Artificial intelligence is increasingly involved in decision-making across different fields, from security to business and operational activities. At the same time, AI systems have rapidly become more complex, and the logic behind the recommendations they produce is not easily understandable. This creates a new kind of challenge: although AI can identify patterns and make accurate predictions, it is becoming increasingly difficult for users to understand what these conclusions are based on.
This raises the question of whether decisions made by AI can truly be trusted. Explainable AI partly responds to this challenge, but the recent Social Explainable AI book argues that current explainable AI methods do not produce explanations that are understandable for non-technical users. For AI explanations to be understandable, they should be presented in language and formats that are suitable for the user, in the same way that large language models adapt their responses according to users’ questions. Explainable AI should be capable of doing the same, while ensuring that the responses genuinely correspond to the AI’s actual reasoning, which is not necessarily guaranteed in the case of responses produced by large language models.
One of the authors of the book, Kary Främling, serves as Research Director at LOUHE as well as Professor at Aalto University and Umeå University. The writing of the book began following a seminar organised in Japan by Kary together with Professors Katharina Rohlfing and Brian Lim. Experts from a wide range of disciplines were invited to the seminar, including not only specialists in explainable AI, but also philosophers, social scientists, psychologists and robotics experts. The idea was to utilise existing knowledge and experience regarding what constitutes a good explanation, specifically from the end user’s perspective.
Current explainable AI methods and researchers do not appear to care very much about the end user, which is also reflected in the limited capabilities of today’s leading explainable AI methods to produce explanations that are understandable for end users. This is contrary to EU principles regarding transparency and openness in AI, which are impossible to achieve using current methods.
User-Centred Explainability Makes AI More Understandable
Explainable AI has largely been developed from a technical perspective: how algorithms function, how models make decisions, and how their outputs can be interpreted. As Social Explainable AI points out, many approaches still pay little attention to the user’s perspective and the practical usefulness of explanations. As a result, explanations may be technically correct, yet difficult to understand or disconnected from the user’s actual needs.
It is precisely this challenge that the book approaches from a new perspective. Explainability does not emerge solely from the algorithm itself, but from how the explanation meets the user: their role, objectives and situation. If this perspective is missing, the explanation easily remains detached from information. When the user is taken into account, explainability becomes a practically useful tool that supports decision-making and helps bring AI-generated information into practical use.
Explainability Is Not a Feature, but Interaction
Social Explainable AI highlights that explainability is not merely a technical feature of AI that can be added to a model afterwards. Instead, explanation is viewed as a social practice in which meaning emerges through interaction between humans and AI. The book describes this as a situation in which explanations are formed through interaction between the user and the system while also evolving alongside the context. In practice, this means that AI does not simply produce a ready-made explanation, but can adapt explanations while taking the user’s needs and situation into account. Explanations are therefore not static answers, but processes that evolve during use.
This perspective also changes the role of the user: they are not merely recipients of explanations, but active participants involved in constructing understanding. Explainability also emphasises the user’s participation in forming explanations, as well as the importance of interaction for the relevance of explanations. When explainability is understood as interaction, its objective is not merely to reveal how an AI model functions, but to ensure that the explanation is genuinely relevant and useful in the specific situation.
Three Perspectives That Change the Way We Think
In Social Explainable AI, explainability is approached through three key perspectives. First, the book emphasises the importance of context: different situations and social roles influence what kinds of explanations are needed. Second, explanations are not one-off events, but are described as gradual processes in which explanations take shape through interaction with the user. Third, explanation is not limited to text alone, but can utilise different forms of presentation, including visual, linguistic or other forms of communication. Together, these perspectives emphasise the contextual, interactive and multimodal nature of explainability.
From Explainability to Competitive Advantage
The value of AI does not arise solely from its ability to analyse data, but from how well the information it produces can be understood and put into practice. When explainability is understood as interaction rather than merely a technical feature, it supports decision-making, improves operational efficiency and increases transparency. At the same time, explainability is becoming an increasingly important factor in the usability and practical value of AI solutions.
This way of thinking is also reflected in practical solutions such as those developed at LOUHE, where AI is utilised to support physical security and risk management. Observation alone is not enough; the system must also be able to explain why an anomaly has been identified and what it is based on. When observations are understandable, organisations can move more quickly towards justified decisions and concrete actions. LOUHE utilises the explainable AI method Contextual Importance and Utility (CIU), originally developed by Kary Främling in his doctoral dissertation. The book also demonstrates how and why CIU specifically enables “social explainability”, and why other current explainable AI methods are unable to achieve the same.
It is precisely in systems such as LOUHE that explainable AI brings explainability closer to practical application: it does not merely support analysis, but enables action.
