- Jason Vearing and Karsten Schulz
Can Student Learning benefit from Machine Learning? Personal thoughts.
Jason: Just recently, I have become the father of a child who has chosen to leave school at this late stage in Year 11. Our son has anxiety, and while he tried his best, it is clear that school is not for him at this stage. At home, he has been teaching himself countries using an interactive mapping system.
He might pick Europe and start choosing countries, and at the end, there is feedback on his % of correctness and what time limit this was achieved. What struck me about this is how intuitively he has approached this program like a Neural Network.
He starts with say 40 – 50% accuracy then he would keep repeating until he can get every country now with 95 - 100% accuracy. There is no emotion about getting the countries wrong, each time he plays he builds an extra few countries to his knowledge base, maybe set up some mnemonics until there comes a point where he gets 95 - 100% accuracy in a region - exactly like Machine Learning. Then he works on the time improvement.
My first thought was 'Why can't he apply himself like this to school subjects?' and the answer seems to be that for whatever reason, he has a fear of getting incorrect answers and low grades and only has one chance to pass exams.
Machine Learning Approach
Could there be an approach where the Machine Learning format is followed? It doesn't matter what the first attempt is, but it is graded and fed back to show where the student is at. Then they have as many attempts as needed to reach a milestone and continue until they have achieved the outcome.
The key here is that the students have to learn not to feel emotionally deflated by a low score; they know they need to repeat and make adjustments to achieve a result. At present (assignments aside), you get one attempt at an exam for ten weeks worth of content.
Weekly Assessment
Could we review the 'read, memorise, repeat' exam model to reflect the current times? Maybe an alternative to exams is that every week there is an online test for competency to understand the week's topics and these all have a 10% weighting each week adding up to 100% weighting for the term. No more frantic revision week and exam blocks.
If you get below say 70% in any week you have to retry before you can even access the second week's assessment. You have as many attempts as needed to achieve a week's competency. This could also help early on in the piece to determine students who aren't cut out for that particular subject. Let's look at this idea in detail:
Cumulative Approach
Week 1: 10 questions online platform that requires a 70% pass rate - students can have more attempts if they want to record a higher mark, overall mark is diluted by the amount of attempts.
Week 2: The 10 questions from week one are asked again as a 'password' access to week 2 material to revise. Then week 2 answers are recorded as a grade.
Week 3: 10 random questions between week 1 and 2 (bias towards week 2) are asked again as password access to week 3 questions. Week 3 answers are recorded as a grade.
Week 4: 10 random questions from week 1 to 3 (bias towards later weeks) are asked to access week 4 questions and so on.
Flaws
There are some flaws which would need to be ironed out. How do we determine 'A grade' students? Perhaps there is recognition for how many attempts there are for those that score 95% or higher on their first attempt or first few attempts.
Another problem could be the people that don't do any work and only do the weekly assessment until they pass. One solution could be that part of the grading is from the teachers observations of class effort.
Some students might screenshot the answers on attempt one - a solution could be that the answers to the incorrect answers are not shown - students will have to research to find the answers.
Constructive Discomfort
I (Karsten) recently heard the term 'constructive discomfort' and that describes that the learning environment should be constructive, but just enough challenging to give a bit of discomfort for the learner to keep going. In a neural network, the back-propagation learning process adds this discomfort. And then the ANN does another round or learning. It will stop when the discomfort has fallen below a pre-set threshold. In my experience, most self-directed learners are like this. They are immensely curious, but generally not satisfied. When children are young learners (pupils), their teachers often have to add the discomfort to get going. But as they become more self-directed learners (students), they can judge whether they need to keep going. As a mature learner, I have come to appreciate the lingering dissatisfaction in the back of my mind as an engine that keeps my learning going. The psychology of the discomfort should be encouraging, but not to lead the students to believe they have already mastered the topic. It is a fine line. But repetition is key to all of this because every repetition cycle extends one's understanding. Repetition is key to learning, but that's not the same as rote learning. However, I have to come to appreciate that my mind has automated many of the mundane tasks of timetables, definitions and equations, so I have them instantly available when I do things. This means I do not need to switch contexts (to do a Google search) when I work. By deeply engraining key knowledge into the deep layers of my brain, it can focus on higher-order thinking.
Conclusion
Machine learning is rapidly changing the world we live in, with computers being able to do some things faster and more efficiently than humans can. Could we also take from Machine Learning the approach itself, and apply it to where it could be demonstrated as an improvement over existing pedagogy?
Going back to the mapping game, if my (Jason's) son had only one chance to get 90% or higher, there would be fear of non-achievement, and he would not engage with the task. Could it be better to let people have as many repetitions of learning outcomes of school assessment as they need?
As with any rethinking of an existing paradigm, there are hurdles to overcome and some areas that are not as effective as the existing model. Our hope in contributing to this blog is to start some conversations and rethinking of how things could be done differently.
About the authors
Jason Vearing is a content creator for Digital Technologies Hub and is also working with Dr Jordan Nguyen on an eye controlled communication device for people affected by ALS.
Dr. Karsten Schulz is the chief nerd at the Digital technologies Institute and father of seven children. He is the creator of the MyComputerbrain AI and the B4 4-bit educational CPU.
Comments