Smart Agents and Automated Discovery Tools, Explained

UPI Study › Courses › Introduction to Artificial Intelligence › Free lessons › Lesson 1.3 Free with a free account

Smart Agents and Automated Discovery Tools

Introduction to Artificial Intelligence · Chapter 1: Foundations of AI and Core ConceptsSection 1.3 · lesson 3 of 6916 min · free with a free account

← Categorizing AI: Types and DistinctionsNext lesson: How Algorithms Power Computer Programs →

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

How smart agents act rationally and discovery tools find knowledge in data.

  • Describe the perceive, decide, act loop that smart agents run
  • Explain rational choice using expected cost and distinguish it from being all-knowing or lucky
  • Identify the five common data mining tasks and the six steps of knowledge discovery
  • Explain how AI speeds scientific discovery and why its results need human verification

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 (2,230 words). Read along, or search it.

Opening Scene

A Robot Vacuum and Hidden Buying Habits

Picture a small robot vacuum in a living room. It bumps gently against a sofa leg, notices its battery is low, and rolls back to its charging dock. Nobody pressed a button. That's a smart agent: it acts on the world. Far away, other software sorts through millions of purchase records, and uncovers buying habits no analyst thought to check. That's an automated discovery tool: it finds knowledge hidden in data. Together, these two power much of the AI people meet every day.

Definition

Agents Sense and Act on Their Own

An agent is anything that can sense what's happening around it, and take action in response. A smart agent, also called an intelligent agent, does this on its own. It works toward a goal, without a person steering every move. It runs a loop. It perceives its surroundings, decides what to do, and acts. Then it does it all again, many times a minute.

Agent Loop

Perceive, Decide, Act, Repeat

Perceive. The agent gathers information through sensors, like cameras, microphones and temperature probes, or through data feeds, like traffic reports, emails and sales records. What it senses is its only window onto the world. Decide. This is the thinking step. The agent compares the possible actions, and picks the one most likely to move it toward its goal. The choice can follow a simple rule, like: if the room is below 20 degrees Celsius, turn on the heat. Or it can use patterns learned from data. Act. The agent carries out its choice through motors, screens, messages or commands. Actions change the surroundings, so the agent perceives something new, and the loop starts again.

Key Qualities

What Makes an Agent Smart

So what makes an agent smart? Three qualities separate it from an ordinary program. It acts on its own, without waiting for a person to start every step. It's goal-driven, choosing actions because they serve a purpose. And it responds to change, adjusting when conditions shift. Many modern agents also learn from experience. A calculator waits for you to press keys, and never senses its surroundings. So it isn't an agent. A thermostat keeps checking the room, and keeps acting by itself. So it is one.

Everyday Examples

Thermostats, Spam Filters, and Fraud Alerts

Agents are everywhere, once you look for them. A smart thermostat senses room temperature and whether anyone is home, decides whether heating is needed, and switches it on or off. A spam filter senses the sender, wording and links in each message, decides whether it looks like spam, and moves it to the spam folder. A navigation app senses your location and live traffic, picks the route likely to be fastest, and shows it, updating as things change. A bank fraud alert senses a purchase's amount and place, decides whether it's unusual for you, then blocks it and texts you. Agents come in many designs, from very simple to highly advanced. Their detailed parts, and the different kinds, come later in the course.

Rationality

Best Expected Result at the Moment of Choice

Now, what does it mean for an agent to act rationally? In everyday speech, rational means sensible. In AI, the word has a precise meaning. A rational agent picks the action expected to give the best result, measured against its goal, based on what it knows and what it has sensed so far. Notice the careful wording. Expected to give the best result. And based on what it knows. So rationality is judged by the quality of the decision at the moment it was made, not by how things turned out afterwards.

Four Ingredients

The Umbrella Decision Broken Down

Four ingredients shape every rational choice. Take a simple one: should you carry an umbrella today? First, a measure of success. What counts as a good result? Here, staying dry, with as little hassle as possible. Second, the knowledge so far. From experience, you know forecasts are usually reliable. Third, the available actions. You can carry the umbrella, or leave it at home. And fourth, what has been sensed right now. Today's forecast says a 70 percent chance of rain.

Common Mistake

Rational Is Not All-Knowing or Lucky

Rational doesn't mean perfect. No agent can see the future, and three ideas are easy to mix up. Rational means choosing the action with the best expected result, using what you know. You see a 70 percent forecast, and you take the umbrella. All-knowing would mean knowing exactly what will happen, like knowing for certain that rain starts at 3 p.m. No real agent can. And lucky means getting a good result by chance. You skip the umbrella despite the forecast, and stay dry because the clouds drift away.

