Evaluating the Impact of Automation on Work by Using the Behavioral Robotics Paradigm

February 14, 2021

This analysis argues that the existing task-based approach to evaluating how machines substitute for and complement workers is based upon an outdated paradigm of automation. This paradigm says that machines can only substitute for tasks that follow explicit rules. Non-routine tasks, however, are harder to automate because there are no explicit rules to be applied and, therefore, are still in the domain of human work. The traditional approach to evaluating the impact of automation on work leads to a breakdown of tasks into routine and non-routine buckets.

The new paradigm of robotics, however, is behavioral and behavioral robots have made in-roads into automating non-routine tasks through the adoption of sensors that can monitor the environment–non-routine or otherwise. According to leading behavioral roboticist, Rodney Brooks, automation involves a machine that can sense the environment, make a judgement, and take an action. With sensors getting cheaper, more machines are making use of them to interact with the environment. One website reported that the average cost of sensors was $1.30 in 2004 and fell to $0.38 in 2020. So, machines can now sense their environment, but the domain of judgement still lies squarely with humans.

This post takes a look at which occupations rely on sensing and which occupations rely on judgement and tries to determine whether those that rely on sensing are more likely to be automated. I take data from O*Net, an online portal for career exploration and job analysis. O*Net has a “Degree of Automation” index for each occupational classification. I search for jobs that require judgement and take the top 60% of matches (i.e. the most relevant matches are displayed first) and then do the same for jobs that require sensing. I then extract the degree of automation for each judgement occupation and each sensing occupation and analyze whether they are statistically different.

What is a job that relies on judgement skills? Judgement jobs include judges, umpires/referees, and animal trainers. Sensing jobs include remote sensing scientists, robotics technicians, and geographic information systems technologists. The average degree of automation for judgement jobs is 22.44 index points. The average degree of automation for sensing jobs is 33.19 index points. The difference between the averages is 10.75 index points and is very statistically significant. This means that the difference in degree of automation between sensing and judgement is unlikely to be due to chance.

Analyzing automation by sensing and judgement groups is a behavioral alternative to the task-based approach. Instead of breaking down jobs into routine and non-routine tasks, I break down jobs by their composition of sensing and judgement skills. The advantage is that the behavioral method accounts for automation in non-routine tasks, which would otherwise be missed in the task-based approach.

Leave a Reply

Your email address will not be published. Required fields are marked *