Evaluating How We Think: Cognitive Function Analysis

UPI Study › Courses › Introduction to Artificial Intelligence › Free lessons › Lesson 2.2 Free with a free account

Evaluating How We Think: Cognitive Function Analysis

Introduction to Artificial Intelligence · Chapter 2: Intelligence, Cognition, and Agent DesignSection 2.2 · lesson 10 of 6916 min · free with a free account

← Learnable Intelligence: Insights from Perkins' FrameworkNext lesson: Expert Systems and AI Approaches to Complex Challenges →

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

Seven cognitive functions and how to measure them fairly in machines.

  • Identify the seven cognitive functions and how they work together
  • Distinguish deductive, inductive, and abductive reasoning
  • Design a fair evaluation of a cognitive function using tasks, measures, baselines, and new test items
  • Map cognitive functions to AI subfields and explain key differences between human and machine abilities

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 (2,140 words). Read along, or search it.

Opening Scene

Two Seconds at a Street Crossing

Imagine crossing a busy street. You see a car, and notice it's moving fast. You remember that cars don't always stop. You decide to wait, and shout Wait! to a friend. In about two seconds, you used nearly every mental ability you have. So what is thinking actually made of? And how could we measure each part, in a person, or in a machine?

Cognition

Seven Overlapping Cognitive Functions

To build intelligent machines, we need a picture of what intelligence is made of. Psychologists call the mental processes of gaining, storing, and using knowledge cognition. Psychologists split cognition into separate cognitive functions. Seven are standard: perception, attention, memory, reasoning, problem solving, language, and learning. They aren't separate boxes in the brain. They overlap, and pass information along constantly. Roughly, information flows from perception, through attention, memory and reasoning, to problem solving and action. Language can enter or leave at any point, and learning improves all of them.

Perception and Attention

Best Guesses and a Mental Spotlight

Perception turns raw signals from the senses into something meaningful. Your eyes receive patterns of light, yet you experience a friend's face. It isn't a perfect recording. The brain fills gaps and makes best guesses, which is why optical illusions can fool us. Attention selects what to process, and what to ignore. Far more reaches you than you can handle, so attention works like a spotlight. In a noisy room, you can follow one conversation, yet still hear your name across the room. But focusing on one thing means missing others.

Memory

Working, Episodic, Semantic, and Procedural

Memory means encoding, storing, and recalling information, and it comes in several kinds. Working memory is a small mental notepad for what you're using right now, like a phone number while you dial it. Classic studies suggested it holds about seven items, plus or minus two. Later work suggests around four meaningful chunks. Long-term memory lasts from hours to a lifetime. Episodic memory holds events you lived, like your last birthday. Semantic memory holds facts, like the capital of France. Procedural memory holds skills, like riding a bike. And human memory is reconstructive. You rebuild each memory from pieces, which is why memories fade, or change.

Reasoning

Deductive, Inductive, and Abductive Styles

Reasoning is drawing conclusions from what you know, and it comes in three main styles. Deductive reasoning moves from general rules to a conclusion that must be true, if the rules are. All birds have feathers. A sparrow is a bird. So a sparrow has feathers. Inductive reasoning moves from many cases to a rule that's probably true. Every swan I've seen is white, so swans are probably white. Black swans do exist, in Australia. Induction can fail, even after many cases. And abductive reasoning picks the most likely explanation for what you observe. The grass is wet, so it probably rained overnight.

Problem Solving

From Starting State to Goal

Problem solving is finding a series of steps, from where you are to where you want to be. Every problem has a starting state, a goal, and a set of possible actions. Planning a route home, solving a jigsaw puzzle, and finding a bug in a program are all problem solving. It differs from reasoning in its focus. Reasoning asks: what follows? Problem solving asks: what should I do next?

Language

Sounds, Words, Grammar, Meaning, and Context

Language means understanding and producing symbols that follow rules and carry meaning. It works on several levels at once. Sounds or letters are the basic units. Words carry meaning, and bank can mean a riverside, or a place for money. Grammar, or syntax, decides how words combine. Dog bites man isn't man bites dog. Meaning, or semantics, asks what the whole sentence says. The table is hungry is grammatical, but odd. And use in context, called pragmatics, asks what the speaker intends. Can you pass the salt is a request, not a question about your ability.

Learning

The Function That Improves the Others

Learning is a lasting change in knowledge or behavior, caused by experience. It differs from the other six, because it acts on them. Through learning, you recognize more things, remember more, and solve problems faster. People learn from examples, like a child who sees a few dogs and grasps the idea of dog. They learn from rewards and mistakes, by watching others, and from direct instruction.

