The Future of Grading
Ask anyone who’s been a teacher or a TA — grading is the most painful part of their job. It ties you down to a stack of paper, takes forever, and requires a lot of effort to give fair and personalized feedback to students.
We experienced this first-hand in teaching Artificial Intelligence at UC Berkeley. Like many instructors, we were frustrated that a key part of our job involved spending hundreds of hours shuffling papers. As Computer Science grad students, we wanted to do something about it: what if we digitized our students’ work and graded everything online? We would be able to work from anywhere, grade entirely in parallel, and ensure that our grading is consistent.
We quickly built a prototype for use in our course. Within months, most CS courses at Berkeley were using it, as well as instructors at MIT, Stanford, CMU. To date, over eight million questions from a hundred thousand students have been graded on Gradescope.
Instructors report that grading with Gradescope takes about half the time of grading on paper. With our background in AI, we believe that we can get it down to zero — allowing instructors to spend that time with their students instead. And because grading is done online, we are also able to help instructors understand exactly what every student knows at every stage in the course.
Get grading time to zero
Our graduate research in robotics, computer vision, and machine learning has focused on one question: how can machines learn to perform intelligent tasks? Deep learning is the current best approach. It’s an exceptionally powerful technique that understands speech on our phones, recognizes faces in pictures, and reads handwriting on checks.
Pieter’s lab has established deep learning as the leading edge of robotics research, and Sergey is one of the core contributors to Caffe, one of the most popular open-source deep learning frameworks. We believe it’s possible to use deep learning to reduce time spent grading from hours to minutes — or even seconds — without compromising quality.
We call it “AI-assisted grading,” where machines assist humans in grading consistently and quickly. For many questions, Gradescope’s AI will be able to learn how to grade all student submissions from a small number of answers graded by the instructor, such that an instructor would only have to grade about ten answers out of a hundred submissions. For known questions, Gradescope’s AI will be able to grade instantly, without any human input.
Later this year, we’ll release our first version of AI-assisted grading for simple question types, including multiple-choice, numerical, and single-word answers. Over time, we will assist instructors in grading chemical diagrams, engineering drawings, math proofs, and hundreds of other types of questions.
Our long-term vision is that instructors will need to spend at most a few minutes grading highly complex assignments, while providing granular and actionable feedback to students and understanding exactly what they know.
Understand what every student knows
Traditional grading generates an enormous amount of data: from total scores to specific reasons behind every point earned. Normally, all this data is simply lost to the instructors, because it is confined to paper that gets handed back to the student. The final grade is just a one-dimensional representation of a complex body of work — and yet it’s the only thing that gets recorded.
With Gradescope, every grading action is available for search and analytics. With this data, we can help instructors understand whether the question is confusing, or if the students did not understand the concept it was testing. We can track student performance from assignment to assignment, from course to course.
Students cannot easily improve unless they understand what they need to improve on. Likewise, instructors cannot easily adjust their teaching unless they understand what their students are struggling with. Gradescope enables improvement by helping instructors and students understand the underlying reasons for every point on every assignment.
The road ahead
Today, we’re excited to announce that we raised $2.6M in a financing round led by people who believe in our mission to bring education into the digital age by starting with grading: Freestyle Capital, Bloomberg Beta, Reach Capital, and the House Fund, with participation from our existing investor K9 Ventures. Dave Samuel from Freestyle will be joining Manu Kumar from K9 Ventures on our board.
If you’re a professor or a TA who is or will be teaching a class in the near future, we invite you to check out Gradescope. And if building AI for education sounds like fun to you, we’re hiring!
— Arjun, Sergey, Ibrahim, and Pieter
