- LeCun believes human-level AI is decades away, with more immediate progress towards simpler, animal-like AI intelligence.
- Emphasizes the need for AI to process diverse data types, moving beyond text-based learning to include audio, images, and video.
- Both LeCun and Schroepfer express doubts about the near-term practicality and relevance of quantum computing in advancing AI.
December 5, 2023: At a recent event marking the 10-year anniversary of Meta’s Fundamental AI Research team, chief AI scientist Yann LeCun shared insights that offer a different perspective on the future of artificial intelligence and quantum computing.
LeCun, a deep learning pioneer, argued that current AI systems are far from achieving human-level intelligence, predicting that AI will reach “cat-level” or “dog-level” intelligence long before it approaches human capabilities.
He emphasized that modern AI, largely trained on text, lacks the common sense needed to understand basic relational concepts that humans find intuitive. LeCun’s view contrasts sharply with that of Nvidia CEO Jensen Huang, who believes AI will rival human intelligence in less than five years.
LeCun also highlighted the limitations of text as a training tool for AI, pointing out the need for more diverse data inputs.
As part of this vision, Meta is exploring multimodal AI systems that integrate audio, image, and video data. This approach could lead to more advanced applications, such as AR-enhanced tennis training using Project Aria glasses.
The discussion also touched on AI hardware. While acknowledging Nvidia’s current dominance in AI hardware with its GPUs, LeCun anticipates the emergence of specialized deep learning accelerators.
He sees these future chips as potentially more effective than GPUs for AI applications.
On the topic of quantum computing, LeCun expressed skepticism, challenging the practical relevance of quantum computers shortly. He pointed out that many problems assumed to be solvable by quantum computing could be more efficiently handled by classical computers.
Meta senior fellow Mike Schroepfer echoed this sentiment, considering quantum technology still too distant to impact current AI developments.
