Rising AI Adoption Prompts Risk Assessments
Adopters of AI in the enterprise are focusing on specific production workloads centered around supervised and deep learning while the number of organizations using AI in production or evaluating the technology jumped to 85 percent of companies polled in an annual survey.
Another indication of maturing enterprise AI initiatives is a heavier emphasis on data governance, according to an AI adoption survey released Wednesday (March 18) by O’Reilly Media. More than 26 percent of respondents said they are instituting formal governance processes as concerns about privacy and “trustworthy” AI grow. Nearly 35 percent of those surveyed said they expect to launch AI data governance efforts over the next three years, O’Reilly reported.
“AI adoption is proceeding apace,” wrote report authors Roger Magoulas and Steve Swoyer. “Most companies that were evaluating or experimenting with AI are now using it in production deployments.”
Next steps include controlling risk factors ranging from bias in model development and ratty data to “the tendency of models to degrade in production,” the authors added.
The majority of AI projects are within research and development departments following by IT and customer service applications.
Meanwhile, hiring and retaining AI experts remains the top barrier to adoption, the survey found. Other hurdles include lack of institutional support (22 percent) and inability to identify appropriate business use cases (20 percent). Hardest to find and keep are machine learning modelers and data scientists.
The biggest AI-related skills gap was data engineering, cited by nearly 40 percent of respondents.
“The uncomfortable truth is that the most critical skill shortages cannot easily be addressed,” the survey authors noted. Data scientists, for example, require a mixed of technical expertise in areas like statistics along with “domain-specific business expertise,” they added.
As the pace of enterprise AI deployments quickens, early adopters are shifting their focus to potential risks in the form of “unexpected outcomes.” Hence, mature programs are focusing on areas like explainable AI and the transparency of machine learning models.
The growing number of AI deployments is also providing researchers a window into which tools developers prefer. The O’Reilly survey found that four of the five top most popular tools are based either on the Python or tools and libraries based on the popular programming language. Python’s growth for machine learning and other AI-related projects was described as “explosive.”
Meanwhile, developer interest in PyTorch, No. 4 in the O’Reilly tool rankings, continues to grow “quickly from a relatively small base,” the authors said. One reason is its use of dynamic computational graphs. Adding to its momentum, Amazon Web Services (NASDAQ: AMZN) said this week it is adding support for PyTorch models with its Elastic Inference tool. That means PyTorch support for inference will be available on Amazon SageMaker, the machine and deep learning stack, along with other AWS cloud platforms.