AI and Classroom Connections

My journey with generative AI started when…

my friend introduced me to an incredible technology called Stable Diffusion back in October of 2022. I began exploring diffusion models by using a powerful web interface, DreamStudio. With DreamStudio, I could effortlessly craft any image that my imagination conjured up. It felt like the sky was the limit, and my creativity soared to new heights. As I delved deeper into the possibilities of Stable Diffusion, I stumbled upon an intriguing realization: this technology could be used to create animations as well. Eager to try out this new technology, I decided to create a video to accompany my nephew’s song. This is the video you can see below and my first attempt at animating with Stable Diffusion.

AI, short for Artificial Intelligence, is a field within Computer Science that aims to mimic the way humans think and process information. In simpler terms, AI tries to create machines or programs that can think and learn like us.

AI systems are designed to understand, combine, and draw conclusions from information. This allows them to make decisions or solve problems just like a human would.

Machine learning is a subset of AI that focuses on how machines can learn from data and experiences. It uses special techniques, called algorithms, to help machines adapt and improve over time, much like how people learn from their experiences.

Narrow AI (also known as weak AI) refers to artificial intelligence systems that are designed and trained for a specific task or a limited set of tasks. These systems can be very good at performing the tasks they were designed for but lack the ability to understand or learn tasks outside of their scope. All of the AI models today are narrow AI.

General AI (also known as strong AI or artificial general intelligence) refers to a hypothetical AI system that can perform any intellectual task that a human can do. This is hypothetical because no one has yet created AGI (artificial general intelligence). It has the ability to understand, learn, and adapt to new tasks and situations. General AI systems would possess advanced cognitive abilities, such as reasoning, problem-solving, planning, learning, and understanding natural language, enabling them to perform a wide range of tasks.

As of my knowledge cutoff in September 2021, artificial general intelligence has not yet been achieved. Research and development in AI are still mostly focused on narrow AI applications, although some researchers are working on approaches that may eventually lead to general AI.

Our only example of artificial general intelligence come from movies because at this point it is still hypothetical. In Ex Machina, Ava would be an example of AGI and in Her, Samantha, the operating system was an example of AGI.

Before GPT-4 from OpenAI was released to the public, a number of Microsoft researchers had a chance to play around with the model before safety restrictions were added. They wrote a paper entitled, “Encounters with AGI”, before they renamed it to, “Sparks of AGI”. With either title, it is exciting to see the beginning of AGI development. The video above does a great job explaining this paper.

As creative AI improves, it will be harder and harder to distinguish between content created by humans and content generated by AI. Above is an example of a music track J. Medeiros featuring Jay-Z, but in this track Jay-Z’s voice is generated by AI. Even though I know it is generated with AI, I can’t hear the difference between the AI generated voice and Jay-Z’s real voice. This is an example of why we need to be discussing this technology and introducing it to our students. They need to be prepared, as best they can, for a world where deep fakes can be made with the click of a button.

Using these three characteristics that are found in every AI model, we develop vocabulary and we begin to understand how these models work.

1. Data Sets

A data set is a collection of information that an AI uses to learn and make decisions. It's like the fuel that powers the AI. The data set can include things like text, images, sounds, or any other type of data. The better and more diverse the data set, the better the AI can understand and make decisions based on that data.

For example, if you want an AI to recognize different types of animals, you would provide it with a data set containing images of different animals along with their names. The AI would then use this data to learn which animal is which.

2. Rules for Using the Data

Rules are like the instructions that guide how the AI processes the information from the data set. These rules can be simple, like "if this, then that" statements, or more complex mathematical equations. AI developers create algorithms (a series of rules) that the AI follows to make sense of the data it receives.

For example, in our animal recognition AI, the rules could be designed to identify specific features (like fur, fins, or feathers) and use that information to determine the type of animal in a given image.

3. Ability to Reason

The ability to reason is what allows the AI to make decisions or draw conclusions based on the data and rules it has learned. This is what makes AI "intelligent." The AI uses its reasoning ability to analyze new information and make predictions or decisions based on its past experiences (the data and rules it has learned).

For example, after our animal recognition AI has learned from the data set and the rules for using the data, it may come across a new image of an animal it has never seen before. Using its ability to reason, the AI will analyze the new image and make an educated guess about what type of animal it is based on the features it recognizes.

Quick draw is an older website and AI model but I like to use it with my students when I explain about data sets. They have the largest doodling data set in the world (50 million drawings), it is open source, and it is very visual. My students and I usually play the game first to understand how it works, and then we look at all the images used to train the model.

QucikDraw: https://quickdraw.withgoogle.com/

Teachable Machine is a great website to begin to understand how the rules are created and the model is able to reason. You can take images using your web cam and set it to a category. This helps you create a very simple image classifier. This is a fun activity with the students, but at the end I have them reflect, “What rules were set for the data? How was the model able to reason?”

The Teachable Machine: https://teachablemachine.withgoogle.com/