Recent public interest in tools like ChatGPT has revived a long-standing question in the artificial intelligence community: is artificial general intelligence (AGI), AI that performs at human level, achievable? This week’s release of a preprint suggesting the latest advanced large language model, GPT-4, is exhibiting "sparks of intelligence" has only added to the hype surrounding AGI. OpenAI, the company behind ChatGPT, has openly declared its pursuit of AGI, while many researchers and intellectuals have called for an immediate halt to the development of these models, citing “profound risks to society and humanity”. While these calls for a pause in AI research are dramatic and unlikely to succeed, the allure and potential rewards of advanced intelligence are too tempting for humans to ignore. So, are the hopes and worries about AGI justified? How close is GPT-4, and AI in general, to reaching human general intelligence?
AI: A Landscape of Cognitive Capabilities
AI has been steadily conquering vast swaths of the cognitive landscape, surpassing human performance in tasks like vision, image recognition, reasoning, reading comprehension, and game playing. These AI capabilities have the potential to dramatically restructure the global labor market within a decade.
However, there are two main perspectives on the question of AGI. One perspective suggests that AI will eventually develop learning skills and capabilities that mirror those of humans, ultimately achieving AGI level. This view anticipates that AI will eventually replicate the uniquely human ability for continuous development, learning, and transferring learning across different domains. In contrast to current AI, where training in one area, such as cancer detection in medical images, does not transfer to other domains, AGI would possess the ability to generalize across different cognitive tasks.
This potential for AI to surpass human intelligence and rapidly leave us behind is a source of worry for many. The idea of AI exceeding human intelligence and eventually viewing humanity as something akin to how we see ants today is a sobering prospect.
The Limits of Imitation: The Debate Over Intelligence
Several philosophers and researchers challenge the plausibility of AGI, arguing that current models lack understanding of their outputs, essentially lacking awareness of what they are producing. These critics contend that consciousness is unattainable for AI models since they are primarily predictive, automating what comes next in text or other outputs. They emphasize that these models simply stitch together and reproduce data they have been trained on, lacking true intelligence and, more importantly, consciousness, the very essence of life.
While there is no guarantee that consciousness or AGI will emerge, even with continued advances in AI foundation models, the rapid performance improvement of each new model fuels the belief that AGI is right around the corner. The impressive ability of ChatGPT and GPT-4 to excel in tasks like bar exams and academic olympiads reinforces this perception.
Beyond AGI: A Future of Human-AI Collaboration
It is undeniable that AI can outperform humans in many specific cognitive tasks. However, there is mounting evidence suggesting that the optimal way to interact with AI may involve human-machine collaboration, where AI augments our intelligence rather than replacing it altogether. The emergence of work copilots and AI pair programmers for writing code highlights this trend.
The future of work, life, and learning seems destined to be profoundly and persistently shaped by AI. While the possibility of AI being considered intelligent is plausible, it remains a contested space, with figures like renowned linguist Noam Chomsky stating that the day of AGI “may come, but its dawn is not yet breaking”.
Embracing AI as a Cognitive Network
The second perspective on AGI shifts the focus from individual intelligence to the concept of intelligence as it’s practiced in the real world, through networks and systems. This perspective recognizes that humans are intelligent primarily in networks, sharing knowledge within communities. Until now, these networks have been primarily human-centric. While we might draw insights from someone like the author of a book, we don’t treat them as active "agents" in our cognition.
However, with the advent of ChatGPT, Copilot, Bard, and other AI-assisted tools, AI can become integral to our cognitive network. We engage with these tools, ask them questions, and leverage them to restructure documents and resources. In this context, sentience or general intelligence becomes less relevant. AI simply needs the capacity to become embedded within our knowledge network, augmenting and replacing many of the tasks we currently perform.
The Human-AI Synergy: A New Era of Creativity
The existential preoccupation with AGI often overshadows the many opportunities presented by current AI models and tools. Regardless of sentience or consciousness, these tools are already empowering individuals to co-create art, structure writings and essays, develop videos, and navigate life.
Ultimately, the most important question for humankind isn’t whether AI is intelligent when isolated from people. Instead, we should focus on the fact that AI is already augmenting our cognitive capabilities, making us more intelligent, capable, and creative. The future appears to be one of human-AI teaming, a journey that is already underway.
