The following information is based on a presentation that was created for the staff of S. Bruce Smith on Artifical Intelligence. That presentation was based on a presentation and lecture by Dr. Jonathan Shaeffer. It is meant to give a basic understanding of what AI is and how it came to be where it is now.
Be warned I will be simplifying a fair bit of history.
The story of AI starts with Alan Turing in 1950 when he asked the question if a computer could think, he disregarded the question and instead focused on whether or not the machine could function in that capacity in a believable way such that a judge couldn’t tell between a human and a machine response. This led to the creation of the Turing Test or the Imitation Game.
Turing, A. (1950). Computing Machinery and Intelligence. Mind, 49, 433–460.
Later in 1955 John McCarthy coined the term “artificial intelligence” as part of a study he proposed to see if a computer could simulate human intelligence. Key word, simulate.
McCarthy: Simulation of intelligence
McCarthy, J., Minsky, M. L., Rochester, N., IBM Corporation, & Shannon, C. E. (1955). A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence.
After these two, not much happens throughout the 60s and 70s in terms of significant developments. This is known as the First AI Winter
Interest spikes in the 80s computers enter the workforce and automate several tasks, generate lots of science fiction.
The 90s is another cooler period for AI development, often called the Second AI Winter.
In the mid 90s to early 2010s there is interest in data science and the starts of practical AI but we lacked the computers, data sets to execute.
2010, often called the beginning of the modern era.
The term “artificial intelligence” is a challenging label to use since it has many other technologies nestled underneath it, but it is the term in general use. This requires untangling terms that are currently being used synonymously when in reality they are related but different.
One of the other many challenges with AI is debunking the myth about its abilities. AI and by extension computers are not an artificial brain as it does not function the same way as human cognition, relying on pattern recognition, large data sets and algorithms to masquerade as a human mind (Mintz et al., 2023; Schaeffer, 2023).
Data Science: An interdisciplinary field that combines math, statistics, specialized programming, advanced analytics, artificial intelligence (AI), and machine learning with specific subject matter expertise to uncover actionable insights hidden in an organization’s data, guiding decision-making and strategic planning.
Artificial Intelligence (AI): A field of study on creating machines that mimic human logic.
Machine Learning (ML): A part of AI where machines are “taught to learn” patterns from data on their own, much like how we learn from experience.
Artificial Neural Network: Inspired by how neurons in the brain interact to perform a specific task. They include an input layer, a series of hidden layers and then an output layer. They can recognize patterns based on training data—for example, image recognition, sounds, text natural language and speech recognition.
Deep Learning: Neural network models with multiple hidden layers (more than 3 layers) and multiple nodes in each hidden layer. Provides a higher accuracy and performance compared with simple 3-layer Neural Networks. l
Generative Models: ML systems that can generate new data similar to the data they were trained on. They may use Deep Learning training data as a base.
Large Language Models (LLM): A generative model designed to handle text data.
Autoregressive Models: A type of LLM that generates sequences, like sentences, one piece at a time, similar to how we speak or write.
Generated Pre-trained Transformers (GPT): A type of autoregressive model developed by OpenAI. They are called ‘generative’ because they generate output, ‘pre-trained’ since they are trained on a lot of text data before they can be used, and ‘transformers’ as they process data by focusing on different parts of the input to generate an output.
(IBM, 2022; Kainat, 2023; Klingler, 2023; Madden, 2023; Microsoft, 2024)
What we call “AI” is a trained computer program. According to the definition of AI in education and learning (AIED), we have been using these types of systems for longer than we have recognized. (Holmes & Toumi, 2022)
It cannot learn in the traditional sense, humans reduce the error rate with increased computer processing power. Currently the term "AI" is being used to describe machine learning, deep learning, large language models and other assistive technologies, making several conversations around the concept somewhat challenging.
For a computer to “learn” to play checkers, it needs to play millions of games.
For a human to learn, maybe a dozen. A few hundred or thousand to be “good.”
For a computer to win at checkers, it needs to play every game and did in 1994. (40 trillion positions)
This isn’t achievable for a person which is where the machine takes over, doing these tasks in a fraction of the time needed by a human. Computers can calculate faster than we can, but it still cannot learn like us.
