This lesson describes what AI is and how it works through concepts such as machine learning (ML) and generative AI (genAI). You’ll discover how AI is woven into everyday experiences and investigate its potential to transform education. Plus, key applications are presented as examples of how AI can create a more inclusive learning environment and empower all students to succeed.
AI fundamentals and inner workings
As you’ve learned, AI describes computer programs that complete cognitive tasks typically associated with human intelligence. A cognitive task is any mental activity, such as thinking, understanding, learning, and remembering. Cognitive abilities enable people to make effective choices and thoughtfully solve problems; yet, there are limits to how much information humans can process. This is where AI comes in — extending our information-processing skills, supporting creativity and innovation, and expediting routine tasks.
The AI4K12 Initiative is a joint effort by the Association for the Advancement of Artificial Intelligence and the Computer Science Teachers Association, which is focused on developing national guidelines for teaching AI in K-12 classrooms. The initiative identifies “Five Big Ideas in Artificial Intelligence” that describe the fundamental concepts in the field:
-
Perception refers to AI’s ability to receive and interpret information through computer vision, speech recognition, and other sensor-based inputs.
-
AI uses representation and reasoning to solve problems, draw conclusions, and make predictions.
-
ML is a subset of AI that helps make it possible for computers to learn from data, identify patterns, and improve their performance over time.
- When using AI, a key objective is natural and intuitive interaction, which leads to developments that bridge the gap between humans and machines, enhancing communication and collaboration.
As AI pervades various aspects of life, its societal impact requires careful consideration in order to address ethical concerns and bolster the potential for societal good.
Types of machine learning
Before moving on, take a moment to explore machine learning (ML) a bit more in-depth. Machine learning refers to the process of a machine “learning.” Think of it like training a pet. You toss your dog a ball and say, “Fetch.” Then you reward her when she brings it back. Over time, she learns to associate “fetch” with the action, even if you use a different ball or toy.
ML algorithms are similar: They learn from examples and improve their ability to identify patterns as they’re exposed to more information. Put simply, an algorithm is a set of rules that a computer follows to solve problems. So, like a dog playing fetch, ML is used to develop computer programs that can analyze data to make decisions or predictions, without needing explicit instructions for every single situation.
Many AI tools use ML for tasks that require flexibility and adaptability. A common example is spam filtering in email programs, where ML algorithms can identify patterns typical of spam messages, such as an unusual amount of typos. But here’s the twist: ML goes beyond simple rules. The program is also trained to recognize patterns in your usual email traffic, so it won’t mistakenly send an email from a friend or colleague with a few typos straight to spam. It learns what’s normal for you and what’s not, just like your dog learns the difference between a dog toy and your favorite pair of shoes.
Understanding the different types of ML and how they support AI will enable you to use AI tools even more effectively. There are three common approaches to training ML programs: supervised learning, unsupervised learning, and reinforcement learning.
Supervised learning
Supervised learning is a type of ML that uses labeled datasets to train a program to recognize patterns in data. Labeled data is like flashcards for machine learning models. In a standard flashcard, you might have a question on one side, and answer on the other. The supervised learning flashcard would be: Data (question) on one side, Label (answer) on the other. An example of a labeled dataset that a teacher might use is a dataset of quiz questions labeled according to difficulty. A supervised machine learning program could use that labeled dataset to analyze a student’s performance in math class over time.
Unsupervised learning
Unsupervised learning is a type of ML that uses unlabeled datasets to allow a program to identify patterns in data without a specific output in mind. Because the dataset doesn’t contain additional information to help the program learn, the program simply searches for patterns. Examples of unsupervised learning in action include grouping different news articles based on their content, segmenting images into landscapes or portraits for an art project, or generating summaries of text documents and highlighting key points for class discussion.
Reinforcement learning
Reinforcement learning is a type of ML that provides feedback to a program to improve its decisions over time. As the name implies, this occurs when the decisions made by an ML program are reinforced.
An example of reinforcement learning is when an AI tool receives positive feedback from its designers when it gives an accurate output. Over time, the program will know to respond to similar requests in the same way. This trial-and-error approach is typically used for tasks that require a series of decisions, such as a writing assistant tool learning to provide more helpful and tailored feedback as students perform writing exercises.
