More than two years after ChatGPT’s debut, scientist and writer Gary Marcus remains one of generative artificial intelligence’s most vocal skeptics, offering a starkly contrarian perspective amidst Silicon Valley’s optimism. Marcus rose to prominence in 2023, sharing a Senate hearing stage with OpenAI CEO Sam Altman, where both advocated for responsible governance of Artificial Intelligence technologies. However, while Altman subsequently pursued rapid international expansion and massive valuations for OpenAI, Marcus has continued to warn that the field’s current trajectory is fundamentally flawed.
Marcus’s central contention is that generative Artificial Intelligence, especially the large language models (LLMs) powering tools like ChatGPT, are inherently limited and unlikely to fulfill their promised societal transformation. At the Web Summit in Vancouver, he criticized industry leaders for prioritizing hype and funding over real progress, arguing that LLMs will never achieve the broadly human-level intelligence many expect. According to Marcus, the community’s fixation on LLMs stifles investment in alternative methods that could actually create more robust and reliable Artificial Intelligence systems. “I’m skeptical of AI as it is currently practiced,” he stated, citing the technology’s persistent flaws and the unimpressive utility of chatbots and image generators, which often serve as little more than vehicles for memes or deepfakes.
Advocating for neurosymbolic Artificial Intelligence, which seeks to emulate human logic rather than rely solely on pattern recognition from vast datasets, Marcus pointed out the ongoing issue of hallucinations—where Artificial Intelligence produces incorrect but convincing information. He recounted a 2023 wager with LinkedIn founder Reid Hoffman, who predicted hallucinations would quickly disappear, but Marcus insists such flaws are endemic to current approaches. He also raises concerns that once investors recognize generative Artificial Intelligence’s limitations, data monetization and surveillance could become the primary business models, posing risks to privacy and society. Marcus envisions a future where generative Artificial Intelligence may be useful for low-stakes tasks like code suggestions and brainstorming, but believes transformative impact and profitability will remain elusive unless the industry embraces fundamental change.
