Microsoft doubles down on smaller generative AI; AI's next debate is over whose values it should hold; AI will transform life in emerging economies; a Chinese startup is winning the open source race
computerspeak.substack.com
In
a recent paper, researchers from the Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) described a new approach that could change the game even further for the energy efficiency of AI models: spiking neural networks (SNNs). The human brain is a marvel of efficiency, capable of complex calculations and learning with minimal energy consumption. But our current crop of large models that power chatbots or generate images and videos require vast amounts of energy to function. One query to ChatGPT uses the equivalent of 3.96 watts, or a third of the battery capacity of an iPhone 13 Pro. Meanwhile, an adult human brain consumes slightly more than a third of a watt of energy in an entire day day, or approximately 8% the energy of one question posed to a large language model.
Inspired by the brain's biological structure, SNNs process information using discrete "spikes" of electrical activity, similar to how neurons communicate. This is a stark contrast to traditional artificial neural networks, which rely on continuous calculations.
Illustration showing the working flow of SNNs
This shift has the potential to dramatically reduce the energy footprint of AI, making it more sustainable and paving the way for exciting new applications including:
- Long-range search and rescue: Imagine drones powered by SNNs, able to navigate disaster zones and locate survivors for extended periods without draining their batteries.
- Prosthetics: SNNs could lead to more intuitive and natural-feeling prosthetics that better integrate with the human nervous system.
- Edge computing: SNNs' low-power requirements make them ideal for running AI tasks on devices with limited resources, like smartphones and wearables.
While SNNs are still in their early stages of development, the potential rewards are vast. By emulating the brain's efficiency, we could unlock a new era of AI that is not only powerful but also sustainable.