Tacit knowledge and human digital twins

Antero Karvonen & Pertti Saariluoma

The difference between traditional and future industrial processes will be in the extent, role and nature of human thought, perception and action. Currently we are somewhere in between. Traditional technology is predicated on the necessity of human guidance and input for steering the processes. Since at least the advent of computers and servomechanisms, this role has been changing. For example, many harbors are largely automated today. Equally, many planes can take off and land with minimal human intervention.

Replacing and augmenting human thought, action, and perception is the basic task in designing intelligent systems. The logic is straightforward. Currently people take care of certain tasks or responsibilities that require intelligence. Intelligent technology is precisely the expanding boundary of tasks which can be autonomously handled by machines. The process is one of adding more and more sophisticated abilities to the artificial system. The human role typically recedes gradually, first to a supervisory role, then perhaps simply setting goals for the system. Thus, the parallel consideration is how work processes are augmented through this gradual replacement. Where the human is still needed is either where the practical limitations of (artificial) intelligence currently stand or where ethical considerations are needed. However, whether we are replacing or augmenting or both, it is necessary to understand and be able to investigate human information processing. Many design mistakes can be avoided, and new possibilities explored if designers could grasp what it is that they are replacing. Here the cognitive perspective has a role to play. The premise is simple. If currently humans and machines can jointly realize some goals and processes, we can be certain that there exists an intelligent information process taking care of it. However, without concentrated research and the right perspective, large parts and the exact nature of this information process will remain unknown. Importantly, it will remain tacit.

Making this tacit information explicit is necessary for all the core aspects of intelligent technology. It can be used for developing AI; it can be used for re-designing work processes; it can be used to design man-machine symbiotics; and it can be used for organizational and individual learning. At Jyväskylä University, we have been thinking about two new approaches to these issues. The broad framework is called cognitive mimetics. One specific embodiment of the idea is human digital twins (HDTs), which we will address here. The importance of tacit information (for both) will become apparent next.

Tacit knowledge makes experts experts, and it is key factor in successful performance

In nineteen forties expertise researchers were surprised when they understood that experts cannot tell why they are experts. Good chess players, for example, explained they are skilled, because they have played much chess with strong players. Thinking that acquiring high-level expertise takes ten years at least, makes it understandable that people do not know consciously what they have learned. Similarly, one can ask who of us knows the rules and regularities they follow in using their mother tongue. The give example shows and human conscious mind is just a tip of an iceberg.

Expertise and expert performance is based on the tacit domain. On the most inclusive reading, this domain manifests in what we can perceive as skill. When we look at skilled performance, what we find is an extraordinary transparency for the expert on why they are good. They simply don’t know why they are good. Consider the many facets demanded by expertise. An expert can typically “see” what is relevant and what is not in a situation. They can quickly identify possible causes for events, foresee future trajectories, and shortlist actions while knowing their consequences. They have the motor skills and familiarity with the domain (or system) they are using. They understand the fundamentals of the domain deeply. But what they know is largely tacit, meaning they are automatic, efficient, and fast – of course not always correct. Simply put, the effectiveness of the tacit domain comes from not having to think about it.

If tacit knowledge underlies expert performance and if by definition it is “hidden”, then we can understand its’ importance and also the difficulty of studying it. However, in cognitive psychology methods of discovery and modeling have been found.

Tacit knowledge can be studied

While tacit knowledge is by definition something hidden, it can be explicated. The fundamentals of modeling intelligent information processes were developed in cognitive psychology on the basis of Turing’s ideas (which are also the basis of computer science). The key idea was to think of the mind as an information processing system. With the simultaneous advent of computers, it became possible to not only sketch out or describe intelligent information processes, but to also model them on a computer. These models were intended to be run on a computer, where the validity of the model was partly derived from its ability to perform the task that was being investigated. In this way, AI and cognitive modeling have common ground. The methodological problem is how to get a picture of what happens in the mind when doing a task or solving a problem. In cognitive studies think-aloud protocol studies, interviews, and focus groups have been used. In protocol studies, participants are asked to verbalize (without reflecting) what is going on in their minds as they perform tasks. These fragments can be analyzed for traces of thoughts, and evaluated against actions and environmental states. They can then be deepened by interviews. This, in turn, can be modeled in a computer giving a concrete and developable picture of the information processes and contents involved. This is where it becomes apparent how this tradition could be connected to the ideas developed in digital twinning.

Tacit knowledge can be modeled as part of HDTs

Digital twins (DT) are typically digital models or replicas of technical processes and systems, like paper machines. They replicate the components, functions, and physical aspects of complex industrial processes in a digital form. They can be used as tool for designing, managing, maintenance and planning among other things. Human Digital Twins (HDT) in turn are models of human action and thought when interacting with technologies. They are designed based on knowledge gained by cognitive research and implemented as computational models. They are a way of introducing the human component of the joint cognitive system to (traditional) digital twins. We see the roles of DTs and HDTs as complimentary.

By using the methods and ideas from cognitive psychology and cognitive modeling, we can create computational models of humans that include tacit knowledge. HDTs are a way of explicating and embodying the tacit component that is largely responsible for keeping current industrial processes running. One added value is that research results are by design embodied and iterated in a computational system. The process becomes iterative. The models will always contain “terminal points”, which simply mean the extend and current reach of the model – what has been studied and embodied in technology. Methodologically, it directs research to focus on where tacit information should be sought. For example, action patterns such as identify and correct can be modeled as an HDT. Initially, it is open what the action correct actually contains (ie. what are the possible and relevant actions in relation to what has been identified). However, this focuses the next round of searching for hidden information.

These models can be used in different ways

The models can be used for various purposes. First, they can be used to developed intelligent machines. Ideally in fact, the terminal points of the model are its limitations in intelligent behavior. This also identifies the role of the human operator. Terminal points, or uncertain areas of the model remain in the domain of humans or somewhere in joint action. In this way, we can study the structure of work processes and design them for the industry of the future. Furthermore, studying is also a way of learning, because the tacit information can not only be used in computational modeling, but also human learning. The limits and the terminal points furthermore illustrate the differences between artificial and natural intelligence. These make HDTs and the perspectives we have integrated into the idea a useful tool for developing intelligent technologies and society. In the final analysis, in the human world the only intelligence there is, is found in the minds of humans and other animals.

Literature:

Saariluoma, Pertti; Cañas, Jose; Karvonen, Antero (2020). Human Digital Twins and Cognitive Mimetic. In Ahram, Tareq; Taiar, Redha; Langlois, Karine; Choplin, Arnaud (Eds.) Human Interaction, Emerging Technologies and Future Applications III: Proceedings of the 3rd International Conference on Human Interaction and Emerging Technologies: Future Applications (IHIET 2020) (Advances in Intelligent Systems and Computing, 1253. Cham: Springer. DOI: 10.1007/978-3-030-55307-4_15)