We had three presentations at "REAL undervisning". The first presentation was given by Professor in language technology Stephan Oepen from Department of Informatics. In his presentation “Kunstig intelligens, maskinlæring, ChatGPT: Hvordan fungerer språkmaskineriet?” he explained the functioning and training of language models. He also discussed the concept of deep learning and neural networks, emphasizing their structure and the objective of learning input-output relationships. Additionally, he highlighted the use of supercomputers such as LUMI for AI research and the significance of scaling up data and computational resources. Stephan also noted that language models, for the time being, are more of an engineering feat than a display of intelligence, emphasizing the need to be aware of their limitations in terms of actual knowledge.
In the next presentation titled "KI - hva er lov? Hva er juks? Hva er lurt?," senior lecturer Vidar Skogvoll from KURT explored the legal aspects, the potential for cheating, and the intelligent application of AI, in that specific order. The essence of legality revolves around the type of data being transmitted and its destination. UiO categorizes data into four levels of sensitivity, represented by the colors green, yellow, red, and black, each requiring increasing confidentiality. For researchers, the use of non-pre-approved external services is permissible only when dealing with 'green' data. In contrast, for educational purposes, students are restricted to pre-approved services. The UiO IT homepage lists these approved services and specifies the permissible data for each.
Cheating is primarily manifested in three ways: 1) Inaccurate citation, 2) Plagiarism, 3) Unauthorized use of tools during assessments. AI in itself is not inherently cheating, but its misuse can lead to issues of citation or plagiarism. To minimize such risks, it is advisable to initially compose your own text, using AI subsequently for enhancements. As of January 2024, UiO's official stance allows professors to determine the appropriateness of AI use in assessments. If AI usage is permitted, clear guidelines should be provided to students on acceptable practices, including a format for disclosing AI assistance, similar to guidelines in leading academic journals.
Regarding intelligent use of AI, ongoing dialogue with students is crucial. The key is to resist the temptation of using AI for direct answers. Instead, it should be employed to guide towards solutions. Further insights to this were given in the last presentation titled "Kunstig intelligens – på lag med ekte intelligens?" by senior lecturer Andreas Haraldsrud from KURT.
In his presentation Andreas focused on the integration of AI, particularly tools like ChatGPT, in learning natural sciences and mathematics. He discusses how ChatGPT can be used as an advanced solution manual, emphasizing the importance of the learning process over merely finding answers. He highlighted how AI can make students either passive or active learners and the need for guided and disciplined use of AI in education. He delved into learning principles, such as retrieval practice, spacing, interleaving, and feedback, and explored how AI can contribute to these aspects. Additionally, he touched upon the role of AI in error detection and refinement, reducing cognitive load, and creating diverse narratives for explaining concepts. Finally, he encouraged thoughtful and beneficial use of AI in education, balancing technology with the human aspect of teaching.
Andreas has also developed resources for learning promoting use of AI, found on KURTs webpage.