Assessing answer sheets in a college or university examination consumes faculty members’ time. The time and energy required to assess and grade a college or university student can be redirected to research and other faculty work. This saves faculty time for other important tasks rather than the mundane checking of answer sheets. The faculty can use this time to write their own books, consult with research students, collaborate at research conferences, discuss research funding, and promote their research institutions.
How can AI assess your answer sheets?
Do we need AI to assess answer sheets?
Given the specialization you have in college, you can have many people checking your college grades. However, this is not the case in universities; courses are enrolled by a limited number of students. Students are mature and know why they are doing a specialization at a university. Grading can be automated with AI to save professors’ time. Some professors assign grades to their research students. While other Professors check themselves.
In schools, children need a personal touch; therefore, AI cannot yet be used there, as this is a developmental period. We should begin from the top down. First, grade university students; then grade college students; and, when all things mature, grade school students using AI, with a supervised approach by the class teachers if in primary school or by a subject teacher if in secondary school. All AI grading for primary school students can’t be functional until assessments are fully transparent, showing marks for each sentence written by a primary school student. We need special care for primary school children, so let us proceed from the top down: start with university assessments. It may take several years for this to be a success, and only when the system is foolproof can we think of including it in primary schools.
We conclude that we should begin using AI to grade university examinations, then college examinations, and subsequently school examinations. However, simple primary school evaluations may appear. Proper grading needs to be taught to AI. There are several things we need to take into account for this.
For first theoretical subjects, such as management, history, and social science, the answer sheets can be scanned, converted into a document, and evaluated with the help of AI. AI used here can be semantic similarity with the correct answer, so that grammar and words don’t matter. The ideal answer is provided to AI, with keywords in bold. The keywords must be present in the answer in the answer book. The other points must be matched using the LLM’s semantic similarity mechanism, which is trained on the correct answer. Each correct statement must carry marks assignment and intermediate steps must be graded in this way. Full marks are to be given only when all keywords are enumerated, and all logical explanations are provided. A robot can be used to scan the sheets!
If the subject is mathematical or semi-mathematical, such as physics or economics, the text must be converted to an intermediate form, such as mathematical notation or LaTeX. The symbols are understood. There are cases when there is more than one derivation of a problem, for example, a calculus question or a physics derivation. In that case, the answer provided should be more than one. When the answer does not match any of the given expert Professor’s answers, in that case, LLM can try to understand the logical analysis of the same. Marks must be assigned for each correct step toward the answer, as specified by the professor. LLM can infer its own understanding and pose a question to the professor in a unique case when it faces a dilemma.
Be aware that there will be a transition period for all this. During the transition period, faculty would have to review the answers themselves and compare the AI’s marking with their own. Once the Professor and the AI agree on a marking scheme, the AI assessment tool has reached maturity and can be used in other frameworks for test schemes. This can be done in reinforcement learning mode. The Professor’s input is paramount to AI learning in the assessment of students’ examination copies. As AI learns by doing the tasks of assessing in close contact with respected Professors.
Once this model works, we can apply it to college examinations as well. Gold standards of answers are provided by the faculty members. Everything is online, so there is no chance of errors. As in University assessment, college assessment must be cross-verified by faculty teaching in colleges and high schools. These are intended to provide reinforcement learning to the AI. AI learns its mistakes, as we can’t expect AI to do all we want in one go. Once marked correctly in 2–3 examinations, the AI can be integrated into the system. After that, a random answerbook checking system would come into force, where random answerbooks evaluated by AI are checked and corrected, marked, and evaluated by faculty. Then this random sampling is evaluated. AI is given feedback, if any. Again, some skewed answer books are evaluated.
AI in any of the above systems is to measure the mode and median of evaluation. It must bring to the faculty or the Professor’s attention any doubt, say in a mathematical physics paper, the student presented an altogether different derivation of concepts, while the Professor provided another derivation. Such a thing was checked for analogy by LLM and marked full, but this must be brought to the attention of the Professor as a positive outlier. The other side is a negative outlier in which the student produced a completely incorrect derivation; however, his definitions were correct, so granting the professor the right to mark this answer constitutes a shift from AI marking to human marking. This makes marking the answer book a collaborationbetween AI and faculty members.
The assessment of answerbooks is never easy when the time frame of checking the bulk of answer books looms over faculty with a d-day. AI can assist in evaluating answer books with faculty in semi-supervised and consequently unsupervised settings. The AI can request faculty input for outliers and uncertain conditions. The AI should mark each part of the answer and be accountable to both faculty andstudents. This can give students an open context for seeing answer books, where there is no second thought. Transparency in the system is the output of this AI system; both students and faculty can see the breakdown of marks and the evaluation system with clarity.
Artificial Intelligence, Computer Science & Mathematics 