The future of AI: beyond dependence on GPUs | Festina Lente - Your leading source of AI news | Turtles AI

The current phase of AI development relies heavily on intensive use of specialized hardware, such as GPUs, but the future seems to hold more efficient solutions that are less dependent on computational resources. According to Gartner, this trend is not sustainable in the long term.

Key points:

•  Generative AI is useful for only a small percentage of enterprise workloads.
•  Dependence on specialized hardware, such as GPUs, may be temporary and likely to decline.
•  Established AI techniques, such as machine learning, remain critical for many industries.
•  Composite AI, which combines generative artificial intelligence and traditional techniques, is a more practical and secure solution.

In the current AI landscape, there is a phase of intensive use of hardware resources, particularly GPUs, needed to perform the complex computations required by advanced machine learning and machine generation models. However, some experts believe that this dependence on specialized hardware is a symptom of a transitional period referred to as the “brute force phase” of AI development. During this phase, computing power is used as compensation for the lack of optimized programming methods, but in the future, as techniques evolve, AI may rely less on such resources.

According to Erick Brethenoux, head of AI research at Gartner, the current emphasis on using GPUs to cope with complex AI workloads is likely to decline. Brethenoux likened this to a cycle seen before, where specialized hardware was often supplanted by more generic and affordable solutions once technologies became more efficient. In this context, he said that AI techniques that require a great deal of computing power today will likely be made obsolete by new developments that will allow the same goals to be achieved with standard computational resources.

The discussion of generative AI is central to this debate. While generative AI has received a great deal of media attention for its creative capabilities, such as text or image production, Gartner estimates that only a small fraction of enterprise workloads-about five percent-are really suitable for using this technology. Brethenoux pointed out that many organizations, after initial enthusiasm for generative AI, are returning to focus on more established techniques, such as machine learning or rule-based systems, which have proven effective in various business scenarios.

One emerging approach that could provide a balance between these different technologies is that of “composite AI,” in which generative AI is integrated with traditional techniques to improve the quality of results. For example, generative AI can be used to create descriptions or recommendations based on data analyzed by established machine learning models. This approach takes advantage of generative AI’s ability to produce content flexibly, but without losing the robustness and accuracy provided by more mature methods.

Bern Elliot, vice president at Gartner, also expressed caution about the overuse of generative AI. According to him, although the technology has improved greatly, reducing instances where it produces incorrect answers or “hallucinations,” it still remains unreliable in contexts that require absolute reasoning or accuracy. Elliot compared generative AI to a Swiss cheese, pointing out that although it appears solid, it still has unpredictable shortcomings. He also warned that even with a very low error rate, a large-scale application could still lead to a considerable number of wrong answers, creating significant problems in daily operations.

The prudent use of generative AI, flanked by other, more tested techniques, could therefore be the key to more effective and secure integration of these technologies. In areas such as cybersecurity, for example, some firewall log analysis software is beginning to use generative AI to suggest prose recommendations or even write new security rules, but such operations are still verified by traditional AI systems to ensure accuracy and reliability.

Gartner’s view is that although generative AI represents an important stage in technological evolution, its practical applications remain limited and its use must be carefully considered. Established AI techniques will continue to play a central role in business processes, while the brute force phase of AI, characterized by intensive use of specialized hardware, may gradually give way to an era in which software solutions will be able to offer similar performance with more common resources.

Generative AI, while offering new possibilities, is not a universal solution. Organizations will need to combine different technologies to take full advantage of AI without relying too heavily on single approaches or high computational resources.