AI in Everyday Life: Current Applications, Explained

UPI Study › Courses › Introduction to Artificial Intelligence › Free lessons › Lesson 9.2 Free with a free account

AI in Everyday Life: Current Applications

Introduction to Artificial Intelligence · Chapter 9: AI Applications, Ethics, and the FutureSection 9.2 · lesson 63 of 6915 min · free with a free account

← Navigating the Ethical Landscape of AINext lesson: Transforming Healthcare Through Intelligent Systems →

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 voice assistants, recommenders, maps and personal apps use AI daily.

  • Describe the collect, predict, act and learn loop shared by everyday AI systems
  • Explain how voice assistants, recommenders, navigation and ride-hailing apps work, including their key trade-offs
  • Apply collaborative filtering, joint assignment and a standard-deviation alert rule to small everyday examples
  • Evaluate everyday AI features for privacy, accuracy and fairness risks

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 (2,102 words). Read along, or search it.

Opening Question

Four AI Systems Before Work

Before Maya even leaves for work, AI has touched her morning four times. Her phone heard her ask for the weather. Her map steered her around a jam. Her video app lined up a show, and her bank flagged a strange charge. Most of it was invisible. So what is each system doing, what data does it feed on, and what can go wrong?

The Shared Loop

Collect, Predict, Act and Learn

Underneath, all four run the same loop. Each one collects data about you, and predicts something. It acts on that prediction, then learns from how you respond. That loop is what makes these services useful, and it's also where their risks come from.

Voice Assistants

Wake Word, Recognition and Reply

Start with voice assistants. A voice assistant is a speech pipeline packaged as a product, with one extra step in front: is anyone talking to me? A tiny, always-on detector listens for one wake word, and ignores everything else. Then the request is recognized as text, and understood as an intent with its details, like a timer and its length. Finally, the system acts or looks up an answer, and speech synthesis speaks the reply. The wake word is spotted on the device itself. The later stages may run on the device, or on the company's servers.

Wake-Word Threshold

Missed Commands Versus Accidental Wakes

The wake-word detector gives every sound a score, and wakes up when the score passes a threshold. Set the threshold low, and it rarely misses you. But ordinary sounds wake it by accident, and it may record talk never meant for it. Set it high, and it stays quiet, but ignores you more often. Here's a simulation with assumed scores, and sixteen thousand ordinary sounds a day. At a threshold of 0.5, nothing is missed, but it wakes by accident about 99 times a day. At 0.65, it misses under 1 percent of commands, and wakes by accident about once every two days. At 0.8, accidents all but vanish, but it misses about 27 percent of real commands. That's the precision-recall trade-off, with a privacy cost. So designers pick a point between, and add safeguards: on-device processing, a recording light, and easy deletion.

Conversation Levels

From Single Commands to Open-Ended Replies

Assistants also differ in how much conversation they handle. The simplest takes a single command with every detail given: set a timer for ten minutes. Guided dialogue asks for missing details, one at a time. Context-aware systems follow up on earlier turns: what's the weather today? And tomorrow? And open-ended systems generate free replies on almost any topic: a plan, a poem, a problem.

Slot Filling

Booking a Table Detail by Detail

Guided dialogue works by slot filling. The system knows which details a request needs, tracks which ones it has, and asks for the rest. Book a table for four, fills the party size. Which restaurant? Luigi's. What time? Seven thirty. Nothing is missing now, so it confirms. Open-ended replies are fluent, but fluency isn't accuracy. They can state wrong facts in a confident tone, so check the answers that matter. And at every level, unfamiliar accents, background noise and recognition errors remain the main reasons a request fails.

Recommenders

Content-Based, Collaborative and Hybrid

Now recommendation systems. A catalog can hold millions of items, and nobody browses them all. A recommender ranks items by how likely you are to like them. Content-based systems suggest items whose features resemble ones you liked. Their weakness: more of the same. Collaborative filtering suggests what people with tastes like yours enjoyed. It needs ratings from many users, and fails for new users and new items: the cold start problem. Hybrids combine both, plus context like the time of day. They're harder to build and test.

Worked Example

Predicting Alex's Ratings From Similar Viewers

Here's collaborative filtering at work. Five viewers rate six shows from one to five. Alex hasn't seen the last three yet. First, measure how closely each viewer's taste matches Alex's on the three shows Alex has rated. That's a correlation, from minus one to plus one. Bea and Chen score plus 0.97: nearly identical taste. Dara scores minus one, and Eli minus 0.84. Their tastes are opposite, so they're ignored. Alex's predicted rating for each show is a weighted average of Bea's and Chen's: 5 for Robot Friends, 4.5 for Detective Duo, and 1.5 for Garden Diaries. So Robot Friends goes first, Detective Duo second, and Garden Diaries isn't suggested at all.

