This chronology was published in the morning magazine 21. July 2023
I'm open to the fact that my method is not optimal, but it doesn't mean that language models are inappropriate as a tool for sensitising.
The researchers Pierre Lison and Samia Touileb come in Morning magazine last week, with a clear call to their colleagues not to be inspired by me to use language models to save time when they sensorate exams. Of course, you have to be critical of technology and how it is used, and you always have respect for privacy, academic honesty and signal effects. And I'm open to the fact that my method is not optimal, but that doesn't mean that language models are inappropriate as a tool for sensitising.
I think the optimal solution for sensorship in the future will be a combination of artificial intelligence (KI) and human intelligence. Being open about the processes of using KI helps to provide academic honesty and send the right signals to students of the future.
When you are sensoring an examination task in business economy, it may be reasonable to use a calculator or another appropriate mathematical tool. Similarly, it may be reasonable to use a relevant language model to censor a task that primarily contains text.
Ethan Mollick, professor at Wharton University and perhaps the world's most current KI expert in education at the moment, has Very good experiences using KI to summarise, extract relevant information and answer questions that require understanding of the content of a text. It's just... as I have used KI in the assessment processSo let's say that this is the same thing as that.
I did not use Chat GPT, as the title of the researchers' articles in the Morgenbladet signals, but a tool called Humata, used by many educational institutions. Humata claims that using the tool can be used research and learn a hundred times faster! The tool focuses privacy and document data is not used to train models. All data sent to or from the model will be deleted after thirty days. Vebjørn Sønders red, one of Norway's leading privacy experts, copyright and GDPR, has stated on Twitter in connection with this debate: "I don't know that you're uploading the student's name I don't see any particular privacy issues here. As regards copyright, the educational institution must be able to take copies necessary for censorship without consent.’;»
Then the point remains on signal effects. That students are pragmatic is nothing new. Access to KI tools will increase the requirements for students in the future. Students are already using KI on a large scale. It is therefore important that the actors in the education sector understand how KI can influence future education. This is done by encouraging teachers to explore the tools, while actively working with institutions and governments to develop guidelines and rules on how to use them.
By combining knowledge about how Google defines quality, information obtained from tasks using relevant KI tools, as well as traditional manual assessment, I believe that I have saved time and improved the quality of the assessment, without at the expense of privacy or academic honesty.