Categorizing AI: Types and Distinctions, Explained

UPI Study › Courses › Introduction to Artificial Intelligence › Free lessons › Lesson 1.2 Free with a free account

Categorizing AI: Types and Distinctions

Introduction to Artificial Intelligence · Chapter 1: Foundations of AI and Core ConceptsSection 1.2 · lesson 2 of 6914 min · free with a free account

← What Is Artificial Intelligence? Origins, Ideas, and GrowthNext lesson: Smart Agents and Automated Discovery Tools →

Earn 3 college credits for Introduction to Artificial IntelligenceRecommended for credit by NCCRS · Computer Science 373 · self-paced · one payment, lifetime access
Enroll in Introduction to Artificial Intelligence · $250 →Download the syllabus (PDF)

In this lesson

Sorting AI systems by capability, functionality, and how knowledge is built.

  • Distinguish narrow, general, and super AI by the range of their abilities
  • Compare reactive machines, limited memory, theory of mind, and self-aware AI
  • Contrast symbolic and statistical approaches and recognize hybrid systems
  • Classify a real AI system using three sorting questions

Free study notes

Get this lesson's notes by email

The key points and 4 practice questions with answers, in your inbox. One email, free.

One email. Unsubscribe in one click.

Transcript

Everything said in the video, section by section (2,015 words). Read along, or search it.

Opening Question

One Word for Very Different Systems

Ask ten people what artificial intelligence is, and you'll hear ten different pictures. A phone that unlocks with your face. A streaming service that guesses your next favorite show. A friendly robot from a movie. These have almost nothing in common, in what they can do or how they work. Calling them all AI is like using one word, vehicle, for a bicycle, a cargo ship, and a rocket. The word is true, but it doesn't help you choose, compare, or understand anything. So how should we sort AI?

Three Sorting Questions

Capability, Functionality, and Approach

Categories fix that. We'll use three different ways of sorting, and each one asks its own question. By capability: how wide is the system's ability? By functionality: how does it use memory, and awareness? By approach: how did its knowledge get inside it? And here's the key. Any single system can be placed on all three at once.

Narrow AI

Built to Do One Task Well

The first way of sorting looks at range. How many different things can the system do, and how well? Narrow AI, sometimes called weak AI, is built to do one task, or a small group of closely related tasks, very well. A chess engine can beat world champions. Yet it can't recommend a movie, read an X-ray, or even play checkers, unless someone builds a new program for that job. Route planners, face unlock, spam filters and translation tools are all narrow AI.

Common Misconception

Weak Means Limited in Range

And weak doesn't mean low-powered. It means limited in range, like an Olympic sprinter who's unbeatable on the track, but no help in a swimming race. Essentially every AI system in use today is narrow.

General AI

AGI and Transfer Across Areas

General AI, also called artificial general intelligence, or AGI, would learn and perform almost any mental task a human can. It could understand a brand-new problem, pick up unfamiliar skills, and carry knowledge from one area into another. That last ability is called transfer. A person who learns to drive a car can adapt to a van within minutes. A general AI would show that flexibility across far bigger jumps, like moving from accounting to engine repair. No system is widely accepted as having reached this level. Researchers disagree about how far away it is, and even about how to test for it.

Super AI

Outperforming the Best Humans

Super AI, or artificial superintelligence, is one step further: a system that would outperform the best humans in nearly every area. That includes scientific creativity, long-range planning, and social skill. It exists only as an idea, and as a subject of thought experiments. The ethical questions it raises come later in the course.

Capability Ladder

Three Steps From Real to Theoretical

So capability is a three-step ladder. Narrow AI is the bottom step. It exists, and it includes every system in everyday use. General AI is the middle step. It hasn't been achieved, and it's an area of active research and debate. Super AI is the top step, and it's theoretical only.

Functionality Types

Memory and Awareness Inside a System

The second way of sorting asks how a system works on the inside. Does it remember anything? And does it understand anything about the minds around it? A widely used scheme has four types: reactive machines, limited memory, theory of mind, and self-aware, arranged from simplest to most advanced.

Reactive Machines

Same Situation, Same Response

A reactive machine responds only to the situation in front of it, right now. It carries nothing over from earlier moments, so the same situation always gets the same response. Classic chess programs, like Deep Blue, are the standard example. They study the current board, pick a strong move, and don't learn from the games they played before. Reactive machines are fast and predictable. But they can't improve with experience, or take earlier events into account.

