What Is Artificial General Intelligence (AGI), And Is It Already Among Us?

In light of OpenAI’s recent advancements, particularly its reinforced collaboration with Microsoft and strides towards more sophisticated AI models, there’s growing interest and curiosity around Artificial General Intelligence. This growing interest sparks essential inquiries: What truly defines AGI, and are we on the brink of its real-world integration?

What Defines As Artificial General Intelligence?

Artificial General Intelligence, or AGI, stands as a frontier in artificial intelligence research, aiming to resemble human cognitive abilities in software. This concept is different from the more commonly encountered type of AI, which pertains to AI designed for specific tasks.

As envisaged by experts, AGI would theoretically possess the capability to tackle any task a human can, reflecting a level of adaptability and cognitive ability not yet seen in current AI models.

Blaise Agüera y Arcas, a Vice President at Google Research, elaborates on AGI as a system “capable of executing human-level tasks and abilities that no existing computer can achieve.” He points out that while today’s AI can perform many tasks, they do not yet match the breadth and adaptability of human intelligence. This gap between AI’s current capabilities and the lofty goals of AGI is substantial, marking AGI as a pinnacle yet to be reached in AI development.

The defining characteristics of AGI involve more than just task execution; they encompass abstract thinking, understanding cause and effect, and engaging in metacognition. This implies an AGI system would not only perform tasks but would also understand and adapt to them in a manner akin to human intelligence. For instance, an AGI system would excel in sensory perception, natural language understanding, and navigation, surpassing the limitations of current AI technologies.

Peter Norvig, a Distinguished Education Fellow at the Stanford Institute for Human-Centered AI, notes, “AGI should theoretically be able to perform any task that a human can and exhibit a range of intelligence in different areas without human intervention.” This comment highlights the envisioned autonomy of AGI, separating it from the more guided and constrained applications of current AI models.

Despite its theoretical nature, AGI remains a focal point of AI research, with companies like IBM and OpenAI pushing the boundaries of what’s achievable. However, experts caution against overestimating the current state of AI. As Norvig states, “Today’s frontier models are of course not fully qualified to be lawyers or doctors, even though they can pass those qualifying exams.” This is the gap between AGI’s theoretical framework and the practical realities of today’s AI capabilities.

Examples of AGI

The concept of Artificial General Intelligence remains a visionary goal. While true AGI systems do not yet exist, several examples in the current AI space hint at the possibilities and potential directions AGI might take.

Current AI Models and Their AGI-like Features

IBM’s Watson supercomputer exemplifies an AI system that approaches AGI in certain aspects. Watson combines immense computing power with AI capabilities to tackle complex scientific and engineering tasks. “Watson and other supercomputers are capable of calculations that the average computer can’t handle,” notes a researcher in AI. These systems, while not AGI, demonstrate a level of problem-solving and analytical capabilities that are steps towards AGI.

Legal and Medical Expert Systems

Legal expert systems, like ROSS Intelligence, and AI in medicine showcase AI’s ability to mimic human judgment in specific domains. They represent narrow AI’s evolution towards AGI. As these systems can analyse large amounts of data and provide precise responses, they serve as early indicators of AGI’s potential for decision-making in complex, real-world scenarios.

Self-driving Cars and Language Models

Self-driving cars are another example where AI excels in a particular domain, demonstrating advanced perception and decision-making skills. Similarly, language models like OpenAI’s GPT-3 and GPT-4 showcase an AI’s ability to generate human-like language, which some researchers argue is “strikingly close to human-level performance.” Sam Altman of ChatGPT remarks, “GPT-4 could reasonably be viewed as an early version of an AGI system.” These models, though not AGI, reflect the advancements towards achieving a more generalised form of intelligence.

Music AI and Its Creative Potential

Music AI, such as Dadabots, represents another facet of AI’s approach to AGI. These algorithms generate music that approximates human creativity, pushing the boundaries of AI’s capabilities in the arts. While these are not AGI systems, they reflect the creative potential that AGI could encompass.

The development of AGI is a complex and multifaceted one, with researchers exploring various approaches. Ben Goertzel, an AI researcher, explains that AGI development might involve “symbolic, emergentist, hybrid, and universalist” approaches, each contributing to the understanding of how general intelligence can be replicated in AI systems.

How Soon Will AGI Exist?

Elon Musk, the visionary behind xAI, projects a bold timeline for AGI. “I think we can make something better than DeepMind or OpenAI,” Musk stated in a Twitter Spaces session. He anticipates usable AGI emerging around 2029, revealing his confidence in rapid AI advancements.

Shane Legg, Co-founder of DeepMind Technologies, maintains his decade-old prediction of AI matching human intelligence by 2028. “You’ll never have a complete set of everything people can do,” he remarked on a podcast, highlighting the complexity of defining human intelligence. He advocates for significantly scaling up AI training for reaching AGI.

Geoffrey Hinton, a prominent AI researcher, has recently adjusted his forecast, now estimating AGI could be achieved within five to twenty years. “We live in very uncertain times,” he expressed on Twitter, acknowledging the unpredictability surrounding AGI development.

Yann LeCun, a Turing Award laureate, offers a more cautious perspective. “Will AI take over the world? No, this is a projection of human nature on machines,” he mentioned at a press event. He believes key concepts for AGI are still missing, suggesting a gradual progression to AGI.

Sam Altman, CEO of OpenAI, envisions AGI transforming lives. “We’ll be able to express ourselves in new creative ways,” he said, suggesting that AGI could materialise in the next decade.

Demis Hassabis, CEO of DeepMind, echoes this sentiment. “The progress in the last few years has been pretty incredible,” he told the Wall Street Journal, foreseeing human-level AI “maybe within a decade.”

The rate of technological advancements and ethical implications remain critical factors in determining AGI’s future. While the experts offer varying timelines, they converge on the importance of ethical considerations and technological breakthroughs in AGI’s journey.