The rapid development of AI over recent years, and particularly the rate at which Open AI seems to keep churning out new iterations of its GPT model, make it easy to assume that the technology will continue on a fairly linear trajectory as it matures – perhaps similar to the evolution of the internet. However, there are currently forces at play that could seriously disrupt such a path, hampering the development of the technology or limiting access to it, reports Forbes.
Perhaps the biggest and most imminent risk is that demand for computing power to feed AI algorithms could outstrip available supply – and soon. Our World in Data shows that across virtually all domains, compute requirements to feed AI algorithms are increasing exponentially. Meanwhile, there are questions over whether the supply of GPUs will be sufficient – even accounting for factors such as Moore’s Law resulting in increased capacity.
Another challenge is capability. For instance, parallel processing using multiple GPUs is one way to alleviate the problem, but there are limitations on the extent to which this can be done within a single system at scale, due to diminishing benefits and increasing costs.
