Generative AI grows 17% in 2024, however information high quality plummets: Key findings from Appen’s State of AI Report

A new report from AI information supplier Appen reveals that firms are struggling to supply and handle the high-quality information wanted to energy AI methods as synthetic intelligence expands into enterprise operations.

Appen’s 2024 State of AI report, which surveyed over 500 U.S. IT decision-makers, reveals that generative AI adoption surged 17% previously 12 months; nevertheless, organizations now confront important hurdles in information preparation and high quality assurance. The report exhibits a ten% year-over-year enhance in bottlenecks associated to sourcing, cleansing, and labeling information, underscoring the complexities of constructing and sustaining efficient AI fashions.

Si Chen, Head of Technique at Appen, defined in an interview with VentureBeat: “As AI fashions sort out extra complicated and specialised issues, the information necessities additionally change,” she stated. “Corporations are discovering that simply having a lot of information is not sufficient. To fine-tune a mannequin, information must be extraordinarily high-quality, that means that it’s correct, numerous, correctly labelled, and tailor-made to the precise AI use case.”

Whereas the potential of AI continues to develop, the report identifies a number of key areas the place firms are encountering obstacles. Beneath are the highest 5 takeaways from Appen’s 2024 State of AI report:

1. Generative AI adoption is hovering — however so are information challenges

The adoption of generative AI (GenAI) has grown by a formidable 17% in 2024, pushed by developments in giant language fashions (LLMs) that permit companies to automate duties throughout a variety of use circumstances. From IT operations to R&D, firms are leveraging GenAI to streamline inside processes and enhance productiveness. Nonetheless, the fast uptick in GenAI utilization has additionally launched new hurdles, significantly round information administration.

“Generative AI outputs are extra numerous, unpredictable, and subjective, making it tougher to outline and measure success,” Chen instructed VentureBeat. “To attain enterprise-ready AI, fashions should be custom-made with high-quality information tailor-made to particular use circumstances.”

Customized information assortment has emerged as the first methodology for sourcing coaching information for GenAI fashions, reflecting a broader shift away from generic web-scraped information in favor of tailor-made, dependable datasets.

2. Enterprise AI deployments and ROI are declining

Regardless of the thrill surrounding AI, the report discovered a worrying pattern: fewer AI tasks are reaching deployment, and those who do are displaying much less ROI. Since 2021, the imply share of AI tasks making it to deployment has dropped by 8.1%, whereas the imply share of deployed AI tasks displaying significant ROI has decreased by 9.4%.

This decline is essentially because of the growing complexity of AI fashions. Easy use circumstances like picture recognition and speech automation at the moment are thought-about mature applied sciences, however firms are shifting towards extra bold AI initiatives, corresponding to generative AI, which require custom-made, high-quality information and are far harder to implement efficiently.

Chen defined, “Generative AI has extra superior capabilities in understanding, reasoning, and content material technology, however these applied sciences are inherently more difficult to implement.”

3. Information high quality is important — nevertheless it’s declining

The report highlights a vital problem for AI growth: information accuracy has dropped practically 9% since 2021. As AI fashions develop into extra refined, the information they require has additionally develop into extra complicated, usually requiring specialised, high-quality annotations.

A staggering 86% of firms now retrain or replace their fashions not less than as soon as each quarter, underscoring the necessity for recent, related information. But, because the frequency of updates will increase, guaranteeing that this information is correct and numerous turns into harder. Corporations are turning to exterior information suppliers to assist meet these calls for, with practically 90% of companies counting on exterior sources to coach and consider their fashions.

“Whereas we will’t predict the longer term, our analysis exhibits that managing information high quality will proceed to be a serious problem for firms,” stated Chen. “With extra complicated generative AI fashions, sourcing, cleansing, and labeling information have already develop into key bottlenecks.”

4. Information bottlenecks are worsening

Appen’s report reveals a ten% year-over-year enhance in bottlenecks associated to sourcing, cleansing, and labeling information. These bottlenecks are immediately impacting the power of firms to efficiently deploy AI tasks. As AI use circumstances develop into extra specialised, the problem of getting ready the correct information turns into extra acute.

“Information preparation points have intensified,” stated Chen. “The specialised nature of those fashions calls for new, tailor-made datasets.”

To deal with these issues, firms are specializing in long-term methods that emphasize information accuracy, consistency, and variety. Many are additionally searching for strategic partnerships with information suppliers to assist navigate the complexities of the AI information lifecycle.

5. Human-in-the-Loop is Extra Important Than Ever

Whereas AI expertise continues to evolve, human involvement stays indispensable. The report discovered that 80% of respondents emphasised the significance of human-in-the-loop machine studying, a course of the place human experience is used to information and enhance AI fashions.

“Human involvement stays important for growing high-performing, moral, and contextually related AI methods,” stated Chen.

Human consultants are significantly necessary for guaranteeing bias mitigation and moral AI growth. By offering domain-specific data and figuring out potential biases in AI outputs, they assist refine fashions and align them with real-world behaviors and values. That is particularly vital for generative AI, the place outputs could be unpredictable and require cautious oversight to stop dangerous or biased outcomes.

Take a look at Appen’s full 2024 State of AI report proper right here.