Models are not true, but some are useful. Models are either used as mental constructs by individuals or they are used by groups as social constructs. Conceptual modeling has always focused on socially constructed, explicit representations that are useful for gaining shared understanding of an affair or even for designing and implementing technical systems, most of all information systems.

For mental representations of individuals, the working hypothesis of neuroscience is that mental representations are embossed into neural structures and ultimately into electric signals. Socially constructed conceptual models require explication by representations governed by some shared conceptual-modelling grammar (Wand & Weber 2001), i.e., they become social reality by information in a medium.

Different phases of conceptual modeling have been conducted in the past decades. Initially, conceptual models were fully controlled and “closed” representations, e.g., frames. This was followed by a phase of conceptual models that are unrestricted (“open”) for capturing the richness of human knowledge in general, most of all ontologies. Statistical models and machine learning models often lack direct connections to individual knowledge and socially constructed knowledge but emerge from data alone. The underlying assumption is that data is taken directly from reality and is therefore objective. Recent discussions on biases and distortions of data raise the question of the social construction of data as well. Current research on explainability and interpretability tries to build bridges between both fields. Hybrid models are an attempt in this direction by trying to merge socially constructed conceptual models with machine learning models. The success of machine learning models has initiated chip design research to develop dedicated chips that can directly support AI processing. This might have repercussions on preferred designs of information systems and machine learning models. Even more advanced are quantum computing and quantum information theory when transforming data representations into quantum representations that are accessible by quantum computing algorithms.

In this talk, common aspects between all these fields are discussed and some thoughts on research questions will be presented. A focus will be laid on the interplay between conceptual modeling and machine learning models but also some connections to advanced chip designs and quantum computing are given.

Digital transformation is a key trend in which data plays a crucial role. As a result of the ongoing digital transformation, in line with the law of requisit variety, I see increased complexity and variety that helps in dealing with the challenges of today. Complexity is not bad in and of itself, as long as we have enough understanding of the organization in order to manage it effectively, and even use it to our advantage.

Three real-world cases show that organizations struggle with data/data man- agement. Key questions in this are are: (1) Do we know our data well enough in order to use it? (2) How do we balance “grip on data” with “value creation”? I will argue that effective use of models can help answer these questions. This is not a new position: several scholars have made this claim in the past as well. Very few organizations appear to have a mature modeling capabilitiy, leading to high cost, low agility, and hindering digital transformation. It is intering to ponder how this state of affairs came to be: why is modeling such an under valued skill in light of the fact that both scientific theories as well as heuristic frameworks emphasize it so strongly? Going a step further: why is “theory” such a dirty word in most organizations, up to the point where considering the use of heuritics-based frameworks is alreaddy cause for raised eyebrows and serious discussions?

There is no silver bullet that will improve the status quo: if there was, we would have found it by now. I will propose a strategy that combines a fast cycle (learning by doing) and a slow cycle (build the capability, re-learn the value of theory/models) to move forward. This generic approach can only be succesful when tailored to the specific situation in an organization. Last but not least, I will argue that training and experimentation are key enablers: don’t wait to educate people when they are in the field but start already during their university education.

No complex system can be built without effective modeling. And software systems are complex. Hence, modeling is necessary for building software systems, and a range of models are used for different tasks in the software development process – each playing an important role for that task. The nature and use of models in building a system, however, depends on the nature of the system also – if the system being developed is “hardware” which is costly to change, more rigorous and detailed modeling becomes necessary. If the system being developed is “software” which is easy to change and the cost of change is not high, modeling can be more helpful at higher levels of abstraction helping develop the software solution. Also, while models guide the development of the solution, it may be acceptable to have the final solution diverge from the model. In such situations, reverse engineering models from the solution can also be an useful. Detailed modeling is more appropriate for application domains where errors and changes in the solution are much more expensive. In such cases, domain specific modeling can be useful which may lead to executable models. If models are executable, then the modeling language becomes another programming language and will have to compete for adoption with other programming languages and other approaches for rapid solution development that have emerged. Providing support for integrating generated code with other written code can help in using formal models for parts of the solution.

Keywords: Modeling, Software Engineering, Software Development.

Security attacks form a system of specific flow of computation and data by one or multiple threats. Attacks follow a set of steps in a sequence. Threats work together as threat groups. Holistic 360-degree defenses against APTs often interconnect multiple threat intelligence computation and defense mechanisms. Each of these processes have a graph structure inherent to their execution. Graphs can be used to model spatio-temporal dimensions and flows of different facets of security as well as privacy. In our previous work, we have explored use of graphs and hyper graphs for threat, attack as well as defense modeling. Moreover, we have also explored using modeling threat intelligence as a system of graphs and using graph analytics and graph deep learning in order to predict, infer, extract features and information for assuring holistic security. Such work has been developed in the context of autonomous cars, AI, cloud and edge computing. In this talk, we will also explore how to use NLP and NLU on how to automatically construct such graph models for specific systems under protection/attack.