Pundits of so-called artificial intelligence are only starting to surprise us with demonstrations of how this new tool can change the way we look at, and think about children’s art.
Computer scientists use different apparatus to achieve classification and analysis of children’s drawings, their elements, or processes that create them. To put it simply, they use digital visual recognition and mathematical models to build deep learning machines. Inspired by observations and studies made by humans, they sometimes compare the types and number of characteristics and categories which computers can process, with that of human scrutiny, in terms of accuracy.
Researchers who aim at developing tools for ever more efficient analysis of children’s drawings, will often prefer providing touchscreens to participants, and leave aside the pen and paper. This is a bit odd because not only drawing on a screen is a far cry from drawing on paper, but also the proportion of children who have access to a touchscreen is and will remain marginal for quite some time. This raises the serious question of whose drawings they are really talking about.
Take for example a study published in Alexandria Engineering Journal (Vol. 60, issue 1) in 2021, titled Classification of children’s drawings strategies on touch-screen of seriation objects using a novel deep learning hybrid model. The article by Dzulfikri Pysal, Said Jadid Abdulkadir, Siti Rohkmah Mohd Shukri, and Hitham Alhussian is available on Science Direct. It concludes that a quantitative analysis of children’s drawing process from a computational system is both faster and more accurate than a qualitative human analysis.
So be it, but one should keep in mind that the study uses born-digital images. We would think that this gives the computer a head start over humans.
For methodological reasons, researchers who develop machine learning systems for drawing analysis often prefer when children draw on screen. Another case in point is a study by Seth Polsley, and four fellow scientists titled Detecting children’s fine motor skills development using machine learning, and published in 2022 in the International Journal of Artificial Intelligence in Education (Issue 32). Here again, the original drawings are created on electronic devices for the study.
There are AI scientists who dare challenge computers to “look” at and analyze children’s drawings created on paper. Ochilbek Rakhmanov, Nwojo Nnanna Agwu, and Steve Adeshina, of the Nile University of Nigeria did just that, in a study titled Experimentation on hand drawn sketches by children to classify draw-a-person test Images in psychology. Their findings were presented at the 2020 33rd International Florida Artificial Intelligence Research Society Conference (FLAIRS 2020). It would be unfair to try and summarize their detailed article here. They sure deserve our respect for offering the only presentation, out of well over a hundred at that conference, to consider children’s drawings worthy of attention. The point is, it is nice to know that children’s drawings on paper can help research on machine learning.
One of the key challenges AI researchers face is the quality and size of data set they have access to. Researchers increasingly resort to online crowd sourcing to amass significant amount of data for their work. One of most accessible such initiative is QuickDraw, created by Jonas Jongejan, Henry Rowley, Takashi Kawashima, Jongmin Kim, and Nick Fox-Gieg, in collaboration with Google Creative Lab. This game invites anyone to help machine learning, simply by drawing on screen. According to Google, 15 M people have submitted 50 M images so far. The tech giant is transparent and upfront about making these images open source material for research.
Advancements into AI seem to already outpace human ability to keep up. Yet, we must try to grasp as much as possible its potential and impacts. The Association for the Advancement of Artificial Intelligence (AAAI) holds its 11th Conference on Human Computation and Crowdsourcing (HCOMP 2023) at Delft University of Technology (Netherlands), this week until November 9th.
