Artificial intelligence (AI) has been the talk of the town since ChatGPT & Co. at the latest. Long ridiculed, the potential is now being recognized abruptly. This is also due to the fact that the massive training efforts in this technology, which amount to billions of euros, have led to a new level of quality being achieved. In this respect, we are talking about a leap innovation, i.e. an innovation that is more like a revolution than an evolutionary further development of existing technologies. ChatGPT-3 was trained with 175 billion words and word combinations, ChatGPT-4 within only 4 months with 100 times more words. The models become more and more intelligent with extreme speed, so that meanwhile even the protagonists of AI plead for taking a break in development.
But how do ChatGPT (from openAI), Bard (from Google) and Co. work? The generative language model GPT (stands for “generative pretrained transformer”) writes and improves texts, translates, generates photos on demand and will soon produce videos on demand. All of this generation is done using a neural network, i.e., stochastic principles. Based on the words in a sentence part that has already been generated, the most likely next word or sentence part is generated. The generated text is validated with learned data and text. Then it is checked if the content is harmless and if necessary – if not completely harmless – the answer is rephrased. As a result, the system generates grammatically good and correct sentences. The truth content still has to be checked by the reader, but due to the trained data set, the texts will already be very valid, especially in technical areas. ChatGPT-4 scores in the upper 10% range in comparative exams in the Abi area in the USA, while version 3 was still in the lower third.
The interface of ChatGPT&Co. and its integration into the search enable a novel, human-like communication experience – and this is exactly where the potential for intranet use lies. Instead of “just” searching and working out the answer themselves, employees and collaborators will have the answers generated for them in the future. Editors will then neither have to research nor write the texts for them. That will be done by AI. General knowledge will be used, which can also help in everyday business.
The challenge will be to supplement company-specific knowledge. The technologies have not yet reached that point. First, because companies probably have far too little data to train ChatGPT. Second, because it is usually internal, proprietary information, and there are not yet ways for company-specific training to separate the training data from generally available training data. Regarding this point, however, solutions will emerge in the future. For example, by integrating ChatGPT in such a way that the general questions are answered by the generally available AI service and then extended with company-specific additions. For example, an employee’s question about “How do I act if I get a data protection problem reported?” could be answered by ChatGPT with general instructions, and the intranet then supplements the information with the contacts and references to instructions in the company.
Software vendors like us at NEOCOSMO are working to integrate AI solutions more and more into their intranet platforms. What is already possible now is to use AI tools to support editing. NEOCOSMO has fully integrated DeepL as a translation platform in PIIPE. This allows editors to have the intranet multilingual. A text created in German is automatically translated for the other intranet languages and can then be edited. In the future, real-time translation will be available here, i.e. even if the intranet is not maintained in certain languages, users in their native language will be able to view the content automatically translated. Furthermore, ChatGPT will be available from the middle of the year to allow editors to generate and correct intranet content. In addition, there are image generators so that companies can at least partially dispense with the use of stock material.
What also works in existing solutions is intelligent personalization and individual recommendations: Here, the aim is to use AI to increase the relevance of content for the individual, to play out content that fits precisely, and at the same time not to provide too much information in order to avoid so-called “information overload.” For example, new employees have different interests and questions than employees who have been with the company longer. For the newcomers, it makes sense that content on “onboarding” is recommended to them at the start of their career. For employees who have been with the company for a long time, content that is about knowledge transfer is more interesting. At its core, intelligent personalization is about the intranet learning and adapting. It learns where users are located in the company and plays out suitable content. It learns which topics users are interested in and recommends content. That it helps to bring new employees into collaboration with experienced colleagues, for example, when networking within the company.
In summary: The potential of AI for use in the social intranet is huge. In 3 to 5 years, it can be expected that the intranet will consist more of AI-generated content (texts, images) than of man-made content. Companies will save costs on editorial and stock material in the process. Editorial staff will be freed up to deal with company-specific special issues. And, they will have to deal with training the in-house AI, i.e. AI training expertise will play a key role. AI will handle general knowledge answers. A ChatGPT-like search will move to the center of an intranet’s user interface.
by Dr. Volker Zimmermann, CEO of the social intranet company NEOCOSMO GmbH