Posted on

Why Schools Shouldn’t Teach AI

One of the earliest technologies that gained wide user adoption was the calculator. Pocket size  versions of the calculator became available in the 1970s and it didn’t take long before people started wondering whether children should just learn to use the calculator instead of learning mental mathematics. 

Thankfully, the educational system continued to teach basic arithmetic to young students for a very good reason. Students don’t just need to understand the concept of addition or multiplication. They need to practice arithmetic over and over again in order to build and strengthen the right neuronal connections for computational fluency. High computational fluency is correlated with better performance in more advanced math, so doing there drills in early years sets the foundation over which more complex mathematical thinking can be built. 

With the coming of AI, especially Large Language Models (LLMs), we are seeing a similar debate take place. The popular line of reasoning — AI is undoubtedly going to be a big part of our lives going forward and students need to learn how to use it — makes logical sense. 

But perhaps that’s precisely the reason why we shouldn’t teach students to use AI.

What Is The Goal Of Education?

There are different philosophical views about the goal of education, but most people would agree that education is about building the right skills so students can thrive in the real world. The point is never to teach students about all possible tasks that they might have to do, but to build enough of the right foundational skills that will allow them to adapt, learn new things and make meaningful contributions. This is why higher order skills, like creative and critical thinking, feature on the top rungs of Bloom’s taxonomy, a hierarchical model of learning goals that guides our educational philosophy. This makes sense because it is nearly impossible to predict the kinds of jobs that will exist a decade or two later. So teaching specific domains is less important than building higher order cognitive skills that will allow students to be successful, no matter what tasks they might have to face. 

Cognitive Complexity

Consider the two questions below: the first one is taken from College Board’s question for AP World History while the second task asks you to create an AI prompt. 

Task 1:

Directions: Question 1 is based on the accompanying documents. The documents have been edited for the purpose of this exercise. 

Evaluate the extent to which communist rule transformed Soviet and/or Chinese societies in the period circa 1930–1990.

(See the full question for accompanying documents)

Task 2: 

Create a prompt for ChatGPT to help you understand the significance of the American Civil War.

Which of these tasks, in your opinion, is harder?

Most people would agree that the first task is way more complex than the second one. 

To complete the first task, students have to quickly read the documents provided, make notes on the historical context, audience, perspective and purpose. As a side note, the documents are varied in nature (e.g. it could be a propaganda poster or a diary entry) and not directly related to the question. Using the provided material as evidence, students have to construct a defensible thesis. They also have to choose an outside piece of evidence that adds an additional dimension, and provide a broader historical context relevant to the prompt. In short, they have to make reasonable inferences from the provided documents and integrate prior knowledge they have about the subject to create a compelling argument. And finally they have to synthesize all of this into a multi-paragraph essay! (Big thanks to my son, who recently took the AP World History test, for helping me understand what goes into answering this type of question).

For the second task, students simply have to ask ChatGPT to explain the significance of the American Civil War! Even if they want to become more sophisticated with their prompting, the strategies (e.g. role-play or providing additional context) are pretty straightforward and easy to learn. 

It’s no wonder the AP exam gives students an entire hour to answer the first question, while the second one can be done in just a few minutes.

A more telling sign of the low cognitive complexity of using LLMs, is how quickly people from all ages and walks of life have adopted LLMs to assist them with work tasks. In contrast, not many people have attempted coding, despite the many tools and tutorials that have been available for many years. Teaching STEM skills in schools continues to be a challenge due to the lack of qualified teachers. 

Teaching students to tackle complex tasks like the first one has a higher payoff than teaching them simpler tasks. After all, if our education system helps students develop critical thinking skills and the ability to handle tough questions, then easy tasks like the second one will be a breeze for them. Much like the calculator, students don’t need to be taught how to prompt LLMs to get an answer – they need to build skills to answer it themselves. 

What Aspects Of AI Should We Teach Students?

It’s clear that teaching students to use LLMs isn’t very useful because we would almost certainly replace a more complex learning goal with a relatively trivial one. 

A more worthy goal of teaching AI would be to teach them how AI works so students can build AI models as opposed to simply using them. But this brings us back to square one, because that requires a grounding in programming and computer science fundamentals. So focusing on STEM, coding and computational thinking in the K-12 curriculum is still the right approach to prepare students for careers in technical fields. 

Beyond the technical aspects of AI, it’s probably more important for students to understand how to learn better and how AI can impact the learning process. While there are a few scenarios where AI can improve the learning process, there are also learning traps that can harm creative and critical thinking in the long run. Or, they might be better served in building their ethical reasoning skills to tackle the numerous challenges that are bound to arise as AI usage becomes more widespread in society. 

All of those would require them to think harder and deeper than simply learning how to use AI. 

Posted on

Will AI Make Us Dumber?

Back in the early 1980s, a researcher named James Flynn was studying old IQ tests when he noticed something strange. Between 1932 and 1978, IQ scores had shot up by almost 14 points! This became known as the “Flynn Effect,” and researchers kept seeing IQ scores rise by about 3 points every ten years. But then, in the early 2000s, things took a turn. Studies in a few Scandinavian countries found that IQ scores were actually dropping. And get this – a recent US study confirmed the trend, with the biggest decline in young adults aged 18-22.

