Monday, November 26, 2018

AI to correct answer sheets?

The other day, I was chatting with a professor about his teaching workload in the ongoing semester. He admitted that he had a fantastic run with an elective course that had more than a hundred students. With the examinations already underway, he was dreading the deluge of the answer scripts that would follow! He was also engaged in a research project involving blockchain applications in solving a community problem. Our discussion veered toward the blockchain technology and its applications in the real world today.

Most applications of blockchain are centered around finance, (virtual) money, governance, public health etc. Education has found an interesting application too in ensuring that one can access the certificates and marksheets securely and with full confidence about their authenticity. You may want to listen to this McKinsey podcast discussion about blockchain concepts.

The conversation with the professor started me thinking about more applications of new technologies in education. One of the most draining tasks in a professor's tenure is the evaluation of answer scripts. The task is quite repetitive, not full of academic discoveries barring those rare gems and takes away time from what could have been more rewarding activities like research or consulting. This looked like the perfect recipe for a tech intervention. Artificial Intelligence has come a long way already in making sense of a paragraph beyond the elementary spelling and grammar checks. The idea isn't novel and has been in the making for at least a couple of years. This article from Aug, 2017 mentions about the experiment in US and Australia and ends with a word of caution from some teachers.

AI systems, unlike their predecessors in rule-based computing algorithms, learn from their experiences and over time, gain a much deeper understanding than a human brain. Currently, they may lack the understanding of creativity but they are very good at doing the repetitive tasks in a quicker manner. We are moving beyond the mundane matters of OMR-based answer sheets to now make sense of a written word in the right context. Some simple algorithms existed already in trying to "rate" the essay in Statement of Purpose that colleges receive every year from thousands of aspirants. But these were simple keyword searches and couldn't have been used as definitive assessments. We are now able to design AI-based systems that do medical diagnosis from MRI scans! On another front, the AI systems are also able to compete (and defeat the best humans) on games like Chess and Go.

It may sound simple as compared to these winning examples but evaluating answer scripts from different student minds is not an easy task. The "model" or "sample" answer script, against which the answers are evaluated, can't be an exhaustive sheet, specially, in the fields of humanities or social sciences. It involves an open mind to identify originality, creativity and underlying understanding of theory rather than applications of mathematical formulae.Since most exams are still conducted on paper today, it adds an added complexity of recognizing human handwriting through Optical/ Intelligent Character Recognition (OCR/ICR) technology.

If we can crack the code to use AI in evaluating answer scripts, and assuming that we move to type-written answer scripts, the opportunity of setting up a central BPO operation for multiple colleges/ universities is huge. More importantly, this will lead to a more personalized feedback mechanism to aid the teaching assistants and tutors to help the student better. Data-driven analytics will feed back into the faculty's teaching to improve quality too.

A 2016 TEDx Talk by Prof. Ashok Goel from Georgia Institute of Technology demonstrated use of an AI-powered chatbot to handle student queries effectively and virtually freed up the Teaching Assistants' time. When we hear contrarian voices about people losing jobs to AI and robots, we need to remember that human brains need to focus on the "new" rather than the "old and regular" work.

Instead of worrying about the job losses, which are quite inevitable, we need to find out what our brains are capable of beyond intelligence and let technology take over the routine stuff that it can learn over time.