Siraj Raval

This Computer Is Made of Real Human Neurons (I Programmed It)

The video introduces Cortical Labs' commercial biocomputer that houses 800,000 living human neurons on a multi‑electrode silicon chip, accessible through a Python SDK and a cloud‑based “wetware‑as‑a‑service” platform. The system forms a closed‑loop where electrodes record spikes and deliver stimulation, allowing developers to train the culture to perform tasks such as Pong or Doom. Energy consumption is striking—roughly the power of an LED—far lower than comparable silicon AI accelerators. Pricing options include a $300‑per‑week cloud rental or a $35,000 on‑premise unit, with cultures remaining viable for up to six months. The presenter demonstrates a simple program that records baseline activity, detects spikes, and triggers stimulation on another channel, causing the neurons to reorganize their firing patterns. He cites Brett Kagan’s 2022 Dishbrain project as the first closed‑loop demonstration and emphasizes that the SDK abstracts hardware details, offering a simulator for developers without physical access. The emergence of a programmable, energy‑efficient biological substrate could reshape AI compute economics if it scales, while the lack of consciousness in the cultures does not eliminate ethical concerns. Early adopters can experiment today, but regulators and the broader tech community must address governance before commercial deployment expands.