Worked Example

Every Function in One Crossing

Back to the street crossing. Seeing the car was perception. Ignoring a billboard was attention. Recalling that cars may not stop was memory. Judging it too fast to stop was reasoning. Waiting on the curb was problem solving. Shouting to your friend was language. And picking a signal crossing next time is learning.

Quick Check

A Dead Laptop and a Dark Charger

Quick check. Your laptop won't turn on, and you notice the charger's little light is off too. You decide the wall outlet probably has no power. Which style of reasoning did you just use? Pause, and decide. Abductive. You started from what you observed, a dead laptop and a dark charger light, and picked the most likely explanation. Inductive is the tempting choice, but you didn't build a general rule from many cases. And it isn't deductive, because the conclusion could still be wrong. Maybe the charger itself failed.

Measurement Recipe

Six Steps to Test a Function

How do we measure any of this? We can't look inside a mind, or a machine, and read off a perception score. Instead we follow a recipe. Name the function being tested. Choose a task that mainly needs it. Pick measures, such as accuracy or speed. Set a baseline to compare against. Test on new cases the system hasn't seen. Then compare, and report the averages, the spread, and where each one fails.

Tests and Benchmarks

Standard Tasks for People and Machines

Psychologists use standard tasks for people. AI researchers build benchmarks: shared test items with agreed scoring, so systems can be compared fairly. For perception, people identify objects shown briefly. Machines face image contests like ImageNet, or speech tests scored by word error rate. Attention has the Stroop task. Memory has digit span: repeating longer and longer lists of digits. Reasoning has matrix puzzles, and for machines, math word problem sets like GSM8K. Problem solving has the Tower of Hanoi. Language has reading tests, and question sets like SQuAD, where answers come from a passage. Learning is tracked with a learning curve.

Classic Test

The Stroop Task Isolates Attention

The Stroop task shows how a clever test isolates one function. Say the ink color of each word aloud, not the word itself. When the word red is printed in blue ink, you have to say blue. Reading is so automatic that it gets in the way. People are slower, and make more errors, on these conflicting items. That extra effort is attention, holding back an automatic response.

Measures

Accuracy, Speed, Capacity, and More

Then choose measures. Accuracy: what fraction of answers is correct? Nine of ten images labeled right is ninety percent. Speed: how quickly does it respond? Capacity: how much can it hold at once? Robustness: does it hold up when conditions get messier, like speech in a noisy cafe? Generalization: does it still work on cases it hasn't seen before? And efficiency: how much practice, data, or energy does it need?

Baselines

What Makes a Score Meaningful

A score means little on its own. Is eighty-five percent good? It depends on the baseline you compare it with. On a yes-or-no task, chance alone gets fifty percent. So compare with chance, with average people, with experts, and with earlier systems. Always test on new cases, items that weren't used for practice or training. Otherwise the test measures memorizing, not the function. And report the spread, and where each one fails. Two systems with the same average can fail very differently.

Learning Curves

Few Examples Versus Many

Learning is measured by watching performance change with practice. Plot accuracy against the number of examples, and you get a learning curve. Here's a typical contrast. It's a shape to remember, not measured data. People tend to improve quickly, from very few examples. Machine learning systems usually need many more. But on narrow, well-defined tasks, they can eventually reach very high accuracy.

Comparison Pitfalls

Five Traps in Human-Machine Comparisons

Comparing people and machines is full of traps. A narrow test: a top score on one task is read as general ability. Test leakage: the system already saw the test items in training, so it memorized instead of reasoning. Unfair conditions: a machine re-reads a passage endlessly, while a person reads it once. A weak baseline: comparing against the wrong people, or too few of them. And averages hide failures. A scanner that's right ninety-nine percent of the time can still fail on one unusual machine type. A fair comparison gives both sides the same task and new items, similar time and tools, and reports the spread, including hard cases.

Quick Check

Math Puzzle Scores That Collapse

Quick check. A homework app scores ninety-five percent on a published set of math puzzles. Researchers then write fresh puzzles of the same kind and difficulty, and its score falls to forty percent. Which pitfall best explains that first score? Pause, and decide. Test leakage. A published set can end up in training data. The app had most likely seen those puzzles, so it recalled answers, instead of reasoning. Fresh puzzles of the same difficulty exposed it. A weak baseline is tempting, but no comparison group was even involved. The fault was in the test items themselves.

Function-to-Subfield Map

A Table of Contents for AI

Many AI subfields are built to reproduce one cognitive function. That gives you a table of contents for the rest of this course. Perception maps to computer vision and speech recognition. Attention, to attention mechanisms in neural networks. Memory, to knowledge representation, and data storage and retrieval. Reasoning maps to logic and knowledge-based reasoning. Problem solving, to search, planning and game strategy. Language, to natural language processing. And learning, to machine learning. The map is approximate. Subfields overlap, and machine attention and memory are borrowed names, for calculations that do a similar job.

