Statistics is the science of description and prediction. The more systems got complex more hardness of correct description and prediction get serious. The latest gift of AI to statistics is machine learning. Although AI had its roots back to even 17th century, ML is not as old as its ancestors. Nature of statistics comes from backed predictions on factual matters. However, on the ML side things are not straightforward. Although there is a strongly correlation between prediction and machine learning, this correlation does not suggest a causation. Causation comes from their shared ancestors: artificial intelligence.
History is full of people that ask the question: why we need it? Not very surprisingly they disappeared from the stage. In fact, same situation is observed in ML, too. Some of the “future experts” didn’t even bother to understand the capabilities of ML whereas some saw the potential. Strong point here is that the correct question is how can we use it? Thus, asking and looking for potential use of the AI, and transposing the humanoid factors to AI resulted in tons of different use of ML from data science to public health. Therefore, the correct approach is always have to be capacity solutions and potential enhancements, not looking for exact meaning of newly found concepts.
The shared value of both AI and ML is prediction as described above. Hence, the coincided concept is a more general one. Prediction by itself means nothing, when put into a concept of data and/or paradigm, it becomes the only solution of a complex problem. That is, descriptive analysis has the power of stating the correct sides, where predictive side shows you the possible outcomes, their roads, and alternative solutions. Thus, prediction analysis with correct tools and discourse reveals what is unseen. The part where description and prediction meet each other named as decision. That is, after statement of problem and listing of possible outcomes and action would be taken, namely decision. Subsequently, there is no need to stress on the matter that they are, decision and prediction are highly correlated.
The exact sense of ML slowly finds its way as we state the terms such as description, prediction, and decision. Even a basic examination of the ML theory shows that basic I/O systems are redefined by ML approach that after individual’s firs few decisions AI learns if there exists a pattern and acts accordingly. Decision theory suggest a probability and a utility. Thus, probability here is defined by prediction by the system and utility is defined by past actions etc. Hence, whether it is a humanoid or an AI element, a decision-taker views universal induction as a sole solution of the matter where abovementioned variables are parted into different pieces of tools and approaches such as Bayes theory for prediction and Turing approach for utility examination.
Therefore, make it a simple I/O system or a complex model for any problem. Both probability theory and decision-taking patterns are observed in inductive initiations. This looks like a human intelligence testing model where the observer’s ability to make correct decision are through induction and tools used are probability and past knowledge on the matter. Thus, the 21st paradigm of AI suggests that AI ability to make decisions becomes consistent, increases its probability of choosing correct action, since it copies the human induction and action-taking pattern. Sooner or later what becomes machine learning from Artificial Intelligence will become deep learning since dominance of predictive and descriptive analysis are converges to more complex concepts of intelligence and decision taking.