Why we need Causality?
The future of AI
The Reasons for "Why?", seems like its the most important question here!!
Knowing and asking "why" is really about the data is important, Causality can help us answer this.
Answering some really important questions:
What leads to the change in outcome?
Does a change in the system improve the outcome?
What are the overall effects of a particular treatment?
Awareness and knowledge about
Spurious correlations
Correlation is not Causation
Confounders in the data
Causal Inference can help generate models that are pretty accurate as well as inherently Explainable. We need causality because there are limitations to statistical models.
If we need more data or not? It helps us in knowing if we can calculate the causal relationships from the data, or if we need more variables within our data for better estimation.
Helps us understand the Data Generating Process(DGP), which then leads us to ask more relevant questions (Diagramatic representation of data is very important and helpful)
The Graphical approach of Causal Inference is more clear with respect to explicitly defining the objectives, assumptions, and conditions.
Causal inference can lead us to better results for the Decision-Making Process.
Helps us understand the importance and necessity of appropriate/relevant data. Even after having infinite data, we might miss important confounders, which hence can introduce bias in the model.
Importance of Domain knowledge, which come into use in formalizing the Causal Structure
Importance of Human Intervention, in this case, Assumptions. Thus it combines the strength of humans along with Machines, which leads to more productivity and harnesses the full potential of AI.
Implementation of causality can remove the algorithmic bias by highlighting the biased features within the data or the examples. Causal Inference meets the criterion for Bias and fairness
It can help build a Trustworthy & Responsible AI
The difference between humans and machines is that humans can distinguish between cause and effect. Adding causality to machines can help us achieve Human-Level Intelligence.
To build a strong AI we need to emulate how humans think about the cause and effect.
It is the natural way of making an inference, the closest one to the real-world model.
Last updated