Since 2018, artificial intelligence (AI) developers have been creating broadly capable foundation models—AI models trained on large and diverse datasets that can be adapted to perform a wide variety of tasks. The most advanced of these models have improved rapidly over just a few years and have already demonstrated very impressive performance at certain tasks, but they also consume exponentially increasing amounts of resources, such as electricity and advanced computer chips. This growth raises the question of who might pay for the large costs of developing and deploying new AI models if resource requirements continue to increase. Some stakeholders suggest that world-changing returns to productivity and well-being—along with significant first-mover advantages—can be unlocked if governments can coordinate national and global resources to support massive financial investments. Other stakeholders suggest that the private sector might well be capable of marshaling the resources required to achieve these outcomes. In this paper, the authors lay out a framework of plausibly imaginable futures for the development and training of AI foundation models, focusing on who will pay for such training. The resulting framework can help both government and private stakeholders consider when and whether various funding models are appropriate.
Who Could Fund Future Artificial Intelligence Development? | RAND
