In September, during a briefing with journalists, OpenAI's president Greg Brockman proclaimed, "Welcome to the AGI era," referring to the company's new AI model and expressing his personal belief that we have arrived at this milestone.
Tech industry leaders, including notable figures on social media platform X, are increasingly referencing a term that once signified a far-off aspiration in artificial intelligence: AGI, or artificial general intelligence. This concept generally describes an AI system capable of performing a multitude of tasks at a level comparable to or exceeding that of a human being. Alongside Brockman, other influential voices like Elon Musk, Sam Altman, and Jensen Huang have also suggested that we are on the brink of achieving AGI.
However, dissenting opinions from within the industry have emerged, with experts arguing that the term lacks a clear definition, rendering it ineffective as a meaningful metric for progress. Some have gone so far as to label claims of AGI being realized as dubious marketing strategies.
Peter Voss, one of the early adopters of the term AGI, dismissed it as mere "marketing talk" and expressed skepticism about the current state of AI. He argued that, if true AGI existed, it would be evident to everyone.
The origins of the term AGI can be traced back to 2002, when Ben Goertzel faced challenges in naming his book related to versatile AI technologies. After consulting peers, he ultimately chose "Artificial General Intelligence." Voss, who contributed to the compilation, believed that such a system represented an unattained dream, highlighting that today’s chatbots remain constrained in their capabilities and struggle to learn new skills with the adaptability of humans.
According to Voss, while humans can quickly adapt to roles like call center work after minimal training, current AI systems falter in similar scenarios. "Achieving even basic customer support functionality requires massive investments in engineering and vast datasets, and the outcomes are subpar," he explained.
Despite these challenges, the excitement around AGI persists among tech leaders and investors. Following the unveiling of OpenAI's Astra model, Jensen Huang celebrated the moment with a post on X, confidently declaring the arrival of AGI. He acknowledged the rapid advancements from ChatGPT to Astra over just four years and commended OpenAI's team.
Elon Musk joined the conversation this week, sharing his experience of feeling the presence of AGI after viewing a short film generated by Claude. Similarly, Factory CEO Matan Grinberg remarked on the Sources podcast that "we are living in a post-AGI world" and suggested that we have moved beyond a critical technological threshold.
In contrast, other industry leaders, such as Dario Amodei of Anthropic, have been more cautious. Amodei pointed out that AGI has never been clearly defined and has often seemed more like a marketing device than a definitive concept.
Emerging AI companies face substantial pressure to demonstrate progress as they eye potentially enormous IPOs. Emily Bender, a professor at the University of Washington, asserted that by invoking futuristic terminology reminiscent of science fiction, these firms create an unrealistic perception regarding the capabilities of current technology.
Bender argued that the public's expectations are shaped by fictional portrayals of advanced machines, and she contended that the term AGI lacks connection to any existing technology, dubbing it a marketing piece.
Alan Chan of GovAI, a think tank focused on AI policy, echoed this sentiment, criticizing AGI as an ambiguous benchmark for measuring advancement. He highlighted the confusion caused by conflating an AI's ability to learn new tasks with its capacity to execute established tasks effectively.
If one accepts Voss’s definition, it could be a considerable wait before we see AI systems capable of independently navigating tasks such as driving a car after a single day of training. He noted that the recent fascination with large language models (LLMs) among investors may be stifling alternative avenues of research and development.
Yann LeCun, a prominent figure in AI research, cautioned against assuming that increased data and computational power would naturally lead to smarter AI, challenging the prevailing trend. Meanwhile, OpenAI cofounder Ilya Sutskever posited that a return to fundamental research within the AI landscape will be necessary, highlighting that current models struggle to generalize tasks as well as humans do.
Bender concluded that the creation of a general-purpose thinking machine appears to be beyond human capability, citing an engineering approach that simply is not feasible.
