This was a HeLF webinar facilitated by Christopher Trace at the Surrey Institute of Education, to provide us with an introduction to KEATH.ai, a new generative AI powered feedback and marking service which Surrey have been piloting.
It looked very interesting. The service was described as a small language model, meaning that it is trained on very specific data which you – the academic end user – feeds into it. You provide some sample marked assignments, the rubric they were marked against, and the model can then grade new assignments with a high level of concurrence to human markers, as shown in the chart above of Surrey’s analysis of the pilot. Feedback and grading of a 3-5,000 word essay-style assignment takes less than a minute, and even with that being moderated by the academic for quality, which was highly recommended, it is easy to see how the system could save a great deal of time.
In our breakout rooms, questions arose around what the institution would do with this ‘extra time’, whether they would even be willing to pay the new upfront cost of such a service when the cost of marking and feedback work is already embedded into the contracts of academic and teaching staff, and how students would react to their work being AI graded? Someone in the chat shared this post by the University of Sydney discussing some of these questions.