One technological sticking point that faces online education startups like Coursera, which I profile today in MIT Technology Review’s November digital education report, is how to grade assignments from tens of thousands of students. This inability for computers to easily grade short answer questions, essays, or even drawings sets a limitation on course offerings, especially for humanities and social science classes. Right now, Coursera is solving this by setting up a peer-grading system for some classes so students can evaluate each other’s work.
While artificial intelligence has not advanced to a state that the robo-graders can take over, Coursera might end up solving some of these problems in unexpected ways, including by training a new generation of data scientists.
Consider the case of University of New Orleans’ mechanical engineering major Luis Tandalla, a native of Quito, Ecuador. Tandalla, who is 22, took Coursera co-founder Andrew Ng’s wildly-popular machine learning course online through Stanford University last year, and has since taken several more computer science classes offered through Coursera. He took these classes online, rather than his own university, partly because the teachers were some of the top professors in their fields. “They are the professors that wrote the books,” he says.
With his new knowledge, he entered a contest sponsored by the Hewlett Foundation on the crowdsourcing site Kaggle and this October won the $50,000 top prize. His accomplishment? Writing the algorithm that most improved how accurately computers can grade short-answer essays without human help.
When I spoke with Tandalla, he told me he plans to use the prize money to pursue graduate studies in computer science. In the end, Coursera and other online learning platforms could help train a new crop of computer scientists, like Tandalla, who improve the quality of online education across a far wider range of fields.