Biology Meets Silicon: Neural Networks of Non-Traditional Computing

Artificial intelligence has long imitated the brain only in mathematics—layer by layer, matrix by matrix, chains of linear algebra. Now researchers are beginning to copy it literally: using biological mechanisms in neural networks, building hardware that computes like a living cell. Non-traditional computing—molecular, optical, neuromorphic—is becoming the main contender for the architecture of strong AI.

When Silicon Hits a Wall

Modern GPUs are built on transistors that keep getting smaller, but not infinitely. Moore's Law is slowing down. Training large models requires monstrous energy: a single search on Perplexity consumes as much electricity as a car drives 5 km. Training GPT-4 cost hundreds of millions of dollars and released as much carbon into the atmosphere as a small city. AGI will require planetary-scale power—if built on traditional silicon. Engineers are running up against physics. Non-traditional computing promises a way out: instead of manipulating electrons in a microchip, you can use photons in optical systems (parallelism at the speed of light), DNA molecules (incredible information storage density), or electrochemical processes in neuromorphic chips (energy efficiency like a biological brain).

Neuromorphic Chips—Brain Architecture in Electronics

Intel Loihi, BrainScaleS, SpiNNaker—these are not GPUs simulating neurons in code. These are chips physically structured like a brain: spiking neurons, synapses with adjustable weights, asynchronous event-based processing. Instead of matrix multiplication of all weights across a batch—only active neurons send impulses when needed. The result is striking: energy consumption is a thousand times lower than GPUs of the same scale. The human brain at 20 watts does what modern neural networks require 20 kilowatts to do. A neuromorphic chip replicates this result.

• Real-time learning (without waiting for a full batch of data)
• Stream processing instead of matrices (ideal for sensors and cameras)
• Adaptation to new tasks without re-optimizing the entire model
• Robustness to noisy inputs thanks to biological pulse logic

Companies like Intel and academic laboratories are already using neuromorphic chips in robotics and sensor data processing. Scaling is more difficult, but the potential is enormous.

Molecular Machine, Test Tube of Computation

DNA is not just a repository of genes. It is a parallel computer. A single molecule can encode gigabits of information and perform computations simultaneously with all molecules in solution. Researchers at MIT, Caltech, and other laboratories are already using DNA computing for machine learning tasks: classification, matrix multiplication. It's harder to scale than with chips—you need methods to control billions of molecules synchronously, extract results without errors. But if it works—the density of computation will exceed any silicon analog, and energy consumption will drop several times over. In parallel, researchers are investigating optical computers (photons instead of electrons) and quantum systems. The picture is clear: the monoculture of silicon GPUs is fading. AGI will be a hybrid.

What This Means

The boundary between biology and silicon is blurring. Non-traditional computing is not exotic laboratory science, but an engineering workshop for strong AI. In a year or two, we will see the first hybrid systems: neuromorphic accelerators in data centers next to GPUs, molecular computers for specialized tasks, optical processors for linear algebra. When AGI arrives—if it arrives at all—its hardware will be 90% composed of what living nature proved effective billions of years ago.