"The word “artificial intelligence” is a misnomer. All artificial intelligence, including generative AI, is merely a set of complex algorithms. But unlike human cognition, computers can’t think. They process information. Humans think. They generate content. We create.
There’s a difference."
Arthur Walker, a science fiction writer and game designer came up with the following analogy in 2011 when he began researching AI. He wanted to give himself a starting point for thinking about sapience relative to humans and completing tasks for humans.
"A thermostat with a mercury switch is technically an intelligent agent, reacting to stimulus, automating a task for humans."
Before the mercury switch, furnaces were stoked by hand and a temperature gauge was monitored by a human.
The switch took over both roles.
A modern, “smart” thermostat could be included in the current definition of AI as it can “learn” your patterns and know when to heat and cool your home.
This allows us to explore AI through a familiar lens and experience that the majority of us are comfortable with in terms of acceptance. One question we need to consider is where are we in the "thermostat evolutionary timeline" regarding AI in Education.
Computers operate on processing instructions. These instructions are sometimes categorized as a "flop." Due to the speed of how fast computers process instructions, they are usually talked about as petaflops or "1015 flops". As you can see from the graph on the right, there has been a recent and massive increase to the ability for these kinds of instructions to be processed, which has helped produce more complex forms of "AI" like large language models.
Not only has speed played a factor, but the overall capabilities of computers and their reduced cost shown the graphs below illustrate the width and depth of access that now exists for this computing power.
Moore’s Law: The number of transistors on computer chips doubles approximately every two years.
Origin: First described by Gordon E. Moore, the co-founder of Intel, in 1965.
Cost: The cost of computers and storage and dropped significantly over time.
Every minute (2022)
231.4M Emails
5.9M Google Searches
2.43M Snaps
347.2K Tweets
$443K ordered on Amazon
104,6K hours of Zoom Meetings
Every minute (2023)
241M Emails
6.3M Google Searches
$455k ordered on Amazon
102MB of data per person
The majority of this data, as well as the rest of the internet, is being used to train large language models like Chapt-GPT. The more data, the more potentially "accurate" the model can become.
To see more data facts, click the links below in the reference to see the full-sized infographic.
Supervised Learning
Uses labeled datasets to train algorithms to classify data or predict outcomes accurately.
i.e. handwriting recognition, email filtering for spam, or the answer key in Google Forms quiz mode
Unsupervised Learning
Similar to supervised learning but lacking a critic. Uses classes to categorize data on hidden features.
i.e. recommender systems in Spotify or product suggestions in Amazon (clustering)
Reinforcement Learning
The algorithm attempts to learn actions for a given set of states that lead to a goal state. Errors are received after a reinforcement signal.
i.e. an ideal algorithm that can learn how to make decisions in an uncertain environment
Machine learning has helped computer programs recognize patterns. By training on large sets of data and being told by humans what certain objects look like, it is able to "train" itself to recognize these objects. These could be people, animals, objects or even medical conditions on MRIs or X-rays.
Let’s consider how image recognition works through this example of identifying an elephant.
The individual lines on their own need to be grouped together through a hidden and complex layer of choices and patterns to form possible conclusions. This hidden layer has got the point where it does not need supervision.
As these pieces connect, the picture becomes clearer and the likelihood of correct identification increases.
The end result can then be verified by the human that the lines form an elephant and it will attempt to identify the object.
This can be seen in the image on the right. These images will often have a number from 0 to 1 by the box to indicate the likelihood of correct identification. 0.9 for example would be a high likelihood.
AI has become a "buzz word" and a term of business. Many things are said to include AI that may not actually include it to attract new customers or overpromise what their technology can deliver.
This flowchart on the right can be a tool we can use to determine if AI might be present.
This slide deck is part of the Amii K-12 Learning Kits. It is designed to be used with students but it is a good primer for teachers as well.
To learn more about their resources and access the full kits, please visit:
For an overall comprehensive guide on AI from top to bottom in easy to understand terms, check out this website to the left.
For more information on how AI is trained, check out this article created by a researcher at IBM that describes the three main types: supervised, unsupervised and reinforcement learning.
For more information on the differences between ChatGPT 3.5 vs 4, check out the article on the left.