Gary Marcus discusses AI risks and limitations

Gary Marcus, a researcher in cognitive psychology, neuroscience, and artificial intelligence at New York University, addressed the discourse surrounding AI threats in a discussion with investor Steve Eisman. Marcus noted that while media coverage often adopts a catastrophic tone, similar panics have occurred regarding technology in the past.

While some researchers in AI laboratories estimate a 10% probability that the technology could turn against humanity, Marcus questioned the logic of continuing such development if that risk is accepted. He argued that many claims regarding an ‘AI war’ or existential threats lack concrete mechanisms and rely on vague analogies.

Furthermore, Marcus highlighted the limitations of current large language models. He noted that while these models appear omnipotent, they lack certain forms of intelligence, particularly in physical problem-solving and real-world experimentation.