Reasons you might think human-level AI is unlikely to happen soon

Abstract

Asya Bergal claims that the possibility of human-level artificial intelligence (HAI) being developed in the near future (20 years, in her valuation) is fairly unlikely. She bases this claim on three lines of reasoning: (1) a privileged class of experts disagree that HAI can be soon produced, (2) we may run out of “compute” (processing resources) before HAI can be reached, and (3) current methods, such as deep learning or neural networks, may by themselves be insufficient to create HAI. Regarding the first point, Bergal notes that the opinions of experts, while generally optimistic about the possibility of HAI, vary widely; as such, it is difficult to determine a clear consensus. On the second point, she argues that while AI progress has until now been powered by increases in computing power, these increases are likely to slow down in the future, which may hinder AI development. On the third point, Bergal considers that current methods may have fundamental limitations or be impractical due to insufficiencies in their efficacy or in the amount of data that they require to train. – AI-generated abstract.