GenAI often uses a combination of supervised, unsupervised, and reinforcement learning. And all three approaches play distinct roles in conversational AI tools, which adapt to conversational context, understand human language requests, engage in natural dialogue, and generate responses in a meaningful way. Here’s how the three types of ML support conversational AI:
-
Supervised learning equips the tools with foundational dialogue data, enabling them to respond appropriately to common communicative cues.
-
Unsupervised learning enables them to interpret nuances in language, such as colloquialisms, that occur naturally in conversation.
-
Reinforcement learning further strengthens them by allowing the AI tools to improve their responses based on user feedback.
Note that, for the purposes of this course, conversational AI tools will be referred to as simply an “AI tool.”
ML advancements helped pave the way for generative AI (genAI) — AI that can generate new content, such as text, images, or other media. The influence of genAI also extends to a wide range of diverse sectors, including drug research and discovery, industrial design, architecture, and fashion. For instance, by generating novel molecule structures, genAI can speed the development of life-saving medications. It empowers the creation of unique product variations and visionary building concepts. Further, designers can tap into genAI to inspire one-of-a-kind patterns and personalize clothing for the perfect fit.
ML in AI tools
Now, discover more about how ML makes AI tools possible. Consider these examples:
Voice-activated AI assistant
Right on your smartphone, a voice-activated AI assistant can set reminders and integrate with your smart devices. To do this, the assistant uses supervised learning models trained on labeled data, such as voice commands paired with corresponding actions.
Navigation tools
Navigation tools use AI to analyze traffic data and suggest optimal routes, predict arrival times, and identify road closures. This is a combination of supervised and unsupervised learning: Supervised learning helps predict the best possible drive using real-time traffic data; meanwhile, unsupervised learning identifies unusual and recurring traffic patterns.
Streaming service recommendations
Streaming services use AI to recommend movies, relying on both supervised and unsupervised learning to curate personalized recommendations. This involves everything from asking your friends for suggestions to considering characteristics of the content itself, such as the genre, actors, or director. It’s like having both your friends and a talented film critic help you choose a movie!
Smart home thermostats
Smart home thermostats identify patterns in your behavior, discovering routines and preferences and adjusting automatically. Unsupervised learning enables occupancy sensors to identify when you’re home or away, and this information optimizes device operation and saves resources.
Financial uses
Financial organizations use AI to identify suspicious transactions, prevent fraud, and protect your accounts. This can involve all three types of ML: Fraud detection models are trained on labeled data of past fraudulent transactions, unlabeled transaction data helps detect unusual patterns or deviations from the norm, and models receive feedback on their flags to improve accuracy over time.
As you continue learning about AI, you’ll understand more about why AI requires human involvement to function properly. For example, ML programs can’t learn independently; they need people to continually update their training. Also, shortcomings in training data can reflect or amplify biases, leading to skewed or unfair outcomes. And, of course, there are many tasks that require a personal touch, such as handling sensitive issues. You’ll explore these issues more in-depth soon; for now, just understand that critical evaluation is essential to address potential challenges and limitations.
Is it AI?
It can be challenging to distinguish AI tools from non-AI tools. The following is a simple breakdown to keep in mind and help you more easily differentiate the two:
Ai Tools:
AI acquires knowledge, then adapts and improves over time. It tackles challenges that have many variables and uncertain outcomes, then makes decisions based on data. And it uses data patterns to make predictions and choose actions, not pre-programmed responses.
Recall the navigation tools analyzing traffic data to suggest optimal routes …
The streaming services that base movie recommendations on your likes and dislikes …
And the financial institutions gaining knowledge from past fraudulent transactions.
Non-Ai Tools:
Rather than learning and adapting, non-AI tools work by executing instructions. Because these systems handle tasks with set rules and predictable outcomes, their responses are predefined.
Think of a simple calculator returning “10” when you enter “5 + 5” …
A sprinkler system that waters a lawn according to a pre-programmed schedule …
And a basic spell-checker identifying misspelled words based on a preprogrammed dictionary.