MIT Professor: Only 1/4 of AI-related tasks will achieve cost-effective automation in the next 10 years
In the next decade, only a quarter of AI-related tasks will achieve cost-effective automation. Even with significant breakthroughs in AI, its impact will take several years to materialize. AI's impact on all tasks is less than 5%, with a 0.5% increase in US productivity and a mere 0.9% contribution to GDP growth. The architecture of large language models may face serious limitations, requiring higher quality data for AI. AI must be able to solve complex problems, but cost-effective solutions are currently lacking
According to the Zhītōng Finance and Economics APP, Daron Acemoglu, a professor at the Massachusetts Institute of Technology (MIT), stated that in the next ten years, only a quarter of AI-related tasks will be cost-effectively automated.
Acemoglu, speaking on the Goldman Sachs podcast, mentioned that even with significant breakthroughs in artificial intelligence, its impact will take several years to materialize.
Goldman Sachs indicated that this means that in the next 10 years, AI will impact less than 5% of all tasks, only increasing US productivity by 0.5%, and contributing a cumulative 0.9% to GDP growth.
He said, "The evidence shows that the current Large Language Model (LLM) architecture is more impressive than many people predicted, but I think that having something as smart as the robot HAL from '2001: A Space Odyssey' just by relying on predicting the next word in this architecture requires a lot of confidence."
Acemoglu stated, "Our current LLM architecture may face very serious limitations."
He also questioned whether AI can achieve its goals faster simply by investing more GPU capacity.
He added that higher quality data will be needed more and more, rather than capacity, and it is currently unclear where this type of data will come from.
Jim Covello, Global Head of Stock Research at Goldman Sachs, stated that in order to see returns from the expected capital expenditures on AI in the coming years by companies such as NVIDIA (NVDA.US), Microsoft (MSFT.US), Google (GOOGL.US), Meta (META.US), Amazon (AMZN.US), and Super Micro Computer (SMCI.US), AI must be able to solve complex problems.
He said, "We have been studying for years, and at this point, nothing is cost-effective. I think there is an incredible misunderstanding about the role of this technology. The problems it can solve are not big problems. There is no cognitive reasoning involved here."