Classification solutions using machine learning are not prediction as these terms are used in the field. These applications are generally based on taking an unknown pattern and classifying it based on finding the closest match to a training set of labeled patterns. The patterns can be images, sounds, words, phrases, etc., or more abstract ones that we don't usually encounter in daily life. Most of these do not have to find a solution that has to do with the future or causality in general.

To answer Matthieu, there are many predictors that are not machine learning. Linear and nonlinear predictive programs, for example, are merely mathematical curve fitting algorithms. These find least squares fits and project or interpolate beyond the available data. These algorithms were (and still are) in use for decades before machine learning, as we recognize it today, was being invented.

Dear Matthieu,

Really thanks a lot for this detailed reply.

I read carefully your answer. The example of vision is interesting for me to understand the separation of actions from decision making. However, the sentene 'Known stimuli are recognized from past experiences' means selection of pattern and this is an action of selection !

Maybe in my mind I still having a confusion of the 'action' or 'acting'. For me it doen't necessarily related to movement or mechanic movement. Choosing a pattern (recognition) in a conceptual way is similar to moving an object in concrete world. Both of them are actions !

Thanks again.

Best regards

Dear Rohit Yadav ,

Thanks a lot for your contribution and clarification.

Best regards

Just a methodological comment. Machine learning, prediction, pattern recognition, decision making, control, etc. concern an interaction of the agent (estimator, predictor, ... decision maker, controller ...) and its environment (system, world part, brain part, ...), which changes the state of this closed loop. Anything what is opted by agent (estimates, prediction, ..., inputs) are actions influencing this closed-loop state. I found this view quite useful. For instance, it forces us to ask: Why we estimate? Why we predict? etc., i.e. it forces us to relate the action with the aim of its selection. Best regards MK

Matthieu Molinier ,

Thanks for your reply again.

I understand. I have to expand my literature to try understanding the point. Thanks for for your reactions.

Best regards

Miroslav Karny,

Yes, I had a similar idea about actions and prediction and the other related terms. But I have to expand my literature to more depict the epistimology of those paradigms.

Thanks a lot.

Best regards

Dear Fateh Boutekkouk,

I think that your opinion is incorrect. To counteract my opinion, please, give a trivial example of, say, decision-making, in which you need not consider possible consequences of your decisions, i.e.\ in which you do not predict what will happen...

Best regards MK

Classification solutions using machine learning are not prediction as these terms are used in the field. These applications are generally based on taking an unknown pattern and classifying it based on finding the closest match to a training set of labeled patterns. The patterns can be images, sounds, words, phrases, etc., or more abstract ones that we don't usually encounter in daily life. Most of these do not have to find a solution that has to do with the future or causality in general.

To answer Matthieu, there are many predictors that are not machine learning. Linear and nonlinear predictive programs, for example, are merely mathematical curve fitting algorithms. These find least squares fits and project or interpolate beyond the available data. These algorithms were (and still are) in use for decades before machine learning, as we recognize it today, was being invented.

Hello all,

A very important question really,

If we look to the core functions of the majority of ANN and AI models, it is very clear that it reflect mathematical probability and statistics functions and methods. In reality if we apply those function our output will be a predictive result. It is very helpful for Big Data and complex systems that are difficult to model.

However, intelligence is not limited to prediction, even if todays results are very satisfying for classification and prediction. The brain is not a predictive engine, and thinking, reasoning and acting are based on knowledge and facts. Furthermore, emotions, ethics and social interactions are not mathematically defined.

There is alot of fields to be approached from different perspectives before defining intelligence and artificial intelligence. But today we are in the age of prediction in AI.

Thanks,

Dear s. Wilber

Really thanks. It's clear and helpful.

Best regards