July 26, 2020
Causal inference in AI refers to the process of drawing a conclusion based on the causal connection amongst the conditions of the occurrence of an effect. The main goal of causal inference is to analyze the response of the effect variable when the cause is changed.
Causality is based on how we view and react to the environment around us, to our decision making, and to the advancement of science. Causal inference in AI has advanced rapidly in the last 20 years. This has resulted in the development of a plethora of new methods, both for causal structure learning and for making causal predictions.
CAUSAL INFERENCE IN AI: WHERE IT COMES FROM
Causal inference is an element of inductive reasoning, which involves supplying strong evidence for the truth of the conclusion. Many aspects of problem solving involve inductive reasoning. Inductive reasoning is a means of reasoning from a part to a whole, from the past to the future, from particulars to generals, or from the observed to the unobserved. For example, from the fact that one can hear the sound of piano music, it can be inferred that someone is (or was) playing a piano. Causal inference, capable of exhibiting such features of inductive reasoning, is the central aim of several AI systems as it helps in the fields of medicine, epidemiology and public health.
CAUSAL INFERENCE IN AI: HOW HAS IT EVOLVED?
Historically, even when attempting causal inference, the role of statistics was to quantify the extent to which ‘chance’ could explain the results, with concerns over systematic biases because of the non-ideal nature of the data relegated to the qualitative discussion of the results. The field of causal inference has changed this state of affairs, setting causal questions within a coherent framework that helps in facilitating explicit statements of all the assumptions underlying the analysis and allows extensive exploration of potential biases.
CAUSAL INFERENCE IN AI: HOW DOES IT OPERATE?
CAUSAL INFERENCE DIAGRAMS
Causal diagrams consist of a ubiquitous feature of methods for estimating causal effects from non-ideal data. It focuses on analyzing causal structure of the variables from which results such as conditional exchangeability can be deduced. Such assumptions are typically represented in a causal diagram or graph, with variables identified by nodes and the relationships between them by edges.
Another common feature of causal inference methods is that, as researchers move further from the ideal experimental setting, more aspects of the variables need to be modeled. This would have not been required had the data risen from a perfect experiment. Structural equation modeling is a fully parametric approach. In this approach, the relationship between each node in the graph and its parents is specified parametrically. This approach helps in offering a holistic view of ignorable missing data and measurement error.
Causal inference methods are particularly relevant for studying the causal effect of a time-varying exposure on an outcome. This is because standard methods fail to give causally-interpretable estimators when existing time-varying confounders of the exposure are themselves affected by previous levels of the exposure. In the past few years, artificial intelligence researchers have paid a great deal of attention to causal inference methods. AI systems, based on predictive and big data analytics often entirely neglect this aspect and, in many cases, don’t give you specific information on which one variable affects the outcome. Causal inference is useful in studying causal relationships among variables and helps in correctly predicting new relationships.