Teaching Machine Learning for K-12 with Teachable Machine
By Yang Cheng
As an instructor at our after-school program ReadyAI Lab, I am talking with students aged 5 to 17 about AI every week. It often surprises me how deeply they’ve thought about AI when they ask questions about automation and the future of work. However, as kids today grow up using AI applications like second nature, I am also seeing firsthand the extent to which they trust, relate to, and rely on AI. It’s critical to teach them about how AI works while they are young. The sooner you open up the black box of AI, the sooner students realize the inherent subjectivity of AI, which, we hope, will lead them to think “how can I do this better?’
In my conversations with teachers, I’ve heard why they’ve found it hard to teach machine learning (especially with younger students). The two common themes have been that 1) the idea of a huge dataset is too abstract and 2) the lack of hands-on examples to teach bias and data quality. With these concerns in mind, we designed this lesson using Google’s Teachable Machine in which students actually play with and actively think about data. Many teachers know Teachable Machine as a great tool but they often feel that materials aren’t classroom-ready. At ReadyAI, we speak the language of lesson plans.
This lesson is unique in that we are drawing a comparison by having students and machines “learn” in parallel. We introduce an unplugged classification game where students are given two sets of images of poisonous and non-poisonous fish. The students are tasked to find out what physical traits make a fish poisonous.
To do so, they compare the two sets of images and try to find commonalities by pattern matching. What they end up with is a set of rules which can be expressed in a decision tree. In our pilot class at the David E. Williams School outside of Pittsburgh, students enjoyed solving this puzzle and constructing their hypothesis.
Once students arrived at their conclusions, they give the same problem to Teachable Machine. By uploading data and watching real machine learning algorithms process right before their eyes, students feel a sense of excitement and empowerment — “ I’m training my own model!” Then, to turn the question over to the machine, we ask them, “who do you think would be better at classification, humans or machines?”
Interestingly, Teachable Machine sometimes gets one of the fish wrong: it says a poisonous fish is not poisonous.
This becomes a powerful opportunity to take a closer look at algorithmic bias. What would it mean in the real world that we predict the wrong category, be it mortgage approvals or job applications? Why does the algorithm get it wrong? In our case, students realize that the traits of the fish in question are not well represented in the pool of poisonous fish.
There are more possible combinations of traits for poisonous fish, therefore, it’s harder for any particular combination to be well represented in a small sample size.
While the students, using human logic reasoning, were able to generalize their decision rules across all fish earlier, the algorithm has trouble with it. Not just students, we all have the perception of machine learning and math being inherently objective. We point out that machine learning algorithms are in fact subjective and its usefulness rests largely on how good and representative the input data is.
With the real-world facial recognition example in mind, students extrapolate about positive-impact applications of machine learning. In our pilot class, students brought up interesting ideas such as a plant identification app or a tumor identification software to free up doctors’ time.
But the end of the lesson isn’t the end of learning! We made an interactive lesson + quiz to further engage with individual students and consolidate their learning. Once they finish the quiz, they will earn a badge to add to their AI Journey, where they see all the courses they’ve completed on our site.
Please check out the full lesson plan here (free after establishing an account!).
If you have any questions, feel free to reach out to me at yang@readyai.org.