Filter Bubbles

Exploration Finds Hidden Items

Cold start has a usual remedy: begin with popular or content-based picks, and ask a new user a few preference questions. There's a subtler trap. Recommendations shape what people click, and those clicks feed the next recommendations. A simulation offered fifty equally appealing items, with ten slots a round. Always showing the most-clicked items found only ten of them, ever. With 10 percent of slots shown at random, 38 were found by round 200, and all 50 by round 2,000. That's exploration against exploitation again. Without deliberate exploration, a recommender narrows what people see: a filter bubble. And clicks are easy to measure, but what matters is whether people were glad afterward.

Navigation

Shortest Paths on Live Traffic

Navigation is search. A map app treats the roads as a graph: junctions are nodes, and roads are links labeled with travel times. The fastest trip is a shortest-path search, from the search lessons. The AI is in the labels: live speeds from phones on each road, and models that predict delays. From Home to Work, the highway route takes 6, 8 and 5 minutes, 19 in all. The other two routes take 22. Now congestion triples the highway section, from 8 minutes to 24. That route jumps to 35, and the search switches to a 22-minute route. But send thousands of drivers down the same alternative, and it jams too, so good systems spread traffic. And location data reveals where people live and work.

Ride-Hailing

The Assignment Problem

Ride-hailing must match waiting riders to nearby drivers. That's an optimization problem, called the assignment problem. Here are four riders, four drivers, and each driver's pickup time in minutes. First-come matching gives each rider the nearest free driver, in turn. R1 takes D1, R2 takes D3, R3 takes D4, and R4 gets D2. Total, 26. R3's nearest driver was the one R4 needed most. The best joint assignment looks at everyone together, and totals just 12. Real systems also weigh traffic, driver direction and fairness to drivers. And when prices rise with demand, riders should be able to see why.

Quick Check

First-Come Versus Joint Assignment

Quick check. Two riders, two drivers. Driver one is 2 minutes from rider one, and 3 from rider two. Driver two is 4 minutes from rider one, and 12 from rider two. Rider one asks first. How many minutes does the best joint assignment save, compared with first-come? Pause, and decide. Seven minutes. First-come gives rider one the nearest driver, driver one, and leaves rider two waiting 12 minutes: 14 in all. The joint assignment sends driver two to rider one, and driver one to rider two: 4 plus 3, just 7. Nine is the trap. Rider two does save nine minutes, but rider one now waits two minutes longer.

Smart Homes

Sensors, Actuators, Rules and Learning

A smart home combines sensors, like motion, temperature, door contacts, cameras and microphones, with actuators: lights, heating and locks. Some behavior is hand-written rules: if a window is open, lower the heating. Other behavior is learned from how the household actually lives. Learned schedules warm the house before you usually wake. Presence detection, camera alerts for packages, and energy saving, like running the dishwasher when power is cheaper.

Worked Example

Learning When the Kitchen Light Goes On

Here's a routine being learned. A household logs when the kitchen light first goes on, over ten weekday and four weekend mornings. Weekdays average 7:01, with a spread of about 7 minutes, so the light could start itself at 6:54. Weekends average 8:30, with a spread of about 10 minutes: start at 8:20. Mix all fourteen days together, and the average is 7:26, with a spread of 42 minutes. The suggested time, 6:44, suits neither group. Finding the right grouping is part of what makes learned automation work.

Smart Home Risks

Microphones, Hacking and Data Sharing

Smart homes carry real risks. Always-on microphones and cameras collect intimate data, so look for physical off switches, local processing and clear indicators. Poorly secured devices can be hacked. Keep them updated, with strong, unique passwords, and put them on a separate network. Data may be shared with the maker or third parties, so read the sharing settings. And keep manual controls that work when the internet is down.

Personal Finance

Sorting Merchants With Naive Bayes

Personal finance apps connect to bank and card accounts. They sort spending into categories, track budgets, forecast bills and flag unusual charges. Sorting is text classification: the merchant name is the input, and the category is the label. A naive Bayes classifier was trained on just 24 labeled names. It got 8 of 8 new names right, and it reports how sure it is. Fresh Valley Market: Groceries, at 86 percent. Shell Coffee is right too, Dining, but only at 35 percent, against 33 for Transport. Shell appeared in a fuel station's name. When two guesses are that close, a good app asks you to confirm. Your corrections become new labeled examples, and it improves.