Limited Memory

Drawing on Recent or Training Data

A limited memory system draws on past information when it decides. That can be recent observations. A self-driving car tracks how fast nearby cars have moved over the last few seconds. Or it can be knowledge gained from lots of earlier data during training, like a spam filter that learned from millions of past emails. It's called limited because that memory is narrow, and tied to one task. It isn't a lasting, growing store of life experience, like a person's. So limited describes the kind of memory, not the amount. And almost all AI in use today belongs in this group.

Theory of Mind

Modeling Other People's Beliefs

Theory of mind is a name borrowed from psychology. It means understanding that other people have their own beliefs, feelings and intentions, which may differ from yours. An AI with it would adjust its behavior to match. A tutoring system could tell that a student is stuck because of a wrong idea, not just a wrong answer, and change its explanation. Today's systems can pick up cues, like tone of voice, and respond to them. But that's pattern recognition, not a true model of another mind. Theory of mind is still at the stage of early research.

Self-Aware AI

A Purely Hypothetical Sense of Self

Self-aware AI would have a sense of its own existence, and its own inner states. It's purely hypothetical. Scientists have no agreed explanation of how awareness arises, even in humans. So there's no clear recipe for building it, and no accepted test for recognizing it.

Comparison Table

Which Functionality Types Exist Today

Here are the four types together. A reactive machine uses no past information. Limited memory uses recent data, or training data. Neither one models other minds. Theory of mind would, and a self-aware system would also model itself. Only the first two exist today, and most current systems have limited memory.

Connecting Schemes

How Capability and Functionality Overlap

How do these first two ways relate? Capability and functionality describe different things, so they overlap without being the same. Every system that exists today is narrow in capability, and either reactive or limited memory in functionality. A general AI would probably need strong limited memory, and some form of theory of mind, to cope with the human world. And neither scheme says how a system's knowledge was built. That's the third question.

Quick Check

Classifying a Smart Doorbell

Quick check. A smart doorbell recognizes your family's faces. It learned them from photos you uploaded last month. Each time someone rings, it looks at the face at the door, and unlocks only for family. Which functionality type is it? Pause, and decide. Limited memory. The doorbell draws on past information: the photos it learned from, which are its training data. Reactive is the tempting choice, because it only looks at the face in front of it. But a reactive machine carries nothing over, and this one couldn't recognize anyone without what it learned. And it isn't theory of mind. Knowing a face isn't knowing what that person believes or feels.

Symbolic AI

Hand-Written Rules and Logic

The third way of sorting ignores how wide or how aware a system is. It asks where the knowledge comes from. Two traditions have dominated the field. Symbolic AI, also called rule-based or classic AI, stores knowledge as explicit symbols, like facts and rules, and reasons through them with logic. People write that knowledge in by hand. Tax software that applies the written tax code is one example. A loan checker that follows a bank's eligibility rules is another. Its strengths: it's transparent, because you can trace exactly which rules led to an answer. And it works without large amounts of data. Its weakness is brittleness. Rules can't cover every real situation, and some tasks resist rules altogether. Nobody can write down a rule that captures what a cat looks like, in every photo. Still, symbolic AI isn't outdated. It's the better choice when decisions must follow clear rules, and be explained step by step, as in tax and safety checks.

Statistical AI

Learning Patterns From Many Examples

Statistical AI, also called data-driven AI, learns patterns from large numbers of examples, and uses probability to make judgments. Nobody writes the rule. Show it thousands of photos labeled cat and not cat, and it learns what cat photos tend to look like. It handles messy real-world inputs, like images, speech, and everyday language. And it improves when it's given more, or better, data. The costs: it needs a lot of data, its reasoning is hard to inspect, and it can fail in situations unlike the examples it learned from. How machines actually learn from data is the topic of section one point eight. For now, the key point is this: data takes the place of hand-written rules.

Worked Example

A Spam Filter Built Two Ways

Here's one task done both ways: a spam filter. The symbolic version uses rules a person writes. If the subject says free money, and the sender is unknown, send it to spam. It's easy to understand. But spammers simply change their wording, and someone must keep updating the rules by hand. The statistical version is shown a big pile of emails already marked spam or not spam. It works out for itself which patterns matter. When spammers change tactics, it's retrained on fresh examples.

Hybrid Systems

Combining Rules With Learned Patterns

Real systems often use both. A self-driving car uses statistical methods to recognize pedestrians and lane markings in camera images. And it uses symbolic rules to follow traffic laws, like stopping at a red light. Calling such a system hybrid is common, and correct. One more point: this third way of sorting is about method, not range. A statistical system can be narrow, and so can a symbolic one.

