When you hear the term “artificial intelligence,” you may envision very specific AI applications, such as online chatbots or self-driving cars. In reality, there are many types of artificial intelligence, ranging from the somewhat limited AI tools (think virtual assistants) to sophisticated AI systems that perform real estate market analysis.
What all of these AI variants have in common is that they try to mimic human intelligence, including image recognition, problem solving, and the ability to apply previous knowledge. Here are the core types of AI, grouped by both capability and functionality.
Table of contents
Capability-based AI types
All types of AI use some combination of machine learning (ML), natural language processing (NLP), speech recognition, computer vision, and/or robotics powered by artificial neural networks that mimic the cognitive functions of the human brain. Here’s a breakdown of the three main types of AI, grouped by capability:
1. Narrow AI
Narrow, or weak AI, is what we encounter most often in our everyday lives. Narrow AI excels at performing specific, well-defined tasks with high accuracy. Think of it as a single-minded tool that’s very good at what it’s designed for but is unable to apply its knowledge to entirely new situations.
Examples of artificial narrow intelligence include:
- Spam filters. These AI systems analyze emails to identify and remove a broad range of spam messages from your inbox.
- Facial recognition software. This technology can recognize faces in images and videos and is used for security purposes, such as unlocking your smartphone.
- Chess-playing computers. These AI programs can analyze millions of chess positions to make optimal moves, often surpassing human grandmasters.
- Website analytics tools. These AI tools audit your websites and analyze site usage and customer behavior; the more data they can crawl, the better their insights.
- ChatGPT. ChatGPT is an AI chatbot that is trained on a massive corpus of text data, allowing it to recognize patterns and respond to written prompts.
2. Artificial general intelligence
Artificial general intelligence (AGI), or strong AI, is a type of AI that would theoretically equal or exceed the abilities of the human mind. Computer scientists envision AI machines that can learn, reason, solve problems, and adapt to new situations just like a human can. AGI can also engage in reinforcement learning—where it trains to make decisions by interacting with the environment—to improve its future actions and outcomes. A general AI system would mimic human efforts, although it would not experience human emotions.
AGI is still in the domain of science fiction, but researchers are exploring machine learning techniques that could pave the way for this level of general intelligence. Here’s what AGI might be capable of:
- Creating original works. An AGI system could process human language to write a news article, compose a musical piece, create visual art, or even design a building.
- Understanding and responding to complex questions. Intelligent machines running on AGI could analyze vast amounts of information to answer questions in a comprehensive and informative way, even if the answers require reasoning or making judgments.
- Helping solve complex problems. AGI neural networks could analyze data on matters of global importance—such as water scarcity—and they could work with humans to develop effective solutions or propose solutions without human intervention.
3. Artificial superintelligence
Artificial superintelligence (ASI) represents the most capable forms of AI. It’s also the most speculative type of AI development. Also known as super AI, superintelligence could surpass human intelligence in all aspects, potentially solving problems and creating products and formulas that human beings couldn’t even imagine.
ASI capabilities are purely hypothetical, but some hopes for the technology include ASI that could analyze scientific data and make discoveries that would revolutionize fields like medicine, physics, or materials science. Through a combination of analyzing past data and making new inferences, a computer could not only create new consumer goods, but also invent categories of goods that no human brain has ever considered.
Functionality-based AI types
AI can also be categorized based on its functionality with respect to how it perceives and reacts to the world around it. Here’s a breakdown of the four main types:
1. Reactive AI
This is the simplest form of AI. Reactive machines essentially respond to their environment with pre-programmed scripts. In human terms, imagine having a reflex without the learning or memory. These AI applications don’t store information about past experiences and can’t adapt their behavior based on new situations. Examples of reactive machine learning models include:
- Traffic light controllers. These systems use sensors to detect the presence of cars and change lights accordingly.
- Simple chatbots. Some chatbots, including many ecommerce chatbots, can answer basic questions about a company’s products or services but can’t engage in a conversation that goes beyond their script.
- IBM Deep Blue. Chess-playing programs like Deep Blue can calculate optimal moves based on the current game state without considering previous matches.
2. Limited memory AI
These AI systems have a bit more sophistication than reactive machines. Their basic limited memory can store information about past experiences and use that information to inform their current decisions. Here are some examples:
- Self-driving cars. A present-day self-driving car uses sensors and cameras to monitor specific objects and navigate roads, essentially functioning as limited memory machines.
- Recommendation systems. As part of their learning stage, these AI systems track your past purchases or browsing history and then use this data to suggest products you might be interested in.
- Advanced email spam filters. More advanced spam filters might consider not only the content of an email but also past interactions with the sender to determine if it’s spam.
3. Theory of mind AI
This type of AI is still conceptual, but it would represent a significant leap in capability. Theory of mind AI would be able to process present-moment data (like facial expressions) to understand the thoughts, intentions, and emotions of others. These emotional learning capabilities would allow the technology to anticipate how others might behave and react accordingly.
Here are some of the concepts theory of mind AI research is building toward:
- Social robots. Some data scientists hope to create machines that can understand your emotional state and respond in a way that is comforting or helpful.
- Personalized learning systems. A theory of mind AI system could tailor its teaching approach to the individual student’s needs and learning capabilities, helping students tackle an intellectual task much like a patient, adaptable human teacher might.
4. Self-aware AI
An even more futuristic concept is self-aware AI that extends beyond the deep learning and machine learning algorithms that power today’s AI systems. This type of AI wouldn’t simply acquire knowledge and perform complex tasks; it would have consciousness and self-awareness, meaning it would understand its own existence and its place in the world. While this is far beyond the capabilities of current AI, it’s a concept that philosophers and science fiction writers love to explore.
Types of AI FAQ
What are the seven types of artificial intelligence?
The seven types of artificial intelligence are reactive, limited memory, theory of mind, self-aware, narrow, general, and superintelligent.
What kind of AI is ChatGPT?
ChatGPT is an example of narrow AI, specifically falling under the category of conversational AI, capable of generating human-like text-based responses within a predefined scope. You can also describe it as generative AI.
What are the two classifications of AI?
AI can be classified by capability (narrow, general, superintelligence) or functionality (reactive, limited memory, theory of mind, self-aware).