This Startup Fused Human Brain Cells with Silicon Chips | E2295

Episode Summary

Executive Summary: The episode centers on Cortical Labs’ effort to commercialize biological computing by fusing human neurons with silicon chips. The founders discuss the CL1 hardware, cloud access, real-world deployments at U.S. research institutions, early success in reinforcement learning, and the company’s ethical guardrails around consciousness. The interview then shifts to Pika’s autonomous large drones for agriculture and cargo, highlighting regulatory hurdles, vertical integration, and dual-use logistics applications.

Main Topics: Biological computing with Cortical Labs (Priority: 5/5): Alex and Han discuss the CL1, a rack-mounted biological computer that keeps neurons alive and interfaces them with chips, positioning it as a practical platform for researchers and developers. Commercial rollout and biological data centers (Priority: 5/5): Cortical Labs has moved from prototype to deployment, with CL1 units sold out, U.S. institutions using the hardware, and a Melbourne-based 'biological data center' plus planned Singapore expansion. Performance claims and reinforcement learning (Priority: 5/5): Han says biological neurons outperformed GPU-based reinforcement learning systems by roughly 5,000x in stability/efficiency for a specific task, suggesting possible advantages for embodied AI and robotics. Ethics, consciousness, and religious concerns (Priority: 4/5): The conversation addresses fears about 'tinkering with humanity,' Vatican concerns, and the company’s red line against creating conscious systems because consciousness implies the capacity to suffer. Cloud access, SDKs, and developer adoption (Priority: 4/5): Cortical Labs is pushing cloud access, Python/Jupyter-based tooling, and hackathons so developers can experiment without owning a wet lab or maintaining neurons themselves. Pika’s autonomous drones for agriculture and cargo (Priority: 5/5): The second interview covers Pika’s electric and hybrid UAVs, their use in Brazilian agriculture, cargo transport, and contested logistics, plus the regulatory and supply-chain realities of scaling hardware.

Key Arguments: Biological systems can provide generalized, goal-directed behavior that current machines struggle to match in some contexts, especially reinforcement learning. Cortical Labs is reducing barriers by offering biological computing as cloud-accessible infrastructure rather than only selling expensive standalone devices. The company believes it can pursue commercially useful biology without crossing into conscious systems, which it treats as an ethical boundary. Placing lab and compute in the same data center (as in Singapore) improves self-sufficiency, reduces shipping/supply-chain friction, and lowers operational complexity. Pika argues that vertically integrated hardware/software systems are necessary for reliable autonomy and customer trust, especially in harsh real-world environments. Large autonomous drones are commercially viable where regulations allow broader beyond-visual-line-of-sight operations, but U.S. rules remain a bottleneck. Building hardware in-house increases development time, but can produce tighter integration and better long-term product quality than assembling off-the-shelf components.

Data Points: CL1 price: ~$35,000 per unit - First run of Cortical Labs’ biological computer Initial stock sold: 30 units - Cortical Labs says it exhausted its first kept stock Approximate first-run revenue: ~$1 million - Derived in conversation from 30 units at about $35,000 each U.S. institutions with CL1: 5 - Johns Hopkins, Mass General, UCSF, Dartmouth, and one earlier PDO recipient CL1 neuron capacity: Up to 1–2 million neurons - Maximum discussed for a CL1 device Cloud neuron count: About 200,000 neurons - Commercial cortical cloud offering uses fewer neurons for viability Data center footprint: 6 racks / about 120 units - Melbourne biological data center Singapore capacity target: Up to 1,000 CL1 units - Planned expansion with Day One data center partnership Power use per unit: About 30 watts - Energy consumption of one CL1 device Biology replacement interval: 4–6 months - Tube/filtration sets need replacement; neurons can live longer if maintained Reinforcement learning advantage: 5,000x more stable/efficient - Compared neurons against GPU-based reinforcement learning benchmarks Pika first flight timeline: 11 weeks - Initial large autonomous aircraft flew a week before Y Combinator demo day Pika cargo plane development: 180 days to first flight - Dropship went from CAD to first flight in about six months Aircraft operating range: 5–15 km in Brazil - Typical Pika agricultural operations from launch point U.S. limit for current ag drone: About 4 km - Commercial approval constraint from FAA due to line-of-sight rules Fuel consumption comparison: 55 gallons/hour vs ~2 gallons/hour - Air Tractor versus Pika aircraft when powered by a diesel generator Training time: 2–4 weeks - Time to train a Pelican operator versus 18 months for an aerial application pilot Payback period: 2–3 years - Rough ROI timeframe for Pika’s ag aircraft depending on utilization US farm size threshold: ~20,000 acres - Approximate size needed to fully utilize the aircraft in the U.S. Dropship performance target: 1,000 miles range / 500 pounds payload - Stated customer requirement that the new aircraft meets Dropship power architecture: ~30 kW diesel peak; 2 x 25 kW electric motors - Hybrid propulsion details described for ballistic takeoff and cruise Battery management sourcing delta: 2x cost difference - U.S. manufacturing of NDAA-compliant BMS versus China sourcing for commercial product Pika customer uptime sensitivity: 24 hours downtime is problematic - Commercial and remote customers expect very high reliability

Pivotal Quotes: "We have accomplished super intelligence... Is it generally intelligent? No." — Han: Clarifying the distinction between current AI capability and true generalized intelligence "You do not want to create conscious systems because ethically, a conscious system has the ability to suffer." — Han: Explaining Cortical Labs’ ethical red line for biological computing "Their benchmarks... the neurons we had were 5,000 times more... efficient than their GPU-based systems." — Han: Describing the reinforcement-learning comparison that stood out as a breakthrough

Implications: Biological computing may become a new infrastructure layer for AI and research if safety, tooling, and ethics scale alongside performance. Meanwhile, autonomous aviation could accelerate in agriculture and logistics—but regulation and supply chains will largely determine how fast it reaches mass adoption.