Key facts
- AGI, or artificial general intelligence, is often defined as an AI model capable of performing a wide variety of tasks as well as or better than a human.
- Peter Voss, an early user of the term, stated that current AI systems struggle with tasks like learning new skills with human plasticity.
- Anthropic CEO Dario Amodei described AGI as a "marketing term" due to its vague definition.
- Researchers like Emily Bender argue that AGI, as a concept of a general-purpose thinking machine, is not currently possible to engineer.
- Yann LeCun, former chief AI scientist at Meta, stated that simply increasing data and compute does not guarantee smarter AI.
The term Artificial General Intelligence (AGI) is increasingly being used by tech leaders, but many experts argue it is a poorly defined concept and that current AI systems do not meet the criteria.
OpenAI president Greg Brockman declared in September that the company's latest AI model represented the "AGI era," personally believing they had reached that point. Following this, Nvidia CEO Jensen Huang posted on X that "AGI has arrived." Elon Musk also commented on X that he "profoundly" felt AGI after viewing a short film generated by Claude. Factory CEO Matan Grinberg echoed these sentiments, stating that AGI is already here and we are living in a post-AGI world.
However, others in the industry view these claims with skepticism. Peter Voss, one of the first software engineers to use the term AGI, called the claims "complete nonsense" and dishonest marketing. He argued that current AI models struggle with tasks that require human-like plasticity and learning, citing the significant resources needed to train AI for simple tasks like customer support with still poor results.
Some, like Anthropic CEO Dario Amodei, have been more reserved, referring to AGI as a "marketing term." Researchers, including Emily Bender from the University of Washington, suggest that companies use futuristic nomenclature to create an "illusion" of advancement, propagating fictional ideas rather than existing technology. Bender believes a general-purpose thinking machine is not within the realm of possibility.
Alan Chan, an AI research fellow at GovAI, noted that AGI is vague and often conflates an AI's ability to perform new tasks with its ability to learn and adapt like humans. He pointed out that systems excelling in one area, like coding, may still require extensive training for new jobs.
Experts like Voss and Yann LeCun, former chief AI scientist at Meta, argue that the current strategy of scaling Large Language Models (LLMs) with more data and compute will not lead to AGI. LeCun stated that more data and compute do not automatically equate to smarter AI, and Ilya Sutskever, a cofounder of OpenAI, has noted that current models generalize much worse than humans, suggesting a fundamental limitation.