Explaining and understanding in the age of AI

Nothing in life is to be feared, everything is to be understood. Now is the time to understand more in order to fear less.

Explaining and understanding without AI

In the context of research, the terms "explain" and "understand" are often used to describe different aspects of the analysis and interpretation of phenomena. Wilhelm Dilthey, a German philosopher and historian, made a major contribution to the distinction between these two concepts.

According to him, explaining (erklären) is more closely associated with the natural sciences, while understanding (verstehen) is specific to the humanities (Dilthey, 1883).

• Explaining involves providing reasons or causes that justify or account for a phenomenon. It refers to the identification of the underlying mechanisms or processes that lead to the occurrence of that phenomenon.
  
  In research, explaining a phenomenon often means determining the causal relationships between different variables. For example, a study may seek to explain why a certain public policy has led to an increase in vaccination rates by identifying the specific factors, such as increased awareness or easier access to vaccines, that have contributed to this outcome.
  
  The work of Carl Hempel and Paul Oppenheim also took this notion further, developing the deductive-nomological model of scientific explanation (Hempel & Oppenheim, 1948).
  
  
• Understanding, on the other hand, refers to the ability to grasp the meaning, significance or implications of a phenomenon. It implies a more holistic and contextual appreciation of the phenomenon.
  
  In research, understanding goes beyond the identification of causes to include an appreciation of the perceptions, meanings and subjective experiences associated with the phenomenon under study. For example, understanding why a community is reluctant to vaccinate requires not only identifying the contributing factors, but also capturing the fears, cultural beliefs and values that influence the behaviors of individuals within that community.
  
  The work of Erving Goffman and Max Weber is particularly relevant here, as it highlights the importance of social meanings and symbolic interactions (Goffman, 1959; Weber, 1922).

The distinction between explaining and understanding can also be seen in terms of objectivity versus subjectivity. Explaining tends to be more objective and quantitative, often based on measurable data and statistical analysis. Understanding, on the other hand, is often more subjective and qualitative, requiring methods such as interviews, observations and case studies to capture nuances and deeper meanings. This distinction is crucial in the social sciences, where the emphasis is on interpreting human behavior and social structures (Geertz, 1973).

In terms of the analytical versus holistic approach, explanation focuses on specific, isolated elements of the phenomenon, such as independent and dependent variables. Understanding takes a holistic view, taking into account the overall context and the complex interactions between various factors. The work of Clifford Geertz, in particular on ethnography, illustrates this comprehensive approach, where the aim is to account for the "density" of cultural meanings (Geertz, 1973).

Finally, the aims and objectives of these two approaches differ. The aim of explanation is often to predict and control. For example, by explaining the factors that influence academic success, we can develop interventions to improve student performance. The aim of understanding is to make sense of and inform, often by enriching the perspective of stakeholders or guiding the formulation of new research questions.

This distinction is particularly highlighted in the work of Thomas Kuhn, who explored how scientific paradigms influence the way researchers perceive and interpret data (Kuhn, 1962).

Explaining and understanding with AI

The impact of artificial intelligence (AI) in this context is profoundly changing explanatory and comprehensive approaches. AI, with its ability to process huge quantities of data and identify complex patterns, strengthens the explanatory dimension by enabling more sophisticated and predictive causal analyses. Machine learning algorithms, for example, can reveal hidden relationships between variables that traditional methods might overlook (Russell & Norvig, 2020).

However, understanding, as defined by Dilthey and others, remains a challenge for AI. The machine, while effective in analyzing data, has difficulty in capturing cultural and contextual meanings as deeply as human qualitative approaches. Work on the interpretability of AI models seeks to bridge this gap by making algorithm decisions more transparent and understandable to humans (Lipton, 2016).

AI can automate the collection and analysis of large quantities of qualitative data, facilitating a faster and perhaps more detailed understanding of social phenomena. For example, AI-based text analysis software can identify themes and sentiments in large corpora of textual data, which would be extremely laborious for human researchers (Manning et al., 2008).

However, the contextual and nuanced nature of human understanding poses a significant challenge for AI. AI tools may lack the sensitivity needed to apprehend the subtleties of human interaction and cultural meanings. This limitation highlights the importance of combining quantitative AI approaches with traditional qualitative methods to obtain a complete and nuanced picture of the phenomena under study.

Finally, the integration of AI into research processes raises ethical and methodological questions. How can we ensure that AI algorithms do not reproduce the human biases present in training data? How can we ensure that the conclusions drawn by AI are interpreted correctly and contextually by human researchers? These questions highlight the challenges and responsibilities associated with the use of AI in research (Vayena et al., 2018).

In summary, the impact of artificial intelligence, while powerful in explanation, highlights the complexity of understanding in the analysis of human and social phenomena. The contributions of Dilthey, Kuhn, Goffman and other researchers in the humanities and social sciences show that both approaches are essential for a complete and nuanced understanding of the phenomena under study.

Questions for future developments

1. How can advances in AI improve our ability to understand cultural and contextual meanings as deeply as human qualitative methods?
   
   
2. What new methodologies are emerging that effectively combine the strengths of AI and traditional qualitative approaches in research?
   
   
3. How can we mitigate the biases and limitations inherent in AI algorithms while taking advantage of their advanced explanatory capabilities?
   
   
4. What ethical implications need to be considered when using AI to understand and explain complex social phenomena?

Image by Alexandra_Koch from Pixabay

Sources

Dilthey, W. (1883). Introduction to the human sciences. Leipzig: Duncker & Humblot.

Geertz, C. (1973). L'interprétation des cultures. Paris: Éditions Gallimard.

Goffman, E. (1959). The staging of everyday life. Paris: Éditions de Minuit.

Hempel, C. G., & Oppenheim, P. (1948). Studies in the logic of explanation. [No known direct French translation].

Kuhn, T. S. (1962). The structure of scientific revolutions. Paris: Flammarion.

Lipton, Z. C. (2016). The myth of model interpretability. [Available in prepublication on arXiv].

Manning, C. D., Raghavan, P., & Schütze, H. (2008). Introduction to information retrieval. [No known French translation].

Russell, S., & Norvig, P. (2020). Artificial intelligence: A modern approach (4th ed.). [No known French translation].

Vayena, E., Blasimme, A., & Cohen, I. G. (2018). Machine learning in medicine: Addressing ethical challenges. PLOS Medicine.

Weber, M. (1922). Essays on the theory of science. Paris : Plon.