Worked Example

Expected Cost of Carrying an Umbrella

We can put numbers on the umbrella choice, by scoring each outcome as a cost in points. These are illustrative scores. Getting soaked costs 10 points. Carrying an umbrella you didn't need costs 1 point, a small nuisance. Carrying one on a rainy day costs nothing. The expected cost of an option is its average cost, with each outcome counted by how likely it is. Carry the umbrella: 70 percent of zero, plus 30 percent of 1. That's 0.3. Leave it at home: 70 percent of 10, plus 30 percent of zero. That's 7. Carrying has the far lower expected cost, so it's the rational choice.

Quick Check

A 20 Percent Chance of Rain

Quick check. Same scores, but tomorrow's forecast gives only a 20 percent chance of rain. Which option has the lower expected cost? Pause, and decide. Carry it, even at 20 percent. Carrying costs 80 percent of 1, which is 0.8. Leaving costs 20 percent of 10, which is 2. So carrying still wins. Leaving it at home is tempting, because rain is unlikely. But a soaking is so costly that even a small chance of one outweighs a small nuisance.

Tipping Point

Rain Above About 9 Percent

In fact, there's a tipping point. Call the chance of rain p. Carrying costs one, times one minus p. Leaving costs ten, times p. They're equal when ten p equals one minus p. So p is one eleventh. That's about 9 percent. With these scores, any chance of rain above about 9 percent makes carrying the umbrella the better bet.

Good Decisions, Bad Outcomes

Sensible Choices That Still End Badly

So a sensible choice can still turn out badly. A navigation app picks the route that live traffic shows is fastest. A crash happens after it's chosen. The app couldn't have known. A chess program gives up a piece to set up a stronger attack. If the opponent finds a rare reply, the move was still the best one on the evidence. A spam filter blocks a message that matches many spam patterns. A rare genuine email gets caught, but blocking was the best call on what it could see. Real agents also have limited time and computing power. A chess program can't examine every possible game, so it searches as deeply as time allows, and plays the best move it found. Doing the best you can, with the time and information you have, is still rational.

Data to Knowledge

Data, Information, and Knowledge

Now the second half: automated discovery tools. Organizations collect enormous amounts of data, from every purchase, every sensor reading, every experiment. Nobody can read it all. Discovery tools use computers to sift through that data, and bring forward what's worth knowing. It helps to separate three levels. Data is raw facts, with no meaning attached yet. One receipt: a tent, 120 dollars, Saturday. Information is data organized so it says something. Tents sold 40 times this week. Knowledge is understanding you can act on. Tent buyers usually need sleeping bags, so show the two together.

Data Mining

Five Common Mining Tasks

Data mining is the use of algorithms to find useful patterns in large collections of data. The computer searches by itself, so people don't have to guess in advance what to look for. Five common tasks cover most uses. Classification asks which known group an item belongs to. Is this email spam, or not? Clustering asks which items naturally group together, like shoppers with similar habits. Association asks which things tend to happen together. People who buy tents often buy sleeping bags. Anomaly detection asks which cases look unusual, like a card purchase far from the owner's usual places. And prediction asks what's likely to happen next, like next month's sales of winter coats.

Warning

Ice Cream Does Not Cause Sunburn

One warning. Ice cream sales and sunburn cases both rise in summer. But ice cream doesn't cause sunburn. Both follow the weather. A mining tool will happily report a pattern like that, so people must judge whether it makes sense.

Worked Example

Three Natural Customer Groups

Clustering is easy to picture. A store has records of how often each customer visits, and how much they spend, but no labels. The algorithm finds three natural groups. This data is illustrative. One group visits often but spends little. One visits rarely but spends a lot. And one visits often, and spends a lot. Nobody told the algorithm those groups existed. The store can name them afterwards, and serve each one differently.

Knowledge Discovery

Six Steps at a Grocery Store

Knowledge discovery is the whole process of turning raw data into knowledge people can trust and use. Data mining is only one step inside it. Think of mining as the engine, and knowledge discovery as the entire trip, where algorithms and people work together. Here are its six steps, at a grocery store. Select: choose the data that fits the question. Use last year's checkout records, and leave out staff schedules. Clean: fix mistakes. Remove duplicate scans, and correct items typed with the wrong price. Transform: reshape the data so algorithms can use it. Turn each receipt into a list of items bought together. Mine: run algorithms to find patterns. Certain items often appear on the same receipt. Interpret: experts check whether the patterns are real. A manager spots one that came from a one-time promotion, and discards it. Use: apply the knowledge. Place related items near each other, and feed a recommendation system.

AI in Science

Searching Huge Spaces of Possibilities

Science often faces search problems of staggering size: millions of possible molecules, countless ways a protein can fold, huge archives of telescope images. AI helps in three ways. It scans huge spaces of possibilities far faster than people can. It spots patterns that humans may miss. And it suggests candidates worth testing. In some automated labs, robots even run the experiments.

