Loss of knowledge due to baby boomer retirement: Daniel Fallmann explains how AI can save companies

Germany is facing a massive demographic shift as the baby boomer generation – particularly those born in the 1950s and 1960s – retires. This generation forms a significant part of the workforce in key industries such as engineering and IT, whose valuable expertise cannot be easily replaced.

Daniel Fallmann explains how AI and modern knowledge management systems can overcome the shortage of skilled workers caused by the retirement of baby boomers.

Photo: Mindbreeze

In our interview, AI and knowledge management expert Daniel Fallmann explains how companies can cushion the impending loss of know-how through innovative AI and knowledge management solutions.

Mr. Fallmann, how do you view the current demographic development in Germany, especially with regard to the baby boomer generation, and what challenges does this pose for companies in the engineering and IT sectors?

Demographic trends show that many employees of the baby boomer generation will retire in the next few years. This generation currently represents a large part of the workforce in the engineering and IT sectors and has significant specialist knowledge and experience. Without these specialists, there will inevitably be a loss of know-how that is often difficult to compensate for. It is therefore important for companies to take measures to efficiently make the knowledge and, in particular, best practices of the baby boomers available to the next generation.

How do you assess the role of AI in knowledge management, especially when it comes to capturing and making accessible the valuable know-how of experienced employees?

Artificial intelligence (AI) definitely plays a central role when it comes to preserving the know-how of experienced employees. Imagine a young colleague asks a question to a digital assistant and receives a summary of how, for example, her predecessor handled a matter – even if she is no longer with the company. This requires AI-based knowledge management systems that extract relevant information from the various company data sources, combined with a Large Language Model (LLM) to summarize the content.

Document the knowledge of key employees

Can you explain how Mindbreeze technologies help companies to efficiently and accurately document and share the knowledge of their departing employees? What specific tools and approaches do you offer?

With pleasure. A large part of the knowledge is stored, be it in documentation, manuals or emails. It is now important to make this knowledge findable and accessible. Our insight engine Mindbreeze InSpire makes it possible to search all relevant data sources of an organization quickly and efficiently. We use different artificial intelligence methods to extract the information from the sources, link it and prepare it for employees. This can be done in a 360-degree view or as an answer to a question in text form, as is known from ChatGPT. Of course, we also combine our insight engine with large language models (LLMs).

There is a lot of talk about the risk of the “hallucinations” of AI language models such as ChatGPT. What exactly are these “hallucinations” and why are they particularly problematic in the context of knowledge management?

“Hallucinations” are AI-generated content that appears plausible but is incorrect, outdated or misleading. This happens especially when the AI calculates based on incomplete or incorrect training data, incorrect conclusions or connections, or information is simply misinterpreted. Large language models in particular are designed to provide a generated answer to a query. The focus here is on the generated text and not on the fact that the content must be correct. Insight engines, however, focus on content correctness, validated and trustworthy information as a basis, as well as on enforcing and checking access rights. In a corporate context, it is essential to provide verified answers and at least always to identify the sources underlying the answer so that humans can verify them.

Combining a Large Language Model with an Insight Engine

How can you ensure that only a company's specific knowledge is used as the basis for the answers of an AI language model, and what measures are taken to ensure data security?

To avoid the hallucinations mentioned, companies are increasingly turning to “retrieval augmented generation”, or RAG for short. This is the combination of a large language model with an insight engine. When a question is asked, the insight engine searches the company data sources for relevant facts and information and passes this on to the LLM, which creates an answer in natural language and identifies the relevant sources for traceability. This allows users to trace the answers at any time using the facts provided by the insight engine.

To what extent can AI-powered expert networks help share human knowledge more efficiently, and what advantages do such networks offer over traditional knowledge management methods?

Traditionally, information was stored and maintained in a knowledge database – or not. Mindbreeze as an additive system takes a different approach here. By connecting the data sources via connectors, the information remains in its storage location. We analyze and link the information and make it available when queried according to access rights. This means that connections between key topics and experts can be quickly determined, always with the most up-to-date information. We call this the 360-degree view of an expert. Here, in the query, you will find all relevant information about the person and the documents they have created, such as reports or studies. This saves time when searching and makes knowledge available more efficiently in the company.

Trends and challenges in knowledge management

Given the rapid technological developments, what future trends and challenges do you see in the area of knowledge management and knowledge preservation through AI in the next five to ten years?

We are seeing a development towards generated information that is verifiable and relevant. Analyst houses such as Gartner or Forrester refer to this as “Explainable AI” or comprehensible AI-generated information. If the results are transparent, this is also referred to as “Trusted Information”. In contrast, there are black box results whose origin cannot be traced. We see another challenge in the area of sustainability. Currently, operating and training LLMs costs an enormous amount of energy – especially electricity. These energy costs must be reduced minimally so that computing power becomes cheaper and more sustainable by generating electricity from renewable energy sources.

How important is collaboration between humans and AI when it comes to maintaining and promoting know-how within a company, and how can this collaboration be optimized?

We see AI as personal assistants that support and relieve the burden of daily tasks. Today, many employees are faced with the challenge of having to complete more and more work in a very short space of time. AI systems relieve the burden here by carrying out repetitive processes, such as automatically distributing incoming mail or being available as a sparring partner for answers. It is important to sensitize employees to only use systems that comply with compliance guidelines in order to avoid certain hallucinations and, on the other hand, to ensure that confidential information also remains within the company.

Implement efficient knowledge management early on

What advice would you give to companies that want to prepare for the impending knowledge loss due to the retirement of baby boomers, and what first steps should they take to improve their knowledge management strategies?

We recommend that companies implement efficient knowledge management to make existing information available to new employees. We see these systems as personal AI-based assistants that can provide ad hoc answers. Information that is available in the company is often forgotten during a handover, and this is precisely where Mindbreeze's strength lies – creating answers from the stored data. When introducing systems, we recommend additive systems that can be connected to existing data sources via connectors. This way, the data remains in its original storage location (cloud, on-prem). In addition, it is important to define the use cases and goals before introducing them and to start in one department. For example, many of our customers start with a 360-degree view of information for a department, implement a chat for documents, or automate processes that were previously done manually, such as inbox classification.