The latest findings from Appen have highlighted a significant hurdle facing American companies on their journey towards artificial intelligence (AI) advancements. The "State of AI in 2024" report, released recently, underscores the mounting challenges associated with managing and obtaining high-quality data necessary for training AI models. Sourcing the right data emerges as a critical obstacle, according to insights gleaned from a survey conducted by Harris Poll on behalf of Appen, which involved over 500 IT decision-makers from US firms.

AI’s dependence on data is unequivocal—whether organisations are building their own models, refining existing ones, or employing Retrieval-Augmented Generation (RAG) techniques with pre-constructed models. Successful AI deployments hinge on the availability of large volumes of meticulously curated, high-quality data. Appen, a seasoned provider of data annotation and labelling solutions, has consistently reported on these data sourcing challenges through its State of AI reports, which are now in their fourth edition.

The 2024 report reveals a continued decline in data accuracy and availability, marking troubling trends from previous years. Specifically, there has been a 9 percentage point drop in reported data accuracy over four years, alongside a 6 percentage point increase in data availability challenges since 2023. Appen attributes these issues in part to a transition from projects utilising structured data towards more sophisticated projects involving unstructured data, which has been the focus over the past two years.

Si Chen, Appen’s Vice President of Strategy, attributes the decline partly to the complexity of unstructured data, which often requires domain-specific and subject matter expertise to process effectively. "The data people need today are inherently more complex than before,” says Chen. This complexity challenges organisations to maintain the necessary accuracy and structure within their datasets.

The report also identifies an increasing bottleneck in the AI data pipeline. Companies face hurdles at various stages—accessing data, managing it appropriately, and possessing the technical knowledge to work with it. Since 2023, there's been a 10 percentage point rise in data-sourcing-related bottlenecks, according to Appen.

Chen suggests that the increase in diverse AI initiatives is a contributing factor to these challenges. As organisations explore different use cases for AI, the demand for custom data sets unique to each use case has accelerated. "Every enterprise-specific use case demands tailor-made data sets,” she notes, leading to an increased need for robust data pipelines and skilled personnel adept in handling unstructured data.

The shift in focus towards generative AI has been substantial; Appen’s survey indicates a 17% increase in organisation adoption of GenAI use cases from 2023 to 2024. Presently, 56% of organisations are reportedly engaging with GenAI, with the most common application being the enhancement of productivity in internal business processes (53%). Another 41% utilise GenAI to lower operational costs.

However, as interest in GenAI mounts, the success rates of AI project deployments have seen a downward trend. In 2021, approximately 55.5% of AI projects had reached deployment stages—a figure that has dipped to 47.4% in 2024. The success in achieving meaningful returns on investment (ROI) has also declined from 56.7% in 2021 to 47.3% in 2024, reflecting the ongoing data challenges outlined by Chen. The central role of data in determining deployment success rates is evident, with difficulties in sourcing, cleaning, and structuring data becoming critical factors in project outcomes.

The report categorises AI data usage into three primary types: 27% utilise pre-labeled data, 30% employ synthetic data, and 41% rely on custom-collected data. The latter, according to Appen CEO Ryan Kolln, offers a competitive edge, as the capability to gather unique, custom data sets provides robust advantages within the generative AI space. Kolln highlights the significance of bespoke data—data unutilised by conventional model builders—as crucial for establishing competitive superiority.

In summary, Appen’s "State of AI in 2024" report paints a comprehensive picture of the evolving dynamics and challenges within AI data management. It projects a landscape where traditional AI requires adaptation to incorporate unstructured data and underscores the increasingly vital role that unique data sources play in fostering successful AI deployments.

Source: Noah Wire Services