Erick Brethenoux, Gartner's lead AI researcher, suggests that specialized AI hardware, particularly GPUs, will become less necessary as AI technology becomes more sophisticated. Addressing the Gartner Symposium in Australia, he highlighted that the current phase of AI relies heavily on powerful hardware due to primitive programming strategies. As programming advances, AI tasks will be manageable with more general-purpose computers.
Generative AI's Limited Impact
Brethenoux elaborated on the limited applicability of generative AI, despite its buzz. He mentioned that generative AI is relevant to only a minor fraction of use cases, around five percent. Many organizations have realized that their operations can be significantly enhanced without generative AI, often reverting to more traditional methods.
Composite AI, which blends generative AI with other strategies like machine learning and knowledge graphs, is becoming more prevalent. According to Brethenoux, companies are embracing this approach for tasks like predictive maintenance and firewall log analysis. The method provides descriptive insights or recommendations, ensuring more efficient and reliable solutions.
Caution on Generative AI Dependence
In another session, vice president Bern Elliot advised against over-reliance on generative AI. In a discussion titled “When not to use generative AI,” Elliot highlighted that generative AI mainly produces probabilistic content and lacks reasoning abilities. He suggested using it mainly for content creation and conversational interfaces, while other AI technologies should verify its outputs.
Brethenoux described the period from late 2022 to early 2024 as a “recess” during which IT departments experimented with generative AI, often at the cost of revenue-generating activities. As the hype around generative AI fades, companies are shifting back to AI applications that directly support business goals, marking the conclusion of this experimental phase.
Soaring Costs of AI Development
Also during the event, Gartner cautioned that AI development will send costs soaring for businesses and dent the morale of their employees. Gartner Vice President and Analyst Mary Mesaglio highlighted the potential for organizations to significantly overspend on generative AI, mirroring the unexpected financial implications encountered during the early adoption of cloud services. Mesaglio indicated that AI costs could be substantially underestimated, suggesting a potential increase in expenditures ranging from 500 to 1000 percent.
Several factors could contribute to this financial strain, including heightened vendor charges and excessive cloud resource utilization during AI experimentation. Ms. Mesaglio advised against employing AI for simple inquiries that could be more effectively addressed through traditional search methods, as this could lead to unnecessary expenses. She noted that longer and more complex user queries could drive up costs under token-based pricing structures.
Mesaglio introduced the concept of “productivity leakage,” referring to instances where time savings attributed to AI, estimated at 43 minutes per day per worker, do not necessarily translate into increased productivity. Instead, employees may utilize this time for personal tasks, such as obtaining coffee, potentially resulting in a loss of 10 to 30 percent of AI's anticipated productivity gains.
