The Impact Produced By Ml Techniques In Practice During The Past Decade

The reviewed article makes a thorough analysis of the impact produced by ML techniques in practice during the past decade. One of the central directions of the work is the classification of the present jobs or skills as “SML” or “not (yet) SML”. At the same time, the article highlights the idea that humans’ role will increase in some sectors where ML techniques becomes prevalent. In general, the automation process and human’s implication cannot be separated for unstructured tasks. By comparison, ML adoption tends to keep at minimum the human’s intervention in case of structured tasks with well-defined goals. In addition to the elasticity of the workforce requirements introduced by ML, the business perspective continues to be impacted in what regards the ways how the services are provided and methods used to create the goods.

Although it is generally not trivial to estimate the range of tasks where humans perform better than machines or vice versa, the overall impression is that learning methods implemented through ML can be improved especially in the context of larger training data sets which easier become available. However, the machines have a clear advantage against humans when the rate of change in time of these sets is relative low. ML can spread over the current limitations if creative ways to define the task are identified. As the article shows, presently the performance metrics are defined in detail when the objectives can be learned well by machines. On the other hand, the training experience is enhanced once the technology advances. However, the definition of the task to be achieved is often reduced to an objective function which cannot simulate the complexity of the decisions, comprehensive logical chains or the afferent background information. This is the main reason why “automating automation” remains for now an area where the Artificial Intelligence impact does not have the desired level.

Explanation of the reasoning residing behind the decisional process represents a key field where people continue to outperform the machines. For many this can be a paradox because the final decisions are often better or more accurate when are taken by machines instead of humans. Despite this apparent advantage, machines continue to require humans’ implication and assistance in what regards implementation of more reliable learning methods and simulation of the background information. The needed complementarity between humans and machines is also required by the incapacity of the persons of formulating a set of formal rules when a problem needs to be codified. In essence, humans know more than they can express formally. Besides the technical factors, it is also necessary to consider the weights of the economic factors when redesigning the business processes or studying the implications of ML for the workforce.

In general, humans will be easier replaced by machines in case of those jobs where ML adoption does the tasks better than human; for instance, drivers. Since human error is responsible for almost all vehicle accidents, the giant companies like Tesla, Google, Apple, and so on are investing billions of dollars in self-driving vehicles. Thereby, I believe that in ten years machine will start driving our vehicles. Finally, despite the common perception that the pronounced automation will tend to eliminate humans from the greatest majority of the business processes, ML learning techniques will continue to attract the need for better programmers since the fact that better formulation for the task goals will require better programming and modeling skills.