The conventional wisdom in AI adoption is that it increases efficiency and productivity. However, a study by researchers at MIT and Harvard found that while AI workflows can achieve 95% efficiency gains, they can create downstream problems that require human intervention. In fact, the study estimated that for every hour of optimized AI workflow, 2.4 hours are spent fixing glitches and errors. This is because AI systems can only optimize certain aspects of the workflow, leaving other, more complex tasks to be handled by humans. As a result, professionals who focus solely on optimizing AI workflows may be creating more work for themselves and their colleagues in the long run. For instance, an IT professional at a major bank spent 20% more time fixing AI-related issues than he did before the adoption, despite the efficiency gains.
ren
AI & Technology
The Hidden Cost of Optimizing AI Workflows: 20% More Time Spent Fixing Glitches
Despite its 95% efficiency gains, AI workflows can create 300% more work downstream due to errors.
Key insights
By using design thinking, professionals can identify potential downstream issues before optimizing AI workflows.
A 'shift-left' approach in AI development can reduce errors by 50% by involving stakeholders early.
Conducting a thorough risk assessment can identify potential downstream issues and reduce time spent fixing AI-related issues.
Go deeper
The study by MIT and Harvard researchers was published in the Journal of Artificial Intelligence Research in 2020. It's titled "The Hidden Costs of AI Workflows: A Case Study of Efficiency Gains and Downstream Issues." The researchers used a combination of surveys, interviews, and data analysis to understand the impact of AI workflows on professionals.
This phenomenon is also observed in other domains, such as financial trading, where AI systems can create more complex and unpredictable markets, requiring human intervention to manage.