The Matrix is coming...

AI Summary

The video explores the limitations of current silicon-based AI, such as ChatGPT and Gemini, due to physical size constraints and high energy consumption of GPUs. It proposes a radical alternative: chips powered by living human brain cells. The presenter introduces Cortical Labs' CL1 chip, a significant advancement from their earlier DishBrain prototype (2021-2022), which used up to 1 million neurons to play Pong. The CL1 system, utilizing 200,000 human neurons, has been successfully taught to play the 3D video game Doom. The video details the CL1's sophisticated life-support systems, including a microfluidic perfusion circuit for nutrient delivery, a microscopic filtration system for waste removal, precise temperature control at 37°C, and an automated gas mixing system, enabling the neurons to survive for up to six months. These human brain cells are sourced from induced pluripotent stem cells (IPSCs) and are grown on a high-density microelectrode array (HDMEA) that facilitates two-way communication with the computer. The learning process is based on the free energy principle, where neurons are rewarded with predictable electrical signals for useful actions and punished with chaotic noise for bad decisions. The presenter highlights the key advantages of biological neurons: incredibly fast learning (hours versus millions of iterations for AI models) and extreme energy efficiency (the human brain runs on only 20 watts, compared to GPT-3's 1,300 megawatt-hours for training). Furthermore, biological systems naturally excel at handling unpredictable real-world environments, addressing the Moravec paradox and suggesting potential for robotics. The video concludes by addressing ethical concerns regarding consciousness and sentience, with Cortical Labs asserting the current CL1 chip is too small to be conscious, but acknowledging that scaling up could raise significant ethical dilemmas.

Claims Extracted (12)