Classification Method

Three Questions for Any System

With three ways of sorting in hand, classifying any system takes three questions, one for each. Question one: how broad is its ability? Could it handle a job outside the one it was built for? If not, it's narrow. Question two: what does it draw on from the past? Only the present means reactive. Recent observations, or training data, mean limited memory. Claim theory of mind or self-awareness only if the system truly models other minds, or itself. Question three: how did it get its knowledge? Rules written by people mean symbolic. Patterns found in data mean statistical. A mixture means hybrid.

Worked Example

A Music App and a Chat Assistant

Try it on a music app that suggests songs, based on what you and millions of other listeners have played. It only recommends music, so it's narrow. It uses listening history, so it has limited memory. And its suggestions come from patterns in data, not written rules, so it's statistical. Now a harder one: a chat assistant that writes, translates, and answers questions on almost anything. Its skills are wide, so it's tempting to call it general. But wide coverage isn't human-level flexibility, with new skills learned and transferred. Most researchers still call it narrow, though the boundary is debated. It's limited memory, and statistical, built from patterns in huge amounts of text.

Quick Check

Classifying a Greenhouse Controller

Quick check. A greenhouse controller follows forty watering rules that a grower typed in. Every few minutes, it applies those rules to the current soil, light and temperature readings, and switches the sprinklers. It also saves each reading for the grower's monthly report. Classify it on all three questions. Pause, and decide. C: narrow, reactive, and symbolic. It does one job, so it's narrow. It decides from the current readings alone. The saved log goes into a report, not into the decision, so it's reactive. And a person wrote its rules, so it's symbolic. Option A is tempting, because the controller saves its readings. But nothing saved changes what it does. And B fails, because it never learned anything from data.

Summary Table

Five Familiar Systems Sorted

Here are five familiar systems, each sorted with the same three questions. A classic chess program, and a hospital checker that applies rules doctors wrote, are both narrow, reactive, and symbolic. The music app and the chat assistant are narrow, limited memory, and statistical. The self-driving car is narrow, limited memory, and hybrid. Now look down the capability column. Every one of them is narrow.

Key Takeaways

Only Narrow AI Exists Today

To sum up. Capability asks how wide a system's ability is: narrow, general, or super. Only narrow AI exists. Functionality asks about memory and awareness. Only reactive and limited memory systems exist, and most are limited memory. Approach asks where the knowledge comes from: rules people write, patterns in data, or both. And three questions classify any system: how broad, what from the past, and how it got its knowledge.

Reflection

Classify a Tool You Used

Here's something to think about. Pick an AI tool you used this week, and answer the three questions for it. Which answer were you least sure of, and what would you need to know to settle it?

About Introduction to Artificial Intelligence at UPI Study

Introduction to Artificial Intelligence (Computer Science 373) is a self-paced online course of 69 video lessons in 9 chapters, from intelligent agents and search to game strategy, neural networks, logic and reasoning, machine learning, language, robotics and AI ethics. It carries a NCCRS credit recommendation for 3 semester credits.

Every lesson transcript is free on pages like this one, and the videos are free with a free UPI Study account. Enrolling adds a practice quiz after each lesson, the graded work and final exam that earn the credit, lifetime access, and an official transcript on completion. UPI Study has helped over 130,000 students with college credits. ACE and NCCRS have over 2,100 cooperating colleges that will consider accepting credits; the final decision is with the universities.

Enroll now → Course details Download the syllabus All Computer Science courses Every course with PRO →

Compare: AI vs Current Trends in CS and IT

Why take it here

Why students take Introduction to Artificial Intelligence with UPI Study

How machines search, reason and learnIntroduction to Artificial Intelligence covers how AI systems are built: intelligent agents, search and heuristics, game strategy, logic, neural networks, machine learning, language and perception, and the ethics of AI. It is a common upper-level course in computer science degrees.
Credit at a fraction of the priceThe same 3 credits can cost $750 or more at a US college. Here it is one payment, self-paced, with lifetime access.
Finish your degree soonerTransfer credits reduce the courses you still owe, so you can graduate earlier or lighten a heavy semester.
Strengthen an applicationA completed college-level course on an official transcript shows admissions and scholarship committees you can do the work.
Study on your own scheduleNo start dates, no deadlines. Watch the lessons, practice, and take the final when you are ready.
130,000+students helped with college creditsACE and NCCRS have over 2,100 cooperating colleges that will consider accepting credits; the final decision is with the universities.

