Costs, skills gaps, data quality, data security and privacy concerns were the top impediments to scaling AI securely/sustainably, noted respondents

In a Dec 2023/Jan 2024 series of 24 in-depth interviews and 1,220 quantitative interviews* about scaling AI initiatives responsibly — with decision makers involved in IT operations, data science, data engineering and software development for AI initiatives —  several trends were discerned and disclosed for public consumption.

The first key trend was that respondents deemed as AI Masters* had optimized their data infrastructure for transformational AI initiatives by facilitating easy access to corporate datasets with minimal preparation, and by designing a unified, hybrid, multi-cloud environment that supports various data types and access methods. They also exhibited more ambitious AI goals and nevertheless experienced data-related failures including infrastructure-based data access limitations (21%), compliance limitations (16%), and insufficient data (17%). This attribute is termed “intelligent data infrastructure”.

Secondly, respondents deemed as AI Emergents indicated similar challenges but also experienced budget constraints (20% Emergents vs 9% AI Masters), insufficient data for model training (26% vs 17%) and business restrictions on data access (28% vs 20%).

Other findings

Thirdly, some 48% of respondents deemed as AI Masters indicated they had instant availability of their structured data and 43% of their unstructured data, while AI Emergents indicated having only 26% and 20% respectively. This pointed to respondents’ organization’s “data infrastructure flexibility”.

Also:

    • AI Masters (65%) and AI Emergents (35%) indicated that their current data architectures can seamlessly integrate their organization’s private data with AI Cloud services.
    • 8% of AI Emergents had completed and standardized governance policies and procedures across all AI projects, compared to 38% of AI Masters. (Note: this is deemed as a factor behind effective data governance and security processes.)
    • 51% of AI Masters indicated they had standardized policies in place that were rigorously enforced by an independent group in their organization, compared to 3% of AI Emergents. (Note: this is deemed as a factor behind effective data governance and security processes.)
    • 43% of AI Masters indicated they had clearly defined metrics for assessing resource efficiency when developing AI models that were completed and standardized across all AI projects, compared to 9% of respondents deemed as AI Emergents. (Note: this is deemed as a factor behind “efficient use of resources important for scaling AI responsibly”.)
    • 63% of all respondents reported the need for major improvements or a complete overhaul to ensure their storage is optimized for AI, while 14% indicated they needed no improvements. (Note: this is deemed as a factor behind “efficient use of resources important for scaling AI responsibly”. Editor’s note: in the context of the survey, “scaling AI responsibly” has specific meanings (focusing on data architecture flexibility; governance and security; resource efficiency) that are distinct from those of the term “responsible AI”.)

According to Jonsi Stefansson, Senior Vice President and Chief Technology Officer, NetApp, the firm that sponsored the interviews, organizations scaling AI initiatives need “the flexibility to access any data, anywhere with integrated data management to ensure data security, protection, and governance and adaptive operations that can optimize performance, cost and sustainability.”

The report also concluded: By prioritizing security, data sovereignty, and regulatory compliance, organizations (with similar AI implementation needs as the various classes of respondents defined in the survey report) can mitigate risk in their AI and generative AI initiatives and ensure that their data engineers and scientists can focus on maximizing efficiency and productivity. Also, “it is critical to acknowledge the impact on compute and storage infrastructure, data and energy resources, and their associated costs. A key measure of AI maturity is the definition and implementation of metrics to assess the efficiency of resource use in the creation of AI models.”

*via web surveys, based on a model of classifying respondents into one of four arbitrary maturity levels based on their current approach to AI in terms of data and storage infrastructure; data policy and governance; resource efficiency focus; and stakeholder enablement and collaboration. From the lowest (predefined) maturity level to the highest, respondents were either “AI Emergents”, “AI Pioneers”, “AI Leaders”, and “AI Masters”. No information was supplied about the respondents’ organization size or other industry attributes.