The paper presents the basics of causal inference, which is important today for achieving more advanced artificial intelligence…
ABSTRACT
Artificial intelligence (AI) is standardly divided into symbolic (often called logical AI) and subsymbolic (referred to as statistical AI).
Today, neural networks are in the center of attention, as the most important part of subsymbolic AI. However, in certain applications (e.g., AlphaGeometry), excellent results are achieved by combining neural networks with symbolic AI (known as neuro-symbolic AI).
One subset of neuro-symbolic AI is a synergistic combination of causal inference and neural networks (note: if causal inference and neural networks are used relatively independently, it is not considered “true” neuro-symbolic AI).
Today, causal inference is considered important to achieve more advanced AI, e.g. this opinion is shared by one of the “fathers” of deep neural networks, Yoshua Bengio (one of the winners of the 2018 Turing Award).
The presentation/paper will cover some of its basics. It includes both causal learning, which involves building a causal model, and causal reasoning, where the causal model is used together with data for causal reasoning.