Companies Increase AI Automation After Mistakes, Study Finds
Despite AI failures, companies are accelerating automation and reducing human oversight in AI deployments. New research reveals a disconnect between trust and real-world outcomes.
Enterprises that have experienced failures with AI systems in production are accelerating the reduction of human involvement in deployment decisions, according to new research from VentureBeat Intelligence. This trend is occurring even as trust in automated evaluation processes is on the rise.
In July, 13% of 108 surveyed enterprises reported trusting automated evaluations, a significant increase from 5% the previous month. Simultaneously, concerns about the misalignment between tests and real-world results dropped by 10 percentage points to 19%. Despite this, 49% of companies reported that an AI feature, which had passed its tests, went on to cause a customer-facing problem – a figure that remained largely unchanged.
The situation presents a paradox: companies that have encountered issues with test-passing AI systems are more likely to doubt the automated review process. Yet, 85% of these "burned" companies are pushing for automated deployment without human approval, compared to only 61% of companies that have not experienced similar problems. This suggests that companies are increasing automation in response to failures, potentially to enhance deployment maturity.
According to the study, companies that have experienced failures are moving faster towards a model where deployments occur without human approval. Only 11% of these companies reject fully automated deployment for the future, versus 24% of those without such issues. This development indicates a clear trend towards increased autonomy in AI systems.