ML systems cannot respond to new stimuli that they have not been trained on. i.e. if a ML isn't specifically programmed to learn specific trends in data given to it, it will be unable to learn optimally from the data.

Adaptability

ML classifiers/systems are opaque in how they work i.e. hyperparameters are rarely understandable in the scope of the objective, and thus why these models work is difficult to explain/understand.

Explainability

Refers to understanding the relationships between different variables, how they affect one another, they effects of changing one variable on the others, asking 'What if?' questions.
Cannot be understood by statistics alone.

Understanding Causal Relationships

Obstacles that machine learning faces : 
1. Adaptability
2. Explainability
3. Understanding Causal Relationships

University of Michigan - Ann Arbor

Pearl, Judea. “The Seven Tools of Causal Inference, with Reflections on Machine Learning.” Communications of the ACM, vol. 62, no. 3, 2019, pp. 54–60., doi:10.1145/3241036.

https://dl.acm.org/doi/pdf/10.1145/3241036

The Seven Tools of Causal Inference, with Reﬂections on Machine Learning

Three obstacles for Machine Learning

A Structural Causal Model has three parts : 
1. Graphical Model
2. Structural Equations
3. Counterfactuals and interventional logic

Three parts of an SCM

The seven tasks the author thinks SCMs accomplish are : 
1. Encoding Causal Assumptions : Allow Transparency and Testability
2. Allows control for confounding and Do-calculus
3. Helps generate algorithms for counterfactuals
4. Mediation analysis and assessment of effects
5. Adaptability, sample selection, validity
6. Allows for missing data
7. Allows causal discovery

Learn Before

Related

Learn After