02.11.2023 | News AI colleagues need clear limits

Would you hire someone for your company without having a real idea of what the person is supposed to do, whether they are trustworthy and what they are even capable of? No, me neither. And yet you hear from more and more companies that they are desperate to buy generative AI. The technology is the future, and one must not miss the boat, management occasionally says. Of course, these voices are not entirely wrong: AI holds enormous potential indeed - for all sectors, authorities, and companies. However, the will to hire this new employee alone does not bring the "return on investment" that every company is striving for: AI is not a foregone conclusion.

Studies on the topic of artificial intelligence, such as a recent Bitkom study, show that German companies have a rough idea of where generative AI plays to its strengths. According to the study, an overwhelming majority of the companies surveyed believe AI has great or very great potential for text analysis and understanding, and would also use generative AI to create reports, translations, or other texts. What they simply lack in many places, however, is a concrete plan. Companies need to be clear about what problem AI is supposed to solve and what processes it is supposed to improve - and they need to be clear about this before they plan to implement it. Otherwise, AI threatens to become an expensive and pointless amusement that does not generate added value but rather creates confusion.

Beyond the specific intended use case, there is a need for further clarification before generative AI can find its way into the enterprise. First, the term generative AI is not synonymous with ChatGPT. Yes, this "conversational bot" is on everyone's lips due to the hype surrounding its underlying Large Language Model GPT (Generative Pretrained Transformer). But there are also alternatives, some of them even as open-source variants. On the other hand, stakeholders absolutely have to deal with the issue of data protection. Especially companies that store and process sensitive data must select their tools according to GDPR conformity. This is the only way they can avoid being confronted with fines at some point. It is best to choose AI helpers that they can use on-premise, i.e., within their own IT infrastructure. But simply ensuring and retaining sovereignty over their own data and not trusting AI cloud service providers is also an important component of the overall concept.

Since generative AI excels in text understanding and text creation, this is also where the most useful use cases for companies lie. Corresponding tools can selectively provide information from a wide variety of source files – for example for question/answer applications from a company's treasure trove of data. They can even break down long texts to the essentials and summarize them. Their inherent text understanding also makes them the perfect basis for further applications, for example as a digital assistant in first-level customer support.

To be able to carry out these tasks, companies may also have to train the artificial intelligence accordingly and grant it access to all company data. This is a sensitive issue, especially regarding access rights: Those who are not careful can make confidential information accessible to unauthorized persons. To avoid falling into this trap, companies should therefore use AI in conjunction with an enterprise search solution that supports rights- and role-based access control. The big advantage is that enterprise search is already the rights-checking information integration layer on top of which AI can be built - as an ideal, modern complement. The AI then only provides answers from the corporate knowledge that the questioner is allowed to see. In this way, clear limits can be set for the new, digital colleague without undermining its competences and abilities. It is therefore clear that anyone who embeds artificial intelligence in a trustworthy framework can benefit from it with a clear conscience.