AI Fundamentals for Educators

AI will have the same transformative impact on society as cars. Most drivers don't know how their car works but they use them constantly And even if you know how a car works, it doesn't stop the externalities.. Personal vehicles transformed American culture and shaped the constructed world around us, for better or for worse. Teachers need to know the fundamentals of AI-driven programming to best prepare their students for the world they will inhabit. Just because we can drive almost anywhere doesn't mean it isn't worth it to walk once and a while.

What Even Is A Computer? What Even Is Intelligence?

AI programs work on fundamentally the same hardware that any other computer does, albeit with different scales or priorities Oversimplification, but AI programming prefers hardware that can multitask simple processes in parallel rather than tackling a single complex problem at a time.. So to understand the strengths and limits of AI code, it is essential to remember the strength and limits of computers in general. In its code, a computer is an adding machine What is addition if not repeated counting? What is multiplication if not repeated addition? What are subtraction and division if not reversed addition and multiplication? Most of the math we do is just fancy counting!. Modern computers appear to go beyond addition by storing specific mathematical processes and routines; it's similar to the progression in how math processes are taught from adding to multiplying to taking averages and so on. So then how can a superpowered calculator end up writing an essay or generating a video? It doesn’t, it tricks us into thinking it did!

What makes human intelligence special is that we can transform data into information. Data is simply that which is given, like beads on a rod or lights on a screen. Intelligence is being able to understand that as numbers counted on an abacus or the display on a calculator. Right now you are looking at a grid of lights in different colors, that’s all, and that’s all the computer needs to know. But you understand the patterns of these lights as letters, the letters as words, words to sentences, and you are creating information from that data. Computers, and thus AI programs, create displays of data and your brain does the heavy lifting of turning those patterns into information.

Teaching a Machine

How does a computer make such convincing patterns in data? It finds the patterns that have been there all along! AI programming relies on using a large amount of data to “train” the computer program to output data that looks similar, but not identical to, the original data. I could never do a more efficient job of explaining it compared to this CGP Grey explainer:

An AI program is code that uses statistical processes to predict outcomes that resemble what was present in the training data. There are two major takeaways here: these programs are using probability and they are fully reliant on the data used to set those odds. Probability is a tricky thing; it’s easy to talk about the odds of a coin toss but you have never seen a coin land on 50% heads and 50% tails I frequently tell my students that I beleive that probability does not exist! But like the concept of energy in Science, it is still important to study.. We are used to computers providing answers with certainty but AI methods add an element of chance to these trusted processes.

It's All Data

The datasets that are used in the training process serve as the "answer key” for the program output, which means the programs can only be as accurate as the data that feeds it. There is an abundance of data available out there, but quantity is not quality nor is it accuracy. Data comes from humans and humans carry biases and flaws. Scale and mathematics do not clean these issues out of a dataset, they exacerbate it This is most apparent in the way language models treat names and other proper nouns that carry connotations, like this instance with student names. It can also cause problems with less popular names and concepts that do not appear in data as much..

The prevalence of AI-driven technology has also created an enormous push for more data collection. There is a demand to datafy everything, and that changes the way we interact with technology. It can be frustrating, but it also provides us with a lens to better understand what AI can actually mean. When you see something “powered by AI”, ask yourself what data it is relying on. What patterns does the code rely on to generate outputs? Is this a situation where you want the most popular answer - because that is what the statistics of the program will lean towards.

For Teachers

Teaching about AI is no different than teaching about any other information technology. As teachers we need to emphasize critical consumption of information online, whether it is from an AI program or any other source. The humanized presentation that AI applications use make them extremely convincing Digital tutors are probably the most aggressive examples of this. Nothing in their code makes them accurate but it can read like an authoritative figure. Chatbots are also a significant issue. and so it is essential that we prepare students to approach this information with a skeptical lens. It doesn’t mean they need to automatically assume everything is wrong, but there should always be a deliberate effort to verify information beyond just feeling like it’s correct.

A more difficult task that we should aspire to is to do our best to not anthropomorphize technology, especially AI programs. When we give human qualities to computers, it changes the way we evaluate the data they present. It creates an artificial sense of trust and effort when the human elements of the technology are much farther removed. We can’t forget the humans that made the hardware and software we frequent, but that does not mean that their creations are the same as the creators. This is an uphill battle Media is also guilty of making this problem nearly impossible to avoid., especially with the prevalence of AI agents and assistants marketed to appear mortal, but the effort is essential for our students’ understanding.

My Hot Take

AI should not save you any time. Instead it should improve the work you are capable of in the same amount of time. Any AI generated output lacks veracity, you need to put in the work to proofread and verify anything from writing to images to code. If you use AI to give feedback to students, you still need to go through every page to make sure it is right and to see where your class is headed.

If AI does cut down on the time something takes, that task probably did not require that much energy anyways. Maybe that email could have just been one sentence. That slideshow didn’t need images on every single page. It would have been faster to read an article than to listen to a generated voice. Maybe you missed something important when you read only the summary.

When I started teaching, assigning a poster project would be a multiday event. Accessible graphic design tools can cut the visual elements down and I get a choice - do I make this a one day assignment or ask for higher quality work? This is what AI does to our practice across mediums.

More Resources to Explore

For a more technical dive into AI, it is tough to beat Stephen Wolfram's analysis. 3Blue1Brown's series on neural networks and LLM also presents similar depth in video form.