Limitations of AGI: Challenging the Multimodal Assumption

The recent successes of generative AI models have led some to believe that Artificial General Intelligence (AGI) is on the horizon. However, these models often defy basic intuitions about human intelligence. According to Terry Winograd, by projecting language as the model for thought, we overlook the tacit, embodied understanding that underlies human intelligence. The emergence of these models is largely due to their ability to scale, rather than being thoughtful solutions to the problem of intelligence. This challenges the assumption that AGI must be multimodal, highlighting the need for a more nuanced understanding of intelligence.