What can machine learning do? Workforce implications

What is the problem being addressed?

This paper addresses the problem of identifying the scope and scale of tasks performed by statistical machine learning (SML) and the role that this scope plays in determining the effects on the economy. It aims to thoroughly investigate the speculations about humans getting replaced by machines and decreasing employment due to ML by discussing the limitations of the field. The objective is also to predict the impacts of a heavy adoption of ML on the labor demand and emphasize the requirements to cope with the changes.

Why is that a problem worth solving?

Machine learning is rapidly transforming the way this world is operating. There has been a surge in the utilization of automation practices. However, although the potential of ML has been realized, their scope is not, nor have their impacts on the economy. It is apparent that ML is bringing profound changes throughout the economy, and the effects will only increase exponentially in the future. However, it is difficult to predict which parts of the workforce will be impacted the most. Thus, a study of the applicability and implications of ML demands attention from businesses, leaders, researchers, and lawmakers. Addressing these issues is crucial to restructure businesses and make policies that can accommodate the changes and sucessfully redirect the workforce.

Approach:

The authors recognize that not all the tasks are suitable Machine Learning (ML) applications and identify key criteria to determine if the tasks at hand can be solved effectively using ML. For example, any task that can be solved by learning a function that takes inputs and gives outputs may be a suitable task to solve using ML. It is necessary to have large data sets or experience for such a task for ML algorithms to learn from. The goals should be well-defined to be able to solve the task optimally and the labels in the training data sets should conform to these goals. The tasks that involve complex chains of reasoning and planning may not be suitable machine learning tasks, especially if their simulations are hard to obtain. As machine learning lacks the ability to explain its decisions, the tasks should not require the ability to do so. Moreover, ML algorithms are probabilistic that can never achieve decisions with probability 1, so a small error should be permissible for the tasks. The distributions in the test data should be similar to that of the training data and hardly change in short periods. Lastly, advanced physical manipulation and navigation should not be required. The tasks that adhere to these criteria are fit for solving using machine learning.

The authors also shed light on how much more complex the implications of automation are on the workforce compared to the chain of thoughts concerning humans getting replaced by machines. To address this they pin down 6 non-technological economic factors that affect the impacts including the substitution of humans by machines, the elasticity of price, income, and labor supply, complementary skills, redesign of business processes, and finally, the invention of new goods and services. Understanding these factors is of utmost importance as their relationships can often change the magnitude of impact on the economy. Thus, along with the criterion to distinguish the suitable machine learning tasks from those which are not, the paper also solves the task of pointing out the factors that affect the impacts of ML on the labor demand. 

Limitations:

• The paper only accounts for machine learning applications and does not focus on any other artificial intelligence applications e.g. robotics, sequential decision making, etc., even though they’re also going to impact the economy at large in the future.
• The authors discuss the impact of ML only through an economist perspective. Other issues such as safety and trust concerning the use of ML have not been discussed. In my opinion, steps needed for redesigning business processes to accommodate changes in the workplace needs to be researched more. 

Strengths:

• I liked that the paper acknowledged the idea of humans in the loop with machine learning systems, which I think highlights the reduced requirement of time and effort to solve tasks, and at the same time, necessitates the demand for human labor.
• The authors rightfully note the significance of skills complementary to automation to thrive in the work environment in the future.
• The paper does a very good job of identifying the capabilities and limitations of machine learning systems along with the factors that affect their impacts on the economic state.
• It was particularly interesting to read how ML would transform the workplace and the factors that play out in the transformation.