As society continues to adjust to the rapid evolution of artificial intelligence, Sam Altman, CEO of OpenAI, has made a bold prediction: Artificial General Intelligence (AGI) could arrive as soon as 2025. This announcement marks a significant shift in the timeline many experts expected, raising questions about how close we really are to achieving a form of AI that can match or even surpass human intelligence in almost all domains.

AGI refers to an AI system that possesses cognitive abilities comparable to, or more advanced than, humans in virtually every intellectual area. For years, it has been the “holy grail” of AI development, with predictions suggesting it was at least a decade or more away. However, Altman now believes that AGI is not a distant dream but a pressing engineering challenge that could be solved sooner than expected.

According to Altman, the path to AGI is “basically clear,” and OpenAI is making faster progress than anticipated. But the definition of AGI is still a topic of debate in the AI community. Some experts argue that AGI must go beyond simply replicating human knowledge; it must also be capable of learning, adapting, and performing tasks in ways that go beyond its initial programming and training data. This would mean the AI could generate original solutions and insights, not just work within the confines of pre-existing information.

While Altman’s optimism about AGI’s arrival is high, recent benchmarks such as FrontierMath suggest that AI systems are still struggling with certain fundamental aspects of reasoning. In tests measuring how well models handle problems outside their training data, advanced models like GPT-4 and Google’s Gemini 1.5 Pro were able to solve fewer than 2% of the problems presented. This suggests that, based on some definitions, current AI models are still a long way from meeting the criteria for AGI, particularly the ability to reason autonomously and generate solutions outside of their initial datasets.

However, sources within OpenAI have suggested that the upcoming version of GPT-4 (referred to as o1) represents a significant leap in terms of reasoning capabilities. Rumors also point to improvements in the next generation of Google’s Gemini models, indicating that we may soon see a new wave of AI systems that are better equipped for tasks requiring complex problem-solving.

OpenAI’s approach to AGI is structured into a series of levels, with AGI itself sitting at the final stage. According to the company, AGI will be achieved through five levels of AI development, with each level representing a major step forward in terms of capabilities:

• Level 1: Basic chatbots that simulate human conversation through simple text generation, like the models we’ve seen over the past few years.
• Level 2: “Reasoners” capable of performing more complex tasks, such as OpenAI’s own o1 model.
• Level 3: The rise of “agents” — AI systems that can autonomously carry out tasks without human input, such as Google’s rumored “Jarvis” or the AI systems developed by Anthropic, like Claude.
• Level 4: Innovators — AI models that can not only perform tasks but also generate new ideas, inventions, and creative solutions. This is where models begin to push the boundaries of what humans can do.
• Level 5: AGI — An AI capable of reasoning, learning, and performing tasks at the level of an entire organization. It would not only solve problems but also innovate, adapt, and operate independently.

While Altman believes AGI could be reached in the next few years, he also touched on the concept of Artificial Superintelligence (ASI)—AI that surpasses human intelligence in all domains and could potentially unlock the secrets of the universe. However, Altman was quick to point out that ASI is still “thousands of days away,” emphasizing that AGI itself is a monumental milestone on the road to more advanced AI systems.

As AI evolves, the line between AGI and ASI will become increasingly blurred, but the two concepts remain distinct in their capabilities. AGI is expected to function at a human level across a wide range of tasks, while ASI would go beyond that, achieving higher forms of cognition and insight.

Altman’s confident timeline for AGI could be influenced by OpenAI’s relationship with Microsoft. The current partnership between the two companies includes a deal that would conclude once AGI is officially achieved. This would require Microsoft to renegotiate terms and potentially pay more to access OpenAI’s models, raising the stakes for both companies. According to a New York Times report, the relationship between OpenAI and Microsoft is reportedly “fraying” due to this agreement, adding another layer of pressure to the race toward AGI.

While the technological leap to AGI may seem imminent, it is important to note that the transition will likely be gradual. Just as generative AI has slowly been integrated into a wide range of industries, AGI will likely emerge incrementally, improving over time until it becomes ubiquitous in daily life. Instead of a sudden revolution, AGI will develop steadily, incorporating new capabilities as they become available.

In fact, the integration of AGI could be more subtle than anticipated, with society adjusting slowly to the increasing presence of intelligent systems. The pace of AI development suggests that we may not even realize when we’ve crossed the threshold into AGI, as it will likely become part of the infrastructure that powers everything we do.

By Impact Lab