This drop in critical thinking skills is like what happened to creative thinking scores over a decade ago. Professor Kyung Hee Kim discovered that student creativity, as measured by the Torrance Test of Creative Thinking (TTCT), has been tanking since the 1990s. Her analysis even led to a super popular Newsweek article called The Creativity Crisis. She found that important aspects of creativity, like coming up with original ideas and being able to brainstorm lots of ideas, have seriously declined over the years. And the decline has gotten even worse in the last ten years.

So, what’s changed that’s causing this decline in higher-order thinking skills? It’s not a huge leap to think that the rise of technology over the past couple of decades might have something to do with it. Both studies suggest that certain aspects of technology can actually hinder the development of creative and critical thinking skills.

With AI rapidly gaining adoption — over 45% of students in high school already use AI to help with their assignments — what are the possible implications on cognitive development of children. 

How We Learn

To understand how thinking is essential for learning, let’s take a closer look at what happens in our brains when we encounter new information.  Learning can be seen as a three-step process. First, our brain encodes new information and stores it in our working memory. It’s like translating a high-resolution image of an apple into a simpler icon or symbol. Next, this information might need to be moved to our long-term memory. Imagine you see a purple apple for the first time. Your brain needs to pull up your existing mental image of an apple, update it, and then store it back. This shows that working memory isn’t just storage – it’s more like a mini-computer with multiple functions. The latest model of working memory includes three specialized memory subunits and an executive controller that manages attention and communication with other parts of our brain.

One memory subunit is the visuo-spatial sketchpad, which lets you hold and modify visual concepts. For example, try picturing a red apple, then change its color to green and add some leaves to the stem. If you could do this, it all happened in your visuo-spatial sketchpad. Another subunit, the phonological loop, processes auditory information. It has a small temporary memory that needs constant refreshing.  When someone tells you their phone number and you repeat it to remember it, you’re using the phonological loop.  Working memory is where the magic of creativity and learning happens. It creates specialized models to speed up future processing. By comparing, contrasting, and finding connections, our working memory helps us understand the world better. It’s also where new ideas are born.

The challenge is that deep thinking takes time. Learning something new or creating something original means changing your internal mental models, and that requires serious brainpower.

Biological Bandwidth of Learning

The late MIT professor, Patrick Winston, was renowned for his lectures, especially his super popular annual talk, “How to Speak.” One of his top tips? Ditch the slides and use the board. Slides are good for exposing people to a topic, but the board is better for informing them. Why? Because writing on the board forces you to slow down, giving your audience more time to process what you’re saying. As Winston puts it, “The speed with which you write on the blackboard is approximately the speed at which people can absorb ideas.” 

This simple trick perfectly illustrates the concept of “biological bandwidth” when it comes to learning. Your brain needs time to absorb and store new information—kind of like how long it takes to physically write stuff down.

So, what happens when we try to cram too much information into our brains too quickly? Our working memory doesn’t have time to process everything properly. Instead of analyzing, comparing, and updating our mental models, it takes shortcuts, relying on things like stereotypes or gut feelings. This leads to quick decisions, but they’re often wrong because they’re not based on solid reasoning.

How AI Can Make Thinking Skills Worse

The problems with AI are similar to the problems of relying on too much tech, but way worse. It all boils down to “cognitive offloading,” which is when we let AI or other tech do the thinking for us. If we keep this up, our own critical and creative thinking skills will get rusty, and could even disappear for good. It’s not hard to imagine how bad that would be for society.

Students are hit even harder. During the teen years, our brains are growing and rewiring big time, especially in the prefrontal cortex. If we don’t use certain abilities, our brains just prune those connections away. So, if you’re not flexing those creative and critical thinking muscles, you might not be able to think as deeply as an adult.

Lots of schools and students are already using AI, and while it can be helpful in many scenarios, there are also some traps that are easy to fall into.

The Effort Trap

It’s obvious that using AI to write an essay without putting in any effort is bad for learning. But there are also sneakier ways that AI can trick us into thinking less. This is the effort trap: when you think you’re being critical and analyzing AI’s output, but you’re actually using less brainpower, or even thinking in a totally different way.

Here’s an example that can mess with your creativity in the long run. People often use AI to generate ideas, which they then refine. But if they had taken the time to think for themselves first, they might have come up with totally different ideas. Coming up with that initial spark is an important skill in itself, especially in ambiguous situations. By relying on AI, we’re short-circuiting our own creativity.

The Competence Trap

The competence trap is when you think you’re a pro at something, but you’re really just leaning on AI as a crutch. This can trip up both students and teachers. Teachers might think their students are killing it based on the AI-polished work they see, and move on to harder stuff before the students are actually ready.

The Capacity Trap

When information is super easy to get, it’s tempting to consume way more than our brains can handle. We’ve all been in lectures where we thought we understood everything, but then got totally lost when it was time to do the work. This is the capacity trap: it’s easy to keep chugging along without stopping to reflect, but that can lead to a big crash later on.

Making AI Work For You

AI can definitely help us be more productive and learn new things. For example, getting instant feedback from AI is very impactful because you can make changes while everything is still fresh in your mind. Waiting a week for feedback on an essay isn’t as effective because you have to switch gears and get back into that mindset. Also, teachers can use AI to personalize lessons for each student, which helps them stay engaged and understand the material better.

So, how can we use AI in a way that actually helps our thinking skills instead of hurting them?

The key is to be mindful of how much thinking you’re doing yourself and how much you’re leaving to AI. A good rule of thumb is to avoid using AI as a crutch right from the start. Instead, think about the problem first and try to come up with a solution on your own. Then, you can use AI to help you improve your work or get unstuck if you need to.