Worked Example

A Location Reminder on a Voice Assistant

See the map work in one product. You tell a voice assistant: remind me to buy milk when I get near a store. Turning your voice into text is perception, done by speech recognition. Working out that this is a reminder tied to a place is language: natural language processing. Keeping the reminder and your saved places is memory: data storage. Deciding when you're near a store, and triggering the alert, is problem solving: planning and rules. And getting better at your accent over time is learning: machine learning.

Human Versus Machine

Stickers, Forgetting, and Hallucinating

Where do people and machines differ? It follows from how each is built. Perception: people recognize objects in odd light and at odd angles. Machines excel on what they were trained on, but a few small stickers can make one misread a road sign. Attention in a machine is a numerical weighting of its inputs. It isn't conscious focus, and it never gets tired. Memory: machines store vast amounts exactly, but learning something new can overwrite older learning. That's called catastrophic forgetting. And language: systems trained on huge amounts of text write fluently, yet can state false things confidently. That's often called hallucinating.

Seven Differences

Data Efficiency to Confident Mistakes

Seven differences are worth remembering. Data efficiency: a child who sees three pictures of a giraffe can spot giraffes in new photos. An image system often needs thousands. Generalization: people carry what they know into new situations, while machines are reliable mainly inside their training range. Grounding: people link the word hot to the feeling of heat. Speed: machines process millions of records without tiring. Energy: a human brain runs on about twenty watts, like a dim bulb, while training large AI systems uses vastly more. Motivation: people pursue goals driven by needs and feelings, while machines pursue the goals designers give them. And mistakes: machines can be confidently wrong, in unfamiliar ways.

Full Analysis

Evaluating a Chest X-Ray Tool

Put it all together in a cognitive function analysis. Take a hospital tool that screens chest X-rays and drafts a short report. Step one, list the functions: perception, memory of patient history, reasoning, language for the report, and learning. Step two, map them to subfields. Step three, choose measures: accuracy on new images, speed, and how doctors rate the reports. Step four, set baselines: a panel of radiologists, and the hospital's old tool. Step five, find the limits. A new scanner type could change image quality, and the tool may state wrong findings fluently. So the useful question isn't: is the machine intelligent? It's: for which functions, measured how, compared with whom, and with what limits?

Key Takeaways

Functions, Measures, and Fair Comparisons

To sum up. Cognition splits into seven overlapping functions, from perception to learning, and learning improves all the others. To evaluate one, use a task that needs it, clear measures, a fair baseline, and new test items. Watch for narrow tests, leakage, unfair conditions, weak baselines, and averages that hide failures. Each function has an AI subfield, though machine attention and memory are borrowed names. And people and machines differ in data efficiency, generalization, energy, and the mistakes they make.

Reflection

One Test Score, One Function

Here's something to think about. A machine can transcribe speech with fewer errors than a tired listener, yet fail on a sentence a child understands easily. What does that say about using one test score to claim a machine has a cognitive function? And which of the seven do you think is hardest for machines to match?

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 Evaluating How We Think: Cognitive Function Analysis

What are the seven cognitive functions?
The seven standard cognitive functions are perception, attention, memory, reasoning, problem solving, language, and learning. They overlap and pass information along constantly rather than acting as separate boxes in the brain. Roughly, information flows from perception through attention, memory, and reasoning to problem solving and action, language can enter or leave at any point, and learning improves all the others.
What are the main types of memory in cognitive psychology?
Memory is usually split into working memory and long-term memory, and long-term memory includes episodic, semantic, and procedural memory. Working memory is a small mental notepad for what you are using right now, such as a phone number while dialing. Classic studies suggested it holds about seven items, plus or minus two, while later work suggests around four meaningful chunks. Episodic memory holds lived events, semantic memory holds facts, and procedural memory holds skills like riding a bike.
How do you fairly measure a cognitive function in a person or an AI?
Name the function, choose a task that mainly needs it, pick measures such as accuracy or speed, set a baseline, and test on new cases that were not used for practice or training. Then report averages, spread, and failures. A baseline gives a score meaning: on a yes-or-no task chance alone reaches 50 percent, so a score of 85 percent should also be compared with people, experts, and earlier systems.
How does human intelligence differ from machine intelligence?
People often learn from very few examples and carry knowledge into new situations, while machines often need thousands of examples and are reliable mainly within their training range. Machines process huge amounts of data without tiring and store information exactly, but learning something new can overwrite older learning, which is called catastrophic forgetting. A human brain runs on about 20 watts, while training large AI systems uses vastly more energy.

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 →