Crashing waves vs. rising tides: Overturning prior views about how AI could overtake human workers
Anthropic CEO Dario Amodei has said that AI could surpass "almost all humans at almost everything" shortly after 2027. While AI's capabilities are certainly improving, such rapid progress might seem at odds with findings that show AI is still failing at 95%+ of remote freelance projects, and continues to struggle with hallucination, long term planning, and forms of abstract reasoning that humans find easy. But recent work from METR has found evidence that LLMs can gain capabilities in rapid surgesājumping from succeeding almost never to almost always in just a few years. If this is true across the economy, it could mean that workers could be blindsided by AI advances.
In recent years, the rapid advancements of artificial intelligence (AI) have sparked debates about its potential to overtake human workers in various fields. Anthropic CEO Dario Amodei has predicted that AI could surpass "almost all humans at almost everything" shortly after 2027. While AI's capabilities are undeniably improving, such rapid progress might seem at odds with existing evidence. Studies have shown that AI is still failing at 95%+ of remote freelance projects, struggling with hallucination, long-term planning, and forms of abstract reasoning that humans find effortless. However, recent research from the Machine Intelligence Research (MIRI) organization, specifically a project called METR, has uncovered intriguing evidence that could challenge these prior assumptions.
METR's findings suggest that large language models (LLMs) can gain capabilities in rapid surgesājumping from succeeding almost never to almost always in just a few years. This phenomenon, known as "capability gains," has been observed in the context of language tasks, but if this trend holds true across the economy, it could mean that workers could be blindsided by AI advances. The implications of such a scenario are profound, as it challenges the conventional wisdom that AI progress is incremental and gradual.
The discrepancy between Amodei's optimistic predictions and the current state of AI can be attributed to several factors. Firstly, the tasks that AI is currently failing atāsuch as remote freelance projectsāmight not be representative of the broader range of capabilities required for many jobs. These tasks often involve complex, real-world scenarios that require a combination of technical, creative, and interpersonal skills. In contrast, AI's rapid capability gains have been more pronounced in specific, well-defined domains, such as language understanding and generation.
Secondly, the struggle of AI with hallucination, long-term planning, and abstract reasoning might be a reflection of current limitations in model design and training. As AI systems become larger and more sophisticated, they may develop the capacity to handle these challenges more effectively. The recent success of models like ChatGPT and the ongoing advancements in generative AI are testament to this potential.
However, the possibility of AI overtaking humans in a matter of years raises concerns about the economic and social impacts. Workers in industries reliant on these skills could face significant disruption, with many jobs becoming automated or rendered obsolete. This could lead to widespread unemployment, income inequality, and societal upheaval.
Moreover, the rapid pace of AI development poses challenges for policymakers and stakeholders. If AI is capable of making rapid surges in capability, it becomes difficult to anticipate and prepare for the changes that will ensue. Regulatory frameworks, labor markets, and education systems may struggle to keep pace with such rapid advancements.
In light of these findings, it is crucial for researchers, policymakers, and the public to engage in open dialogue about the future of work in an AI-driven world. Understanding the potential trajectory of AI's capability gains is essential for making informed decisions about how to navigate this transformative period.
While the METR study provides compelling evidence of AI's potential for rapid progress, it is essential to approach these findings with caution. The research is still in its early stages, and the extent to which these capability surges can be replicated across diverse domains remains uncertain. Nonetheless, the possibility of AI making significant strides in the coming years underscores the need for continued investment in research, collaboration, and proactive policy-making to ensure a smooth transition for both workers and society as a whole.
In conclusion, the recent evidence from METR challenges prior assumptions about the pace of AI's advancements and its potential to overtake human workers. While the current state of AI might suggest a more gradual trajectory, the potential for rapid capability gains across the economy cannot be ignored. As AI continues to evolve, it is imperative for all stakeholders to remain vigilant and prepared for the profound changes that lie ahead.
