MACHINE LEARNING IN CYBER-PHYSICAL SYSTEMS AND MANUFACTURING SINGULARITY – IT DOES NOT MEAN TOTAL AUTOMATION, HUMAN IS STILL IN THE CENTRE: Part I – MANUFACTURING SINGULARITY AND AN INTELLIGENT MACHINE ARCHITECTURE

Abstract

In many popular, as well scientific, discourses it is suggested that the "massive" use of Artificial Intelligence, including Machine Learning, and reaching the point of "singularity" through so-called Artificial General Intelligence (AGI), and Artificial Super-Intelligence (ASI), will completely exclude humans from decision making, resulting in total dominance of machines over human race. Speaking in terms of manufacturing systems, it would mean that there will be achieved intelligent and total automation (once the humans will be excluded). The hypothesis presented in this paper is that there is a limit of AI/ML autonomy capacity, and more concretely, that the ML algorithms will be not able to became totally autonomous and, consequently, that the human role will be indispensable. In the context of the question, the authors of this paper introduce the notion of the manufacturing singularity and an intelligent machine architecture towards the manufacturing singularity, arguing that the intelligent machine will be always human dependent, and that, concerning the manufacturing, the human will remain in the centre of Cyber-Physical Systems (CPS) and in I4.0. The methodology to support this argument is inductive, similarly to the methodology applied in a number of texts found in literature, and based on computational requirements of inductive inference based machine learning. The argumentation is supported by several experiments that demonstrate the role of human within the process of machine learning. Based on the exposed considerations, a generic architecture of intelligent CPS, with embedded ML functional modules in multiple learning loops, in order to evaluate way of use of ML functionality in the context of CPPS/CPS. Similarly to other papers found in literature, due to the (informal) inductive methodology applied, considering that this methodology doesn't provide an absolute proof in favour of, or against, the hypothesis defined, the paper represents a kind of position paper. The paper is divided into two parts. In the first part a review of argumentation from literature, both in favor of and against the thesis on the human role in future, is presented. In this part a concept of the manufacturing singularity is introduced, as well as an intelligent machine

Application of Machine Learning in the Precise and Cost-Effective Self-Compensation of the Thermal Errors of CNC Machine Tools – A Review

“…It is worth to mention about automated machine learning, which can save time and money spent on searching effective machine learning methods. Searching for thermal error modelling method may be supported by double-loop learning [54], n-loop learning [55,56]. F. Hutter et al performed review of automated machine learning (AutoML) methods [57].…”

Machine Learning in Cyber-Physical Systems and Manufacturing Singularity – It Does Not Mean Total Automation, Human is Still in the Centre: Part II – In-CPS and a View from Community on Industry 4.0 Impact on Society

“…Recalling the fundamentals of the intelligent machine architecture, presented in the Chapter 4 of Part I [22], CPS could be considered as a constructor towards development of the intelligent machine, towards manufacturing singularity, following the architecture presented there, that includes n-loop learning.…”

“…In accordance with the notation in Chapter 4 -Part I [22], we could now denominate the CPS with embedded learning loops as I-CPS, where 'I' stands for 'Intelligent'. Consequently, adding the indicator for the levels of learning loops, we could have the following notations for different levels of 'intelligence' embedded in the CPS: I 0 -CPS, I 1 -CPS, I 2 -CPS, …, I n -CPS, where I 0 -CPS stands for the CPS without learning, I 1 -CPS stands for the CPS with single-loop learning, I 2 -CPS stands for the CPS with double-loop learning, and I n -CPS stands for the CPS with n-loop learning.…”

“…Synthesising the program by learning is synthesise of the program "by any means other than explicit programming" [38]. In the context of following the inductive inference based learning, specified in the Section 4.2 -Part I [22], it means that the learning is performed based on examples.…”

An Approach to Identify the Interactions Between the Control Factors in a Mahalanobis -Taguchi System

“…MTS is used as an analysis tool for complex nonlinear relationships with non-numeric data in the industrial field. The MTS is useful for a tool of analysis and prediction of complex nonlinear relationship in the case of fully automatic AI control, the main features are that it can be used as a black box regarding complex, nonlinear, and interaction-laden causal relationships, and it can be controlled with high accuracy, which is widely used in industry [15][16][17][18][19][20]. However, when MTS is used for anything other than fully automated AI control, the several interactions between the control factors in the MTS make highly accurate control difficult.…”