Discovery Loop

Data, Model, Candidates, Experiments

A common pattern is a loop. Data trains a model. The model proposes candidates. Experiments check them. And the results become new data, for the next round.

Real Discoveries

AlphaFold, Halicin, GNoME, and Adam

Some real examples. In biology, AlphaFold predicts a protein's 3D shape from its building blocks. Its predicted structures, for hundreds of millions of proteins, are shared openly. In medicine, a deep learning model scanned a library of about 6,000 existing drugs, and flagged halicin, a compound that kills bacteria. The same model then screened more than 100 million other molecules, looking for more candidates. In materials science, a system called GNoME predicted about 2.2 million new crystal structures, and about 380,000 were judged the most stable. In astronomy, a neural network helped reveal a planet orbiting the star Kepler-90, in 2017. And a robot scientist called Adam formed ideas about gene functions in yeast, ran the experiments with robots, and read the results.

Verification

AI Proposes, Scientists Verify

But an AI result is a strong lead, not a final answer. A flagged molecule must still be tested in a lab. A predicted material must still be made, and measured. Human experts decide what the evidence really shows. AI proposes; scientists verify.

Quick Check

Grouping Unlabeled Bird Calls

Quick check. A wildlife team has ten thousand recordings of bird calls, with no species names attached. Software sorts the recordings into groups of calls that sound alike. Which data mining task is the software doing? Pause, and decide. Clustering. Nobody gave the software any groups or labels in advance. It found the groups of similar calls on its own. Classification is the tempting choice, because the calls do belong to species. But classification sorts items into known groups given ahead of time, and here there were none. And association would look for things that happen together, like two calls often heard at the same dawn.

Applications

Agents and Discovery Across Fields

Smart agents act on the world, while discovery tools find knowledge in data. Most fields use both. In healthcare, monitors watch patient readings and alert nurses, while mining health records spots risk patterns. In banking, fraud alerts block suspicious payments, while anomaly detection and customer clustering find the patterns. On farms, irrigation systems water when the soil is dry, while discovery tools link weather, soil and crop yield. And at home, thermostats and voice assistants act, while streaming services mine viewing patterns.

Working Together

Discovery Feeds a Recommendation Agent

And the two often work together. Discovery tools supply the knowledge, and agents put it to use. Take an online camping store. A discovery tool mines past purchases, and finds that tent buyers often buy sleeping bags. A recommendation agent then senses what a shopper has in the cart, decides which item to suggest, and shows it. Each new purchase becomes fresh data for the next round of discovery. These tools bring real benefits. But they also raise questions about fairness and privacy, which the course returns to later.

Key Takeaways

Agents Act, Discovery Tools Find

To sum up. A smart agent runs a loop: perceive, decide, act, and repeat. A rational agent picks the action with the best expected result, given what it knows. It doesn't need to be all-knowing, or lucky. Rationality is judged at the moment of choice, so a good decision can still end badly. Data mining finds patterns. Knowledge discovery is the whole process, with mining as one step. And AI proposes scientific candidates quickly, but scientists must verify them.

Reflection

When Nobody Checks the Patterns

Here's something to think about. A shopping app suggests items, based on patterns found in millions of purchases. What could go wrong if nobody checks those patterns before they're used?

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 Smart Agents and Automated Discovery Tools

What does it mean for an AI agent to act rationally?
A rational agent picks the action expected to give the best result, measured against its goal, based on what it knows and what it has sensed so far. Rationality is judged by the quality of the decision when it was made, not by how things turned out. A navigation app that picks the fastest route from live traffic acted rationally even if a crash later slows that route.
What is the difference between a rational agent and a lucky one?
A rational agent chooses the action with the best expected result using the information it has, while a lucky agent gets a good result by chance. Taking an umbrella when the forecast shows a 70 percent chance of rain is rational. Skipping it and staying dry because the clouds drift away is lucky. Rational also differs from all-knowing, since no real agent knows exactly what will happen.
How do you calculate expected cost to make a decision?
Multiply each outcome's cost by its probability, add the results, and choose the option with the lowest total. If getting soaked costs 10 points, carrying an unneeded umbrella costs 1, and rain is 70 percent likely, carrying costs 0.3 times 1, or 0.3, while leaving it costs 0.7 times 10, or 7. With these scores, carrying wins whenever the chance of rain is above about 9 percent.
What is the difference between data mining and knowledge discovery?
Data mining is the use of algorithms to find useful patterns in large collections of data, while knowledge discovery is the whole process of turning raw data into knowledge people can trust and use. Mining is one of its six steps: select, clean, transform, mine, interpret, and use. The interpret step matters because tools report patterns, like ice cream sales and sunburn rising together, that share a cause rather than one causing the other.

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 →