One of the greatest strengths of the Leela Video Intelligence Platform is the hybrid causal/neural network engine that powers it. Derived from research done at the MIT AI Lab on neurosymbolic network models, the technology continues to evolve at Leela AI led by Director of Research, Steve Kommrusch and CTO Henry Minsky. We will be revealing more details about our AI engine in the months to come. Meanwhile, here are two recent video presentations on the subject featuring Steve.
ISO/IEC has released a Zoom recording of the ISO/IEC Artificial Intelligence (AI) Workshop held on Nov. 30. Steve’s 18-minute talk on “Neurosymbolic learning on activity summarization of video data,” starts at two hours and one minute into the recording.
The ISO/IEC presentation explains how Leela’s neurosymbolic causal network layer can analyze video to understand concepts in context. The neurosymbolic layer supervises the core video analytics, which is performed by a more traditional machine learning neural network that focuses on individual images. The interplay between these different ways of “thinking” about a problem creates a self-correction loop.
When applied to manufacturing assembly applications, for example, the combination enables faster and more accurate identification of production problems and a deeper visibility into operations. In his talk, Steve describes how Leela’s neurosymbolic research will enable exciting new capabilities within Leela Platform.
The second video is an interview by Trent Fowler on the Futurati Podcast. In the 50-minute podcast, called “What’s the state of artificial intelligence (and should we be afraid?),” Steve summarizes the neurosymbolic AI research underway at Leela AI and places it in the context of emerging trends in AI. He explains how a hybrid neurosymbolic/neural network system could provide a promising springboard to achieve a common goal in the AI community: creating a constructivist learning system that might begin to reason like a human.
The podcast explores alternative approaches toward a constructivist AI emerging from academia and tech companies like Google and Facebook. These are primarily based on increasingly powerful transformer models that aim to generate higher levels of abstraction based solely on a neural net foundation. Finally, Steve and Trent discuss the meaning of sentience, the potential impact of sentient AI on humans, safety and ethical concerns, and the future of mind/machine interfaces such as Neuralink.