It is an exciting time in the world of technology and innovation. Some very important chapters are being written right now in the development of computers beginning to do things only humans were able to do well so far. Leading scientists, for most of the past century, have imagined building computers that think and do what humans already know how to do well, an idea that we call Artificial Intelligence (AI).1 This has, for the most part, been driven by a desire to make machines think and reason like humans,2 whereas others have been excited by the possibility of humans forever being liberated from doing repetitive and boring tasks and instead being able to focus on the new and exciting. They imagine that this will give our species freedom, power, and knowledge like we have never imagined before.
This has barely been an overnight success story. People have been actively working on this vision since the 1950s,3 if not earlier. But we have only recently made a round of major breakthroughs. A turning point in this journey came in 2010, when, because of a “combination of big data, statistical tooling, hardware,” we arrived in a “vivo state.”4 We reached a level of maturity in technology when some key historical breakthroughs in AI were ready to take flight. These breakthroughs have been a main talking point in human progress for the majority of this last decade.
But These Breakthroughs Haven’t Made AI Smart Enough to Make a Dent in Many of Our Most Important Teaching and Learning Problems
The good news is that some of the very people who were working on that vision for AI in the 1950s were simultaneously or, in some cases, more seriously interested in understanding the inner workings of how the human mind learns.5 Sometimes, working with computers to build AI was an intermediate step to uncovering the mysteries of human learning. And this led to a lot of work over the decades, which you can read a little bit about in Appendix A: A Short History of AI for Improving Teaching and Learning.
But in the preface, I shared my skepticism regarding the power of AI in teaching and learning today. About how we were not there yet. This apprehension in AI’s value in teaching and learning today comes from its limitations. Let’s discuss them.
Many of AI’s advances in recent times are in perception6—the ability to see, hear, or become aware of something through the senses. But we still have a long way to go to fully realize AI’s cognitive capabilities.7 Computers can see, read, and capture sensory data—and recognize patterns on them rather well—but these computers don’t always know what these patterns mean. Having the ability to know and understand something is called cognition.
Andrew Ng, one of the pioneers of this modern era of AI, has come up with a rule of thumb8 to figure out how far AI has advanced. According to Ng, AI’s potential to shine is determined by whether a task takes less than 1 second for a human to accomplish. If a complex task can be broken down into less than 1 second-effort pieces, it could probably be accomplished reasonably accurately by some existing AI technology. And when you think about it, you will realize that a majority of things most of us are confidently able to figure out in less than 1 second fall under the definition of perception.
To solidify this rule of thumb, let’s look at a good example that illustrates how AI’s advances in cognition are weak: AI, today, cannot read a textbook and go on to answer the questions at the back of the book.9 So, current AI technology could literally be deemed to be around as sophisticated as an infant. We know that an infant’s brain is known to have something called core knowledge.10 Core knowledge allows an infant to sense and perceive the basic principles of the physical and social world like objects and sounds and touch. But it doesn’t include many cognitive skills, which the infant has yet to learn. AI, today, is around there.
Fortunately, though, AI’s perception abilities are pretty useful. An important skill in being able to perceive is to recognize patterns. A human’s ability to tell the weather, at a given time, by just looking outside their window, and without any sensation of the air outside, is an example of pattern recognition. Humans have engineered computers to do these kinds of pattern recognition tasks amazingly well, but at many magnitudes faster and cheaper than it would take us humans.
And these pattern recognition capabilities are where AI can begin to solve some of our most painful teaching and learning problems.
Because, As It Turns Out, Teachers Perform Pattern Recognition Hundreds of Times a Day. And That Can Be Exhausting
Yes, teachers recognize patterns all throughout a class day. Sometimes thanklessly, but often endlessly.
Teachers look for patterns in students arriving and leaving the classroom on time and in an orderly manner. Teachers look for patterns in the gaze and attention of students while they lecture or pass out instructions. Teachers look for patterns when independent work is given to students, and often must make split second decisions on whether to intervene or not. Teachers look for patterns in homework and assignment submissions, to see where students consistently struggle.
All this creeps up on them very quickly, contributing to an excessive workload. This leaves them with little or no time to prioritize interventions for student success. Teachers who are unable to make every student shine everyday don’t neglect individual student needs; they actually often just struggle to make sense of the magnitude of information in front of their eyes.
But over several years of teaching, teachers can become really good at these skills. In fact, in my observation, one of the traits that repeatedly stands out in great teachers is the ability to do more pattern recognition and act on it with the least effort.
In the evaluation instrument for Framework for Teaching (FFT),11 also known as the Danielson Framework, one of the most widely used protocols for evaluating and developing effective teaching, author Charlotte Danielson emphasizes that effective teachers are proficient at “responding to and building upon student responses and making use of their ideas.”12
And at a time in history when we face such a dire shortage of teachers, what education reformers are looking for very often is for young teachers entering the classroom to respond to student needs at the pace and ability of highly effective teachers. This takes time, training, and practice.
But you guessed it right. For teachers to get better at such pattern recognition over many years is far from ideal. This is neither a guaranteed natural skill one can master nor does it, in any way, solve the immediacy of the need in classrooms today. I, for one, was a horrible pattern recognizer when I was teaching in the classroom.
It is also something we don’t wish for humans to be doing decades from now: repetitive tasks that take us away from what it means to be truly human. We want to spend more of our precious time helping students, and less of it on figuring out all the different patterns of errors made on tens of questions, while grading stacks of assessments. So, if you want to know where we can begin getting serious about AI in education, this is where you may focus your energy.
Alleviating Teacher Workload from Pattern Recognition Tasks by Using State-of-the-Art AI
We may focus our energy on using the best of what AI has to offer us today to do a lot of pattern recognition. It is that simple. In fact, let me clarify what kinds of pattern recognition tasks I am talking about here, so that we can approach these discussions with the...