Does Any AI-Based Activity Contribute To Develop AI Conception? A Case Study With Italian Fifth and Sixth Grade Classes
(Matteo Baldoni, Cristina Baroglio, Monica Bucciarelli, Sara Capecchi, Elena Gandolfi, Cristina Gena, Francesco Ian, Elisa Marengo, Roberto Micalizio, Amon Rapp, Ivan Nabil Ras)

This paper is useful because it has much better circumstances than many others and is therefore able to draw more general and more accurate conclusions. As opposed to studies performed in "hothouse" environments, e.g. with 8 students in a voluntary after-school camp at a STEM-focused school, this study spans an entire school year, involves over 200 children from two different schools, and even includes a control group who attended the programming portion of the curriculum while skipping the modules concerning artificial intelligence.

One sentence in particular stood out to me: "To foster AI literacy, however, AI tools alone are not sufficient, and structured AI curricula are instead needed." This reflects the seemingly common idea of not reinventing the wheel. It is all well and good to create new programs to teach children with, but without carefully structured curricula in place, children are not going to actually learn anything. 

What was perhaps most interesting about this study was its results and conclusions. It found that in both tests, "Computational Thinking" and "Conception of Artificial and Human Minds," the programming-only students did just as well as the programming-and-AI students. They also found that there was no significant improvement in the "Conception of Artificial and Human Minds" test in either group, meaning that both groups were still equally prone to ascribe human qualities to AI. They attribute this result to a fault in the course, saying that it "may facilitate the acquisition of competences (e.g., programming constructs and the usage of AI libraries), rather than an understanding of the functioning of programmed intelligent behaviours." They conclude that more effort should be put into uncovering black boxes, as in Hitron et al. 2019, and emphasize the importance of children building "accurate mental models" of AI.