Stanford AI Helps Grade Tests That Helped Students, But There's a Flaw

Alec G., Tech Times 20 July 2021, 04:07 am

Stanford invited 12,000 volunteer students in 148 countries to try out an online class dubbed Code in Place.

The Stanford-run course taught the foundations of computer programming for the student volunteers with the help of A.I.

A new artificial intelligence was introduced to students who took the same test built by Stanford researchers using automated system points for an all-new future of online courses.

First-Hand Exposes by Student

According to the New York Times, Philips Pham is one of the 12,000 student volunteers from Sweden who experienced his first encounter with the A.I. after four weeks. He was trying to write a program that could draw waves of minute blue diamonds along a black-and-white grid.

After several days, he was able to receive feedback about his work on the code.

The feedback stated that the A.I. applauded his work, but also showed his error, saying, "Seems like you have a small mistake ... Perhaps you are running into the wall after drawing the third wave."

The critique was what Pham needed to hear to improve, thanks to the help of the A.I. machine.

Stanford Researchers Deploys AI

Chelsea Finn, a Stanford professor and A.I. researcher said, "We've deployed this in the real world, and it works better than we expected."

Together with Dr. Finn and her fellow researchers, they were able to design the neural network's A.I. solely for Stanford's programming class.

However, the techniques that were used could potentially automate student feedback in regards to other situations far beyond just programming.

Oren Etzioni, the chief executive of Allen Institute for Artificial Intelligence, cautioned that the techniques are still a very long way from mimicking human instructors. Critique and assistance from human instructors and tutors are still the preferred methods over artificial intelligence.

Nevertheless, Dr. Etzioni still said that the Stanford project is a "step in an important direction" since having an automated feedback would be better than nothing at all.

Creation of Stanford's A.I. And Its Flaw

The data was collected over a decade's worth of midterm exams at Stanford with programming exercises, and has stored them all in a digital record.

The results were then stored in reams of code that were written by students and the university's instructors.

The A.I. can recognize animals such as a cat by looking at thousands of pictures of the animal, recognize voices from listening to old phone calls, and examining teaching assistants' methods of evaluating coding tests.

There is still a flaw, however, as that it can't be able to help many students with questions as to how they went wrong.

It can only analyze data from thousands of results from the past, but can't correctly assist students with questions that are not in the test.

Dr. Chris Pieche, a Stanford professor, says that the goal of the A.I. is not to replace instructors, but to give advice to students to pinpoint problems in regards to their work as to how frequent the errors are made during exams.

Dr. Piech says, "The future is symbiotic -- teachers and A.I. working together."

This article is owned by Tech Times

Written by Alec G.