Leading AI research organization, Google DeepMind, has unveiled its next-level artificial general intelligence (AGI) and clarified its definition in a recent paper. AGI refers to artificial intelligence that matches or outperforms human capabilities in a variety of tasks. However, the specifics of what constitutes human-like performance and the range of tasks have often been overlooked. DeepMind’s team has now constructed a comprehensive definition by identifying the essential common features from existing definitions of AGI.
To provide further clarity, the team has outlined five ascending levels of AGI: emerging, competent, expert, virtuoso, and superhuman. Currently, only emerging AGI has been achieved, which includes cutting-edge chatbots like ChatGPT and Bard. The higher levels of AGI encompass abilities such as decoding thoughts, predicting future events, and even communicating with animals.
The decision to refine the definition and establish these levels stems from the increasing significance of AGI. With even top-level discussions involving AGI, it is crucial to have a clear understanding of the term and its implications. This marks a significant departure from the earlier days when discussions surrounding AGI were dismissed as vague or rooted in fantasy.
Shane Legg, co-founder of DeepMind and now the company’s chief AGI scientist, coined the term AGI around 20 years ago. He recalls that, at the time, AGI was regarded less as a concise definition and more as a field of study. However, as AGI gains prominence, a sharper definition is needed to avoid confusion and establish a common understanding.
By publishing their paper, DeepMind’s researchers hope to bring some much-needed clarity to the topic. Julian Togelius, an AI researcher at New York University, appreciates their efforts in eliminating the ambiguity surrounding AGI. He notes that too many people casually use the term without giving much thought to its meaning.
DeepMind’s definitions and levels of AGI are a crucial step forward in the field of AI, particularly as discussions about AGI become more common. These clarifications will help researchers and practitioners navigate the landscape with a shared understanding. As AGI continues to progress, it is essential to refine its definition and avoid any misconceptions.
With the ongoing hype around generative models and the surge of interest in AGI, Google DeepMind’s efforts to outline and define AGI levels provide a solid foundation for further research and development. The paper serves as a valuable resource for anyone interested in understanding AGI and its implications for the future of artificial intelligence.
