devices: Reactive AI powers smart home devices that respond to real-time conditions. For example, smart thermostats trigger heating or cooling based on the temperature.

Limited memory

Limited memory AI stores data and draws on it to improve its predictions, making it a valuable type of AI for business executives. Learning is a key process for limited memory artificial intelligence, which requires feedback to improve its predictions. Large language models, like ChatGPT, are a type of limited memory AI.

Other examples of limited memory AI include:

Autonomous vehicles: Self-driving cars use limited memory AI to drive safely, follow traffic laws, and avoid obstacles. Autonomous vehicles require feedback to improve decision-making.
Netflix recommendations: Limited memory AI systems can react to recently watched programs to make recommendations for what viewers might enjoy. Feedback improves this process.
Virtual assistants: Virtual assistants and more complex customer service chatbots built on machine learning models adapt to real-time requests. These virtual assistants can learn from past interactions.

Theory of Mind

Theory of Mind (ToM) AI does not yet exist. However, this type of AI has the capacity to express empathy and understand human emotions. ToM AI would be closer to how humans make decisions rather than the more static predictive approach of limited memory AI.

Generative AI models such as ChatGPT may evolve into ToM AI. These types of AI can respond to human emotion, but they still may face the challenge of adapting to emotion as quickly as humans can naturally.

In the future, ToM AI systems may create robots that can serve as companions to humans. Currently, ToM AI remains theoretical.

Self-aware

Like ToM AI, self-aware AI does not yet exist. This theoretical type of AI would be conscious and aware of its existence. It has also been called artificial superintelligence.

While we do not yet have the technology to make self-aware AI a reality, science fiction has created fictional self-aware AIs, such as HAL from "2001: A Space Odyssey."

The possibility of AI becoming self-aware raises ethical questions. Fictional self-aware AIs have displayed hostility toward humanity. Even a harmonious future where self-aware AI coexists with humanity poses difficult questions about the rights of intelligent machines with consciousness.

Compare the stages of AI

The stages of AI compared
AI stage	Function	Benefits	Challenges
Reactive	Responds to input without using memory.	Can be built on a machine-learning model to complete discrete tasks.	Does not learn or adapt and thus can only complete simple tasks.
Limited memory	Reacts and adapts to input by drawing on memory to make predictions.	Uses natural language processing and machine learning to improve predictions.	Requires continuous learning, either through human-lead or situational feedback.
Theory of Mind	Uses human emotion to react to situations and make decisions similar to a human.	Stronger emotional intelligence can better serve people.	Raises questions about the possibility of emotional manipulation.
Self-aware	Develops a consciousness and can make subjective decisions	Superintelligence can potentially improve human decision-making.	Raises ethical questions about the dangers and rights of self-aware AI.