Conference Presentation, Fireside Chat, Interview
Fei-Fei Li: Spatial Intelligence is the Next Frontier in AI
- AGI is expected to require the integration of spatial intelligence, defined as the ability for machines to perceive and understand the 3D world.
- The strategy involves creating world models to capture 3D structure to address the challenge of spatial intelligence, with use cases spanning creation, robotics, and learning.
- Historical biological timelines are cited, noting that language development took 300 to 500 million years while 3D world understanding required 540 million years.
- Future development relies on a hybrid data approach combining quantity and quality, alongside the convergence of data, GPUs, and neural networks.
- Hardware and software convergence is predicted to occur eventually to support metaverse applications.
- Company strategy involves a mix of open, closed, and hybrid models depending on the business model, with a belief that penalizing open source efforts would harm entrepreneurship and public sectors.
- Research focus is directed toward problems bordering on "delusional" to ensure fundamental inquiry distinct from industry capabilities.
- Academic institutions are viewed as suitable for individuals with deep curiosity to pursue problems unsolvable by industry's existing compute and data, whereas startups are driven by commercial goals.
- World Labs plans to hire engineering, product, 3D, and generative model talents who demonstrate intellectual fearlessness.
- Future AI aims to create machines capable of thinking and acting as intelligently as humans, distinguishing the real world as a continuum of generation and reconstruction rather than purely generated language.
- Risks include uncertainty stemming from the fact that human brain 3D perception is not yet a solved problem and that mathematical models for this capability are currently lacking.
- The approach involves focusing on building, disregarding past outcomes and external opinions to maintain comfort in execution.