The key results of the report are the following:
• More is moving to inference and to the edge. As AI technology progresses, inference, the ability of a model to make predictions based on their training, can now be executed closer to users and not only in the cloud. This has advanced the deployment of AI to a variety of different edge devices, including smartphones, cars and internet of industrial things (Iiot). Edge processing reduces cloud dependence to offer faster response times and greater privacy. In the future, Hardware for AI on the device will only improve in areas such as memory capacity and energy efficiency.
• To deliver generalized, organizations are adopting a heterogeneous calculation. To market the complete panoply of cases for the use of AI, processing and computation must be carried out in the right hardware. A heterogeneous approach unlocks a solid and adaptable base for the deployment and progress of the cases of the use of AI for everyday life, work and game. It also allows organizations to prepare for the future of the AI distributed in a reliable, efficient and safe way. But there are many compensations between cloud computer science and the edge that require careful consideration based on the specific needs of the industry.
• Companies face challenges in the management of system complexity and ensure that current architectures can adapt to future needs. Despite the progress in microchip architectures, such as the latest high performance CPU architectures optimized for AI, software and tools need to improve to offer a computer platform that admits generalized automatic learning, generative, generative and new specializations. Experts emphasize the importance of developing adaptable architectures that meet the current demands for automatic learning, while allowing space for technological changes. The benefits of distributed computation need to overcome disadvantages in terms of complexity on all platforms.
This content was produced by Insights, the personalized content arm of the MIT technology review. It was not written by Mit Technology Review Editorial staff.
This content was investigated, designed and written completely by writers, editors, analysts and human illustrators. This includes the writing of surveys and data collection for surveys. The AI tools that may have been used were limited to secondary production processes that passed an exhaustive human review.
