The world of artificial intelligence is in a state of flux, with Yann LeCun, a leading figure in the field, advocating for a new approach to AI development. LeCun, who recently left his position at Meta to found Advanced Machine Intelligence Labs (AMI Labs), believes that current AI systems, such as ChatGPT, Claude, and Gemini, are limited in their ability to understand and interact with the real world. These models, he argues, are not built to handle the complexities and unpredictability of everyday situations, let alone perform tasks like household chores.
LeCun's vision for AMI Labs involves developing a more flexible AI, one that can create abstractions of the real world and assess the outcomes of actions. This approach, called Joint Embedding Predictive Architecture (JEPA), aims to address the shortcomings of current AI models by filtering out irrelevant information and focusing on useful representations of the world. While JEPA is still in development, it has already attracted significant interest from investors, including Nvidia and Jeff Bezos' private wealth fund, with over $1 billion in funding.
The limitations of current AI models, particularly in robotics, are also highlighted by Ingmar Posner, a professor of Applied Artificial Intelligence at Oxford University. Posner's team has been working on an alternative form of AI called World Models, which aim to answer questions like 'What matters?' and 'What causes what?'. This approach, inspired by a 2018 paper by David Ha and Jurgen Schmidhuber, has led to significant research in the field, including the development of the Dreamer World Model by Google.
Posner's 'mechanistic world model' aims to structure knowledge in a way that AI can efficiently recall, combine, and modify it when needed. While the timeline for developing these new models is uncertain, Posner notes that significant progress has been made in a relatively short period. Other organizations, such as DeepMind and Wayve, are also contributing to the development of World Models, and AI pioneer Fei-Fei Li has founded World Labs to explore a new AI model.
LeCun's and Posner's work reflects a broader shift in the AI industry towards more flexible and human-like intelligence. While current AI models excel at specific tasks, such as coding and generating text, they fall short in understanding and interacting with the real world. As LeCun suggests, the future of AI may involve humans and AI working together, with AI systems assisting humans in tasks that require creativity, problem-solving, and decision-making. This collaboration could lead to a more balanced and productive relationship between humans and AI, where AI augments human capabilities rather than replacing them.
