Rethinking The Impact AI Might Have On Work

It’s been a decade since Oxford’s Frey and Osborne published their hugely influential paper on the susceptibility of jobs to automation. The paper sparked a wave of concern about what impact the latest wave of automated technologies would have on the labor market.

While the dire warnings of job losses in the intervening years have largely failed to materialize, the emergence of generative technologies, like ChatGPT, in the last 18 months has sparked fresh concerns that this time, the fears are justified.

Risk of automation

A recent paper from MIT CSAIL explores whether this time it really is different. The study moves beyond task-based comparisons to assess the feasibility of AI systems performing these jobs and the economic viability of businesses adopting such technologies.

The researchers focus on computer vision, and utilize a new model to help them understand both the technical and economic likelihood that tasks will be automated using the technology. Unlike existing models, it considers three crucial factors: 1) the technical performance required for automation, 2) the cost of developing AI systems to achieve that performance, and 3) the economic attractiveness of automation compared to human labor costs.

The study suggests that, as the evidence from the last decade suggests, the actual risk from automation is a bit less than many of the doomsayers predict.

“We find that at today’s costs U.S. businesses would choose not to automate most vision tasks that have “AI Exposure,” and that only 23% of worker wages being paid for vision tasks would be attractive to automate,” the researchers explain. “This slower roll-out of AI can be accelerated if costs fall rapidly or if it is deployed via AI-as-a-service platforms that have greater scale than individual firms, both of which we quantify.”

Overall, the study anticipates gradual effects on jobs, allowing time for adaptation, and indicates that existing job turnover surpasses potential losses from this type of automation.

Cost-effective automation

The study provides a detailed exploration of the cost-effectiveness of automating a range of tasks using computer vision technology. The researchers developed a model that is able to calculate this cost-effectiveness at both the firm and economy levels. The model allows them to identify which tasks are most suitable for automation by computer vision.

In total, they analyze around 400 different vision-related tasks in terms of the economic attractiveness of automation. This process involves breaking down the costs into fixed engineering costs, performance-dependent costs like data and computing, and scale-dependent running costs.

They also utilize a cost model that encompasses the build, maintenance, and operational phases of running a computer vision. The model calculates total costs under two scenarios – a comprehensive system with significant engineering and a “bare bones” system using existing models and minimal engineering.

Highlighting the value

This allows them to understand the true costs and benefits of the technology over its entire lifespan. The researchers also attempt to gauge the labor costs associated with performing particular tasks to understand whether the technology offers better value than people do.

The results indicate that while 36% of US non-farm jobs involve tasks potentially automatable with computer vision, only about 8% are economically attractive for firms to automate based on cost comparisons. The economic attractiveness depends heavily on the costs of the technology, with accuracy requirements, data costs, and engineering costs having a significant impact.

The researchers also analyze how computer vision technologies can be scaled up within larger firms, and the challenges involved in doing so. They suggest that the current economic feasibility is limited for individual firms but could be enhanced through aggregating demand across firms via larger entities or AI-as-a-service models.

Initial changes

After taking all of these things into account, the researchers found that 23% of vision-related tasks could be automated, but this is expected to fall in subsequent years. The authors also point out that this leaves the vast majority of tasks unaffected.

“Looking at computer vision, where the cost estimates for AI systems are more developed, we find that most systems are cost-effective to deploy when single systems can be used across entire sectors or the whole economy,” the authors caution. “Conversely, 77% of vision tasks are not economical to automate if a system can only be used at the firm level. This contrast makes it clear that the cost-effectiveness of AI models will likely play an important role in the proliferation of the technology.”

This, coupled with the relatively slow pace of diffusion of these technologies should encourage those concerned that not only will vast swathes of the labor market be automated, but that people will be unable to adapt to the rapid pace of change.

Facebooktwitterredditpinterestlinkedinmail