Azure OpenAI is your safe environment with endless space for innovation
How do you let teachers and students experiment freely with AI, while making sure that all data remains safe on your own IT environment? An important question that almost all educational institutions are working on.
Both students and lecturers at the university we did this project for have been experimenting with ChatGPT from the start. On the one hand, the university was very enthusiastic about this, because they saw many opportunities to enrich education with AI and because they understand that working with AI is an important skill for their students. But the risks were also clear: if everyone uses a public AI tool, you have no control over your data at all. Personal data, research data, and sensitive business information will become part of public language models, and you don't know where, when, and how that information will reappear.
A nice poem about all your classmates...
One example was a student who copied his class name list into ChatGPT and asked the AI to write a nice poem for everyone. It sounds harmless, but unfortunately it is a data breach. OpenAI now offers subscriptions where the data is not included in the language model and therefore cannot appear in unexpected places and moments, but in addition, the company can still see who used ChatGPT, when and from which IP address. For many organizations, and also for the university, this is an important objection.
Our mission: AI, data and privacy
So the university asked us for an environment where anyone who wanted to experiment with AI could experiment, but where the user could count on data security and metadata privacy. A question that made us enthusiastic, because here at Blis Digital, together with Microsoft, we had been thinking about how to use AI for education for some time. Technologically, we also knew we could do it, because we had already built our own AI chatbot internally here. We did this on the basis of Azure OpenAI, Microsoft's AI that uses the same language model as ChatGPT and is therefore just as powerful.
User-friendly interface based on Teams
The strength of ChatGPT is that everyone can use it right away. Azure OpenAI is not a one-size-fits-all tool. It is a service that allows software builders to use AI in applications that they build themselves. So a software project is needed before you, as an organization, can use the power, the possibilities and the safe environment of Azure OpenAI. In this case, we chose to build a chatbot in Teams. Because within the university, Teams is where students, lecturers and other staff chat with each other and collaborate on files. We were able to easily add the AI to that. A second advantage is that security and access are then arranged immediately, without extra work.
Two technical routes
To create such a bot, you can follow two different technical routes. The easiest way is to deploy a ready-to-use bot with Copilot Studio. With little work, this results in a chatbot with a lot of possibilities. The disadvantage lies mainly in the costs. For this approach, the university should have given each employee and student a Copilot license. A cost of a few hundred euros per year, per user.
The other route is to create your own bot based on Azure OpenAI. This is the option that the university chose. Because of the costs, but also because this route offers the most functional options. The only fixed costs for this bot are those for hosting on Azure, between 100 and 200 euros per month in total. The disadvantage of this option is that we had to write code for it, which therefore also needs to be maintained.
Result: user-friendly, secure AI
The final result was a chatbot with the same chat experience as ChatGPT but looked like a Teams chat with colleagues or classmates. This way, no one has to learn new skills to experiment with AI. This also makes it possible to add the bot to a meeting or group conversation and get quick access to information. Instead of looking for a document on the intranet, ask the AI: “What is the procedure for posting a vacancy?”
The university has an influence on the tone of the answers and can control costs by choosing a cheaper language model or by imposing quotas. This can be done per chat conversation as well as per user. And, most importantly, all entered data remains within the university's IT environment.
The bot is currently still being used by a test group of several hundred users. In the event of a successful evaluation, we will probably roll it out to the entire university soon.
A basis for the future
Privacy, safety and user-friendliness are important arguments for having your own AI environment. But having your own AI is also an investment in future innovation. For example, he gives you the opportunity to train models on your own data and set the limits of what the model does and does not do. Public models, for example, are useless for medical or psychology students because they block topics such as sexuality. A proprietary AI, especially for these faculties, solves that problem. By the way, Azure is also perfectly capable of working with models other than GPT. In order to save costs or to use certain unique features, some of the users can therefore also be transferred to a competitive or customized model.
Also working with AI?
You probably also have the challenge that this university had: you want to work with AI because you know that there are opportunities there. But GDPR compliance and safety are major hurdles.
Albert-Jan Schot, our CTO, says:
“AI is a hype, but it's not entirely unjustified. AI is going to help a lot of people, in many different occupations. But how exactly, we don't know yet. That's why you have to now start experimenting and trying out, but in a safe environment and between clear boundaries. '
Azure OpenAI is where you can solve this problem. But with your “own AI”, you also lay the basis for further experiments, using your own business data and training your own language models. In addition, you can choose how you interact with the AI. A customized AI environment is therefore now a safe way to experiment and a basis for innovation in the future.
Want to know more?
Do you want to work with AI and are your people ready to experiment with it? We'll help you get started quickly with a first, secure, AI chat and lay the foundation for the next step.
.jpg)
