Stanford University’s Lytics Lab has developed a solution to grading that could make the sleep-starved Teaching Assistant a thing of the past. A software program called Codewebs can analyze the programs written by thousands of students in one pass. The software is expected to increase the efficiency of administering massive open online courses (MOOCs) writes NewScientist’s Hal Hodson.
Codewebs bundles student submissions according to the similarity of their code, allowing an instructor to write appropriate feedback for multiple submissions at one time, and deliver that feedback simultaneously. This bundling can be done at the level of the entire submission, or on individual code sections for more targeted feedback.
Codewebs, currently in use with MOOC provider Coursera, operates by running machine-learning algorithims to index snippets of each code submission. The program then creates a database that can build those snippets back up according to similarity when the instructor is ready to provide feedback.
“The program understands assignments incredibly well,” says Chris Piech, a Stanford researcher working on Codewebs. “If a student hands in homework, not only does it say, ‘Good job, you solved it like your peers’, but if it looks like the student is solving the problem in a way that’s detrimental to their learning, we could give feedback to push them away from that.”
The Stanford lab isn’t the only foray into the artificial teaching assistant. The Swiss Federal Institute of Technology is also developing a code-analysis program to assist in administering courses on the Scala programming language. The Swiss entry relies on keyword matching, delivering fast feedback to student submissions to encourage their engagement.
Programming code isn’t the only target of the AI Teaching Assistant movement. Sanjit Seshia of UC Berkeley is developing a MOOC to offer instruction in robot design in a virtual lab, with software tutors a critical piece of student feedback on construction and design.
Technological learning may not be the limit of the AI/MOOC movement either.
Piech is also considering ways to expand the scope of Codewebs. He started with computer science because of the huge global demand for coding skills, but the program would work for other quantitative subjects. “Math is the next frontier,” he says.
“There’s a chronic shortage of people who can teach computer science across the world, not just in Silicon Valley,” says Piech, who was born in Kenya. “Places like Kenya also need programmers.”
The development and future feasibility of MOOCs may owe a large debt to time-savers of this type. The only victim of the tale, the scruffy grad student left behind with nothing to grade.