Democratizing machine learning has become a goal for CIOs and other business leaders looking to scale AI across the enterprise.
The benefits of ML range from improved demand forecasting to improved fraud detection. But to extend the value of technology, it’s important to bring it to more employees, especially those in non-technical business roles. Enticed by the potential of ML to provide sharper insights to improve decision-making, companies will inevitably want to expand their franchises to a wider audience.
However, bringing machine learning to the mainstream enterprise is easier said than done. September 2023 Forrester Consulting studyCommissioned by Capital One, it pointed out a disconnect within the organization. Forrester Consulting surveyed 100 data scientists and 81 line-of-business (LOB) decision makers to find out who their LOB managers want more ML access and the data specialists responsible for making it happen. We found that there is an expectation gap between
There is also a divide among data owners who instinctively want to maintain data silos rather than share it for the common good of the enterprise.
“Business silos” hinder the democratization of ML
“ML is forcing organizations to fight not just data silos, but business silos,” said Forrester Research analyst Michele Goetz.
With ML and, increasingly, generative AI, data access is not only technical but also political, she said. One group within an organization can understand how its data can support another group’s business scenarios and use cases, or how changing the data might conflict with those scenarios and use cases. When you miss the gender, the political side becomes obvious.
“In order to democratize ML, organizations are realizing that they first need to build bridges between different parts of the organization,” Goetz says.
Vinod Chandrasekaran, vice president of products at Capital One Data Insights, said the report revealed that cultural factors are one of the keys to successful democratization. “This includes collaboration, communication, and training.” He noted that 64% of the report’s respondents agreed that a lack of training slows down the adoption of ML workflows.
Technical issues in democratizing machine learning
Chandrasekharan pointed to the different views on the technical details of machine learning as the most important finding of Forrester’s research.
“What struck me was the disconnect between business leaders’ expectations for large-scale ML deployments and the reality of what engineers and data scientists can actually build and deliver on time and at scale. “There was,” he said.
51% of LOB respondents strongly agree that cross-role data collaboration is increasing in their organizations, and 36% of their data manager counterparts share a similar opinion. Despite growing confidence, LOB administrators “may not fully understand what is needed to support democracy,” the report said.
Chandrasekaran said data leaders surveyed emphasized that “doing machine learning” is not an easy task. He noted that technical challenges to democratizing ML include issues around using the right algorithmic techniques and approaches, a concern cited by his 45% of data controller respondents. did.
Additionally, the report identified the ease of use of AI tools as a significant bottleneck preventing widespread adoption of machine learning. 95% of LOB leaders say ML is important or very important to business success, but 67% are slowing enterprise-wide adoption due to a lack of easy-to-use tools says.
Best practices for ML democratization
Michele GoetzForrester Research Analyst
Goetz said business leaders (CEOs, CIOs, CTOs) recognize that a new approach is needed in their organizations to ensure machine learning is democratized and accountable. He noted that generative AI has led some companies to launch AI governance and literacy programs. Such programs establish policies, protections, and education to guide the appropriate use of models and information. Meanwhile, Goetz said the technology team is revisiting its AI strategy and investments to identify gaps that impact how ML is built, managed, and governed.
For additional guidance, she said the National Institute of Standards and Technology and the Organization for Economic Co-operation and Development are rolling out frameworks and tools to help organizations deploy and manage ML responsibly.
Chandrasekharan said he recommends ML adopters modernize their computing environments and use the cloud at every stage of model development. He also noted that Capital One has standardized tools, processes and platforms. That effort includes “moving teams onto the same stack, focusing on collaboration, breaking down silos, and prioritizing reusable components and frameworks across all his ML efforts.” he said.
Other practices include automating the monitoring and training of ML models while maintaining human oversight, and providing low-code/no-code tools to help employees leverage ML capabilities, he said. added.
John Moore is a TechTarget editorial writer, covering CIO roles, economic trends, and the IT services industry.