Unusual Charges

The Three Standard Deviation Rule

A second feature spots unusual charges. Over 29 ordinary days, dining spending averaged $26.38 a day, with a standard deviation of $8.50. A common rule alerts on anything above the average plus three standard deviations: about $52 here. On day 30, a $214.50 charge lands 22 standard deviations above average, and it's flagged at once. Remember, spending summaries describe the past. They don't know your debts or goals, so big borrowing or investing decisions belong with a qualified adviser.

Personalized Learning

Bayesian Knowledge Tracing Step by Step

Personalized learning tools fit practice to each student. They estimate what you've mastered, pick the next question, and schedule reviews. A classic model, Bayesian knowledge tracing, keeps the probability that a skill is mastered, and updates it after every answer. It uses four numbers. Here, the starting chance of mastery is 0.3, and the chance of learning from each practice is 0.2. A slip, a wrong answer despite knowing the skill, has a chance of 0.1. And a lucky guess has a chance of 0.25. Watch one student answer. A wrong first answer drops the estimate to 0.24. Two correct answers raise it to 0.89. Then a wrong answer at question four. The estimate dips to 0.61, not to zero, because that wrong answer might be a slip. Two more correct answers, and at question six it passes 0.95. The tool moves on to a new skill, instead of drilling the same one.

Strengths and Cautions

What Learning Tools Do Well and Poorly

The strengths are real. Practice fits each learner's pace, feedback is immediate, and teachers see who needs help, with what. So are the cautions. Student data is sensitive. Skills that are easy to measure can crowd out the rest, and over-reliance can replace discussion. And a model trained on an unrepresentative group of students can be less accurate for others.

Quick Check

Which Fuel Charge Gets Flagged

Quick check. A driver's fuel charges average $30 a fill-up, with a standard deviation of $4. The app uses the three-standard-deviation rule. Which of these three charges gets flagged? Pause, and decide. Only the $43 charge. The alert line is 30, plus three times 4: $42. $41 is the tempting one. It's well above average, nearly three standard deviations, but it's still under the line.

Judging AI Features

Questions to Ask Before Relying on It

So how should you judge an AI feature you rely on? Ask what data it collects, and why. More data can mean better service, and more exposure. Ask where the processing happens, and how long data is kept. Ask what happens when it's wrong: a wrong song is minor, a wrong money transfer is serious. Ask whether you can see, correct or delete what it knows about you, and whether it works for you, or for the company's goals. And ask whether it works equally well for everyone, since errors can fall harder on some voices, groups or neighborhoods.

Back to Maya

What Each System Was Doing

Back to Maya's morning. Her assistant woke on a threshold that trades missed commands for accidental recordings. Her map ran a shortest-path search on live traffic. Her show came from a recommender that needs to keep exploring. Her bank's alert was a simple standard-deviation rule. Each one collected data, predicted, acted and learned, and now she knows what to ask of each.

Key Takeaways

Everyday AI in Summary

To sum up. Everyday AI reuses familiar techniques: speech recognition, supervised learning, search, optimization, rules and probability. A wake-word threshold trades missed commands against accidental recordings. Recommenders need deliberate exploration, or popularity locks in a few items. Navigation and ride-hailing are search and optimization, fed by live predictions. Personal tools learn your patterns, and they must be tested for errors and protected for privacy.

Think About It

Happier This Week, Narrower This Year

Here's something to think about. A recommender makes people happier this week, but narrows what they discover over a year. How would you decide whether it's successful? What would you measure, and over what time?

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 AI in Everyday Life: Current Applications

How does collaborative filtering work in a recommendation system?
Collaborative filtering suggests items that people with tastes like yours enjoyed. It measures how closely other users' ratings match yours, often with a correlation from minus one to plus one, then predicts your rating for an unseen item as a weighted average of the most similar users' ratings. It needs ratings from many users and struggles with new users and new items, which is called the cold start problem.
What is a filter bubble in recommender systems?
A filter bubble is the narrowing of what people see when a recommender keeps showing items that were already clicked, and those clicks feed the next recommendations. In one simulation of 50 equally appealing items, always showing the most-clicked items surfaced only 10 of them, while showing 10 percent of slots at random found all 50 by round 2,000. Deliberate exploration counters the effect.
How does a wake word detector decide when to start listening?
A wake word detector gives every sound a score and wakes up when the score passes a threshold. A low threshold rarely misses a command but wakes by accident more often, sometimes recording talk never meant for it. A high threshold avoids accidental wakes but ignores more real commands. Designers pick a point between and add safeguards such as on-device processing, a recording light and easy deletion.
How does a finance app detect an unusual charge?
A common rule flags any charge above the average plus three standard deviations of past spending in that category. If fuel charges average $30 with a standard deviation of $4, the alert line is $42, so a $43 charge is flagged and a $41 charge is not. The rule describes past patterns only, so it does not know a person's debts or financial goals.

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 →