TY - GEN

T1 - Taking Causal Modeling to a Next Level

T2 - Self-Modeling Networks Adding Adaptivity to Causality

AU - Treur, Jan

N1 - https://slaai.lk/icai/2020/

PY - 2020/12

Y1 - 2020/12

N2 - This paper covers the contents of a Keynote Speech with the same title. Causal modeling is an intuitive, declarative way of modeling that due to the universal character of causality in principle applies to practically all disciplines. In spite of this seemingly very wide scope of applicability, there are also serious limitations and challenges that stand in the way of applicability, in particular when dynamics and adaptivity play a role. This paper addresses these challenges by exploiting the notion self-modeling network developed from a Network Science perspective. It is shown how temporal-causal networks allow modeling dynamics based on given causal relations and how self-modeling networks can be used to also model dynamic changes of the causal relations themselves. In this way, causal models are obtained that show dynamics of the nodes based on the causal relations as well as adaptivity of these causal relations. Adaptivity is obtained by adding self-models to a given causal base network; these self-models represent the base network's causal structure by additional network nodes for the causal relations that are adaptive. These self-models are themselves temporal-causal networks as well, are still specified in a declarative manner by mathematical relations and functions, and create a next level for the network by which the adaptation is addressed. Moreover, this construction can easily be iterated so that multiple orders of adaptation can be covered in the form of multilevel causal models, for example, addressing controlled adaptation or metaplasticity. So, this indeed takes causal modeling to a next level in more than one way so that now dynamics and adaptivity are also covered well, which substantially widens the scope of applicability of causal modeling.

AB - This paper covers the contents of a Keynote Speech with the same title. Causal modeling is an intuitive, declarative way of modeling that due to the universal character of causality in principle applies to practically all disciplines. In spite of this seemingly very wide scope of applicability, there are also serious limitations and challenges that stand in the way of applicability, in particular when dynamics and adaptivity play a role. This paper addresses these challenges by exploiting the notion self-modeling network developed from a Network Science perspective. It is shown how temporal-causal networks allow modeling dynamics based on given causal relations and how self-modeling networks can be used to also model dynamic changes of the causal relations themselves. In this way, causal models are obtained that show dynamics of the nodes based on the causal relations as well as adaptivity of these causal relations. Adaptivity is obtained by adding self-models to a given causal base network; these self-models represent the base network's causal structure by additional network nodes for the causal relations that are adaptive. These self-models are themselves temporal-causal networks as well, are still specified in a declarative manner by mathematical relations and functions, and create a next level for the network by which the adaptation is addressed. Moreover, this construction can easily be iterated so that multiple orders of adaptation can be covered in the form of multilevel causal models, for example, addressing controlled adaptation or metaplasticity. So, this indeed takes causal modeling to a next level in more than one way so that now dynamics and adaptivity are also covered well, which substantially widens the scope of applicability of causal modeling.

M3 - Conference contribution

T3 - Communications in Computer and Information Science

BT - Proceedings of the Fourth SLAAI International Conference on AI, SLAAI-ICAI '20

PB - Springer Nature Switzerland AG

ER -