AI Programming by Children


University of Oxford


Henry Lieberman

Abstract. The idea of children constructing AI programs is about fifty
years old. When Seymour Papert and colleagues designed and implemented
the Logo programming language the kinds of example projects by
children that they suggested included robotics, natural language
processing, and more. In addition to the exposure to the powerful
ideas associated with constructing computer programs, the children
could become more reflective about their own thinking. Recently
Stephen Wolfram added a machine learning chapter to a high school
Mathematica textbook and then blogged about how middle schoolers might
program machine learning. Dale Lane recently created the Machine
Learning for Kids website where children can train a model with texts,
images, or numbers and use that model in the Scratch programming
language. Stefania Druga at the Media Lab has been developing
Cognimates Platform for AI education. As part of the eCraft2Learn
project the speaker has enhanced the Snap! programming language with
blocks for speech input and output, image recognition, machine
learning, and word embeddings. A live demo will be presented.

Bio: Ken Kahn did his doctoral research at the MIT AI Lab where he was
first exposed to the ideas of Seymour Papert and Marvin Minsky about
how the proper use of computers could change education dramatically.
After 12 years of research in programming languages and AI he returned
to research on computational learning environments. He designed and
developed ToonTalk - a programming environment for children where
programs are created by demonstration in a game-like virtual world. As
part of the European eCraft2Learn project, he has returned to his
interest in children creating AI programs.