22/01/25

AI in Technical Communication – Trends and Latest Developments
Claudia Sistig, technical writer at Compart GmbH in Böblingen, has taken an in-depth look at the potential applications of AI in a professional context. In her presentation at the Best Practice Workshop for Technical Communication in Medical Technology, which took place on October 23, 2024, in Göttingen, she explained the most significant developments of recent months. The following article summarizes the main points of her presentation.
One notable trend is the increasing range of AI applications, which can be divided into two categories: on the one hand, applications that provide easy access to AI functions and, on the other, applications that allow new AI applications to be developed without the need for in-depth programming knowledge. AI functions are now integrated into numerous popular applications. For example, Zoom now offers an AI-supported companion for recordings, and Adobe has also implemented AI tools in Acrobat and Photoshop. Almost every email program or calendar now contains AI-supported search functions or similar features. In addition, more and more content management systems and editors offer AI functions to generate content or analyze existing information and make suggestions for improvement. Interaction with AI systems has also improved: communication is more fluid, and responses are less stiff, therefore appearing more human. Another important advance is AI agents, a more advanced form of AI chatbots. These agents can access data independently and use various tools to handle challenging tasks and complex requirements.
Chatbots
Chatbots respond directly to user requests. Users ask questions and the chatbot provides answers as quickly and fluently as possible to enable human-like interaction. Typical queries could be: “How can I set up a process with this software?” The chatbot then describes the necessary steps, as long as the corresponding documentation has been stored as context information. The quality of the answer is evaluated by the user. If a user asks a question or makes a correction, the chatbot tries to improve its answer. In the live demonstration, a chatbot was used in the FlowiseAI environment to answer questions about the user documentation of a blood glucose meter. The chatbot, which was based on an OpenAI GPT model, showed that it could accurately reproduce the information from the documentation and was able to present complex information in simpler language. However, chatbots quickly reach their limits when the requirements become more comprehensive and difficult. Support usually ends with the question “Can you generate training documents in HTML format for me on this basis?”
AI agents
In contrast to chatbots, AI agents work autonomously and can make decisions independently within the scope of defined tasks. Users define the task, and the agent accesses the tools and data provided to find the best solution. It can check whether the solution found meets the requirements and adjusts it if necessary. The focus here is less on the speed of the answers and more on their accuracy and quality. The agent-based workflow has a multi-level structure and includes a supervisor who receives user requests and distributes the work orders to several workers. Each worker can use different tools to perform specific subtasks. The workers can access a variety of tools, including code interpreters, API access and databases. This makes the workflow extremely adaptable and powerful.
- Supervisor: Receives the user requests and coordinates the work orders of the workers. The supervisor can also access different chat models, which enables flexible adaptation to the respective requirements.
- Worker: Each worker has specific tasks that are defined by a separate system prompt. In the demonstration example, the first worker was tasked with reading information from user documentation and preparing it for children and young people. The text is simplified, and technical terms are reformulated or omitted.
- Multi-stage process: After processing by the first worker, this worker delivered the processed information to a second worker. Its task was to create training units. It divided the information into interesting modules and added additional elements such as quiz questions. The third worker then formatted the content into a child-friendly HTML page.
The combination of the supervisor and several specialized workers allows complex tasks to be processed efficiently in order to deliver high-quality results.
Risk levels of the EU AI Regulation
For companies, the introduction of artificial intelligence (AI) means complying with the applicable rules and the EU AI Regulation. A draft version is currently still available, but many companies are already using it as a guide. The aim is to define strategies and guidelines as to which AI applications are permitted in the corporate context and under what conditions. The EU AI Regulation defines different risk levels for AI applications:
- Unacceptable risk: These applications are completely prohibited in the EU. Examples include social scoring and biometric surveillance in public spaces.
- High risk: These include applications in the areas of human resources (e.g. analysis of applicant behavior by AI) and lending (e.g. AI-supported risk assessments). Strict logging and verification rules apply to these applications.
- Limited risk: This includes chatbots, for example. These can generate incorrect information. In the case of limited risk, applications must be clearly labeled.
- Minimal risk: This includes AI applications such as recommendation services (e.g. from Amazon and Netflix) or spam filters, which are expected to have only a minor impact. No special rules currently apply to these applications.
Companies can use these risk levels as a guide and only use AI applications with limited or low risk, for example. They must define which data may be used for training and which providers are acceptable. The use of AI agents offers new opportunities but also requires clear rules and strategies.
The integration of AI applications into the day-to-day work of technical editors can simplify and speed up processes. Many companies are already using parts of the EU AI Regulation as a basis for formulating their own clear rules and strategies for dealing with AI.