The Open/Closed Problem in AI

The AI field is racing to build faster, more efficient hardware — but that race may be locking us into a dead end. Today's AI models learn through an "open loop": trained externally, then deployed frozen, never updating themselves. Meanwhile, specialized chips are being built around that assumption. True intelligence, like the human brain, learns in a "closed loop" — continuously, from the inside. By hardwiring today's flawed approach into silicon, we're making tomorrow's breakthrough harder to achieve.