Generatvie AI
Generative AI – what’s happening in the administration?
Expectations were high when access to the corporate version of Microsoft’s Copilot chatbot was opened in early December 2023. By that time, the media had been overflowing with both utopian and dystopian visions of what generative AI could mean for our way of working.
We are testing beta versions of Copilot and other AI chatbots
Most of those who have delved into the technology can confirm that the potential is great, but the technology is also immature and – due to the way it has been trained – filled with errors. When we use the Copilot chatbot or any of the other chatbots on the market, we automatically participate as testers for the chatbot provider. This means that we are continuously presented with updated versions of the various chatbots. This could be a new functionality, more tokens, i.e., more space to play with, or a better language versioning and answers.
Now we have also got the students on board, so today, on par with employees, they can choose to use Microsoft’s Copilot chatbot.
However, there is still a long way to go before we have an effective assistant for, for example, minute writing. This has also been realized in the secretariat at the Faculty of Humanities, where they have experimented with AI generation of minutes. So far, the technique can only handle meeting contributions of up to 15 minutes, say Andreas Eriksen and Louise Toftgaard Fibiger. They generally experience challenges with both hardware, software, and compliance. Challenges that need to be resolved before they can see that generative AI can be a real help in connection with minute writing.
Copilot Pro
Microsoft offers an add-on license to the university’s standard license, the so-called Copilot Pro. It integrates Copilot into Microsoft 365’s office products. It has just been released in a Danish version and is being tested. So far, the quality is not impressive, but improvements are constantly being made. For example, it tends to mix English, Danish, Swedish, and Norwegian. We will therefore expand the test scope with some employees who have English as their working language, to assess the quality of the English-language version. We are initially looking especially at the possibilities in Word, Excel, PowerPoint, Outlook, and Teams. Recommendations for Copilot Pro will be shared later when there is more experience with the license.
Ensuring security and quality via own chatbots
Developing our own chatbots is a way to ensure security and quality. Therefore, a pilot experiment is currently underway at SDU Digital to get a chatbot in Microsoft Azure, which SDU already has a license for and a data processing agreement with. Initially, it is being trained on financial instructions carefully selected by the Financial Service. As the solution runs in a Microsoft setup, additional consumption costs per ‘token’ entered into or retrieved from the chatbot are triggered (a token is a unit of measure for the amount of text). Through the pilot in the Financial Service area, it is expected to get a better feel for the gain from building our own chatbots and the actual costs of using them. We are also investigating the possibility of developing our own chatbot located internally at SDU in collaboration with IMADA and the eScience Center, so we can minimize consumption costs while avoiding trouble with data security.
Guidelines, instructions, and courses
To support the journey into the chatbot universe, a guidance universe has been established: SDUnet Digital Services. The guidance universe is continuously developed. Recently, for example, a podcast with DPO Simon Kamber and Professor Sten Schaumburg-Müller, Legal Institute, was added to the guidance universe: SDUnet Digital Services – Guidelines concerning security, copyright and more when using generative AI.
Since December, introductory courses on how to use AI chatbots have been offered with great success. More courses are on the way, both online and in person.
Good colleagues at SDU UP, the Library, Communication, and RIO are also starting to engage with the generative AI tools. Find info and links to more initiatives in our guidance universe or via the departments’ service pages on SDUnet and sdu.dk.
Are you curious to learn more or just want to share your own experiments with generative AI with the rest of us?
Contact Nina Thorsted Petersen, project manager for the Generative AI in Administration project: ninat@sdu.dk or by phone at 65507602 / 41866140.