Substantive changes to the basic school e-exam in Estonian as a native language will be introduced in 2027. 

Harno is studying whether it is possible to introduce AI into the process alongside the other changes. The agency believes it could be used for some parts of the exam-marking process.

At a meeting with Tallinn University researchers last week, the university's working group presented Harno with the project's results so far and outlined the opportunities for future development.

"Both technical solutions and possible use scenarios were discussed, including which types of tasks might in the future be graded using AI-based automated grading. The meeting provided a good overview of what steps would be needed before the possible wider adoption of automated grading," Toime said.

She said Harno does not yet have a firm view on how to proceed, but in principle, it could be used for assessing different tasks.

"For example, it could be used for longer open-ended responses that are graded according to a clear and previously agreed assessment model. In such cases, artificial intelligence can analyze the content of responses and compare it with the assessment criteria," Toime said.

A final decision cannot be made until the relevant solutions have been thoroughly tested both technically and substantively, she added. It is important to ensure that the systems are reliable, transparent, and support fair grading.

Automated grading would help deliver grades faster

Harno believes the main advantages of automated grading are saving time, publishing the results faster, reducing human error and making the grading process more consistent.

"At the same time, automated grading can never become the only method of assessment. The role of the human grader will always remain central. Artificial intelligence can rather function as a tool that helps support the grader, detect possible deviations, and make the process more efficient," Toime stressed.

At the moment, there is no timeframe for the introduction. Additional discussions are also required with the education sector, especially teachers.

"As a next step, we will begin a more thorough analysis of the solutions created during the project and, if possible, also conduct pilot trials. Based on their results, it will be possible to assess whether and to what extent AI-based grading could be used for both exams and other forms of assessment, such as level tests, and what its role could be in supporting the grading process," Toime said.

A study by Tallinn University researchers found that language models gave similar grade assessments as humans.

Tallinn University lecturer in native language teaching and applied linguistics, Merilin Aruvee, the study's lead author, said language models are particularly well-suited to grading an exam task that assesses the ability to use a source text.

--

Follow ERR News on Facebook and Twitter and never miss an update!