Where the credit goes

Universities that work with NCCRS

Introduction to Artificial Intelligence carries a NCCRS credit recommendation. These universities work with NCCRS; some are UPI Study's signed partners. ACE and NCCRS have over 2,100 cooperating colleges that will consider accepting credits; the final decision is with the universities.

Thomas Edison State UniversityThomas Edison State UniversitySigned partnerUniversity of the PeopleUniversity of the PeopleSigned partnerCharter Oak State CollegeCharter Oak State CollegeSigned partnerStrayer UniversityStrayer UniversitySigned partnerColumbia Southern UniversityColumbia Southern UniversitySigned partnerIndiana TechIndiana TechSigned partnerPenn StatePenn StateWorks with NCCRSSouthern New Hampshire UniversitySouthern New Hampshire UniversityWorks with NCCRSSUNYSUNYWorks with NCCRS +many more cooperating universitiesFind my college →

Find my college → Not listed? Ask Athena in the corner of this page, or book a 15-minute call.

FAQ

Questions students ask us

About Categorizing AI: Types and Distinctions

What is the difference between narrow AI and general AI?
Narrow AI, also called weak AI, is built to do one task or a small group of related tasks very well, like a chess engine or a spam filter. General AI, or AGI, would learn and perform almost any mental task a human can, including transferring knowledge from one area to another. Essentially every AI system in use today is narrow, and no system is widely accepted as general AI.
What are the four functional types of AI?
The four functional types are reactive machines, limited memory, theory of mind, and self-aware AI, ordered from simplest to most advanced. Reactive machines, such as Deep Blue, respond only to the present situation. Limited memory systems use recent observations or training data, as a self-driving car does. Theory of mind is still early research, self-aware AI is hypothetical, and most systems in use today have limited memory.
What is the difference between symbolic AI and statistical AI?
Symbolic AI stores knowledge as explicit facts and rules that people write by hand and reasons through them with logic, while statistical AI learns patterns from many examples and uses probability. Symbolic systems, like tax software, are transparent and need little data but are brittle. Statistical systems handle messy inputs like images and speech but need lots of data and are harder to inspect. Many real systems combine both.
Does weak AI mean the AI is not powerful?
No, weak AI means limited in range, not low-powered. A chess engine can beat world champions yet cannot recommend a movie or play checkers unless someone builds a new program for that job. In the same way, a limited memory system is called limited because its memory is narrow and tied to one task, not because it holds little data. Both terms describe scope rather than strength.

About the Introduction to Artificial Intelligence course

How many college credits is Introduction to Artificial Intelligence?
Introduction to Artificial Intelligence (Computer Science 373) carries a NCCRS credit recommendation for 3 semester credits. ACE and NCCRS have over 2,100 cooperating colleges that will consider accepting credits; the final decision is with the universities.
How much does Introduction to Artificial Intelligence cost and what is included?
$250, paid once, with lifetime access: all 69 video lessons, a practice quiz after each, the graded work and final exam, and an official transcript when you finish. Students who want several courses choose UPI Study PRO instead.
Which universities accept UPI Study credits?
UPI Study has signed partner agreements with universities such as Thomas Edison State University, University of the People and Charter Oak State College, and works with universities such as Penn State, SNHU and SUNY. ACE and NCCRS have over 2,100 cooperating colleges that will consider accepting credits; the final decision is with the universities. Use "Find my college" on this page to check yours.
How long does Introduction to Artificial Intelligence take?
It is self-paced with no deadlines. The 69 lessons total about 16 hours of video; most students add reading and practice time and finish in a few weeks to a few months, on their own schedule.
How are the exams done?
Quizzes and assignments are online in the course. The final exam is taken online with proctoring; a person on the UPI Study team checks assignments, usually within two days.
How do my credits reach my college?
When the course shows 100% complete, request the official transcript from your dashboard. It is sent to the college you name, and you can send it to more than one. Check your college first with "Find my college" on this page, or ask Athena.

Ready for the full course? Earn 3 credits in Introduction to Artificial Intelligence.

69 video lessons, a quiz after each one, and the graded work that earns the credit. Recommended for credit by NCCRS.

Enroll in Introduction to Artificial Intelligence →Download the syllabusEvery course with PRO →
Introduction to Artificial Intelligence
3 credits · NCCRSEnroll →