Daniel Lütgehetmann presents a brain-inspired AI approach called 'digital brains' — biologically accurate simulations of neurons and synapses built over 20 years by neuroscientist Henry Markram's team. Unlike deep learning, these digital brains use spiking neurons with complex geometric structures, event-based communication, and local synapse-level learning (no backpropagation). The system learns 30–50x faster than conventional AI in dynamic interactive environments, demonstrated by mastering Pong in 11 games versus hundreds for deep learning. Current applications include time-series forecasting (partnered with Microsoft) and robotics on factory floors. The architecture naturally merges multimodal sensor inputs and scales to a million neurons — more than an ant — suggesting potential for swarm behavior and autonomous navigation without massive data centers.
Sort: