Democratizing machine learning has become a goal for CIOs and other business leaders attempting to scale AI across the enterprise.
The benefits of ML range from better demand forecasting to improved fraud detection. But getting the technology into the hands of more employees, especially those in non-technical business roles, is critical for expanding the technology’s value. Businesses sold on ML’s potential to deliver sharper insights for improved decision-making naturally want to expand the franchise to a bigger audience.
But bringing machine learning to the corporate masses is easier said than done. A September 2023 Forrester Consulting study, commissioned by Capital One, pointed to disconnects within organizations. Forrester Consulting, which polled 100 data science and 81 line-of-business (LOB) decision-makers, found an expectation gap between the LOB managers who want more ML access and the data specialists charged with making it happen.
There’s also a divide among data owners whose instinct is to maintain silos of data rather than share it for the common enterprise good.
‘Business silos’ hinder ML democratization
“ML is forcing organizations to contend with their business silos and not just the data silos,” said Michele Goetz, an analyst at Forrester Research.
With ML and, increasingly, generative AI, data access becomes political as well as technical, she said. The political aspect reveals itself when one group within an organization overlooks how its data could support another group’s business scenarios and use cases or how changing its data could create conflict with those scenarios and use cases.
“To democratize ML, organizations are realizing they need to first build bridges between parts of the organization,” Goetz said.
The report revealed that cultural components are among the keys to successful democratization, said Vinod Chandrasekharan, vice president of product at Capital One Data Insights. “This includes collaboration, communication and training,” noting that 64% of the report’s respondents agreed that lack of training slows ML workflow adoption.
Technical issues in democratizing machine learning
Chandrasekharan pointed to differing views on the technical details of machine learning as a top finding of the Forrester study.
“What stood out for me was the disconnect between business leaders’ expectations for wide-scale ML deployment and the reality of what engineers and data scientists can actually build and deliver on time and at scale,” he said.
Business and technology leaders aim to drive machine learning’s benefits across a wider constituency within their organizations.
Fifty-one percent of the LOB respondents strongly agreed that data engagement across roles is expanding within their organizations, with 36% of their data manager counterparts sharing that sentiment. While LOB managers grow in confidence, “they may not have a complete understanding of what still needs to happen to support democratization,” the report noted.
Data leaders responding to the survey emphasized that “doing ML” is no simple task, Chandrasekharan said. Technical challenges to ML democratization include issues with using correct algorithmic techniques and approaches, with 45% of data manager respondents citing that concern, he noted.
In addition, the report identified the usability of AI tools as a critical bottleneck impeding wider use of machine learning. While 95% of LOB leaders cited ML as important or very important to business success, 67% said a lack of easy-to-use tools slows cross-enterprise adoption.
Best practices for ML democratization
To democratize ML, organizations are realizing they need to first build bridges between parts of the organization. Michele GoetzAnalyst, Forrester Research
Goetz said enterprise executives — CEOs, CIOs and CTOs — recognize their organizations need a new approach to ensure machine learning becomes democratized and responsible. With generative AI providing a spark, some enterprises have launched AI governance and literacy programs, she noted. Such programs establish policies, protections and education to guide the appropriate use of models and information. Technology teams, meanwhile, are revisiting their AI strategies and investments to identify gaps that affect building, managing and governing ML, Goetz said.
As for additional guidance, the National Institute of Standards and Technology and the Organization for Economic Co-operation and Development have rolled out frameworks and tools to help organizations responsibly deploy and manage ML, she said.
Chandrasekharan said he recommends ML adopters modernize their compute environments to use cloud in every stage of model development. He also noted Capital One has standardized tools, processes and platforms. That effort includes “moving teams to the same stack, focusing on collaboration, bringing down silos and prioritizing reusable components and frameworks across all ML efforts,” he said.
Other practices include automating ML model monitoring and training while maintaining human oversight and providing low-code/no-code tools to help employees take advantage of ML capabilities, he added.
John Moore is a writer for TechTarget Editorial covering the CIO role, economic trends and the IT services industry.
![Mitigating Hallucinations of Large Language Models via Knowledge Consistent Alignment. (arXiv:2401.10768v1 [cs.CL])](https://aigumbo.com/wp-content/uploads/2023/12/arxiv-logo-fb-235x190.png "Mitigating Hallucinations of Large Language Models via Knowledge Consistent Alignment. (arXiv:2401.10768v1 [cs.CL])"){kind=logo}
![Machine Learning in Automation [A Guide]](https://aigumbo.com/wp-content/uploads/2024/01/machine-learning.webp-235x190.jpeg "Machine Learning in Automation [A Guide]")
