Computing the Impossible #TechVision Digital Ambition, to transmit information in a way that is more on the battery but on what can be done with Inspired by Nature power-efficient than traditional CPUs. Also, this the system. A library of 100 words for natural architecture is optimized for the execution of language processing, for instance, is going to While HPC may be more familiar, there’s Spiking Neural Networks (SNNs), a different be a lot less computationally intense and power another class of technology reshaping what approach to neural networks than the Artificial hungry than a library of 2,000 words – which businesses can do. Biology-inspired compute Neural Networks (ANNs) that power today’s AI means power considerations will directly affect takes advantage of the most mature systems systems. The SNN leverages simulated neurons uses like human-to-machine interactions. in the world: nature. There are two subdivisions to transmit input and output data, while an to this class: biomimicry, or systems that draw artificial synaptic layer strengthens (or weakens) Another option is for AI processes to run in the inspiration from biological processes, and the connections between each neuron – cloud, but then engineers run into a different bio-compute, which are systems that directly allowing the system to learn very similarly to the set of limitations around bandwidth and latency. utilize biological processes to perform way the human brain operates. No one wants a drone or a car that makes a computational functions. decision half a second too late. This is where Stepping back to what these machines actually neuromorphic computing provides a clear Biomimicry has been used in areas ranging from let us do – consider robotics. Currently, to advantage – it can run AI systems that allow for chip architectures to learning algorithms, and design autonomous or semi-autonomous learning, more natural interaction, and more, in a successful pilots have shown this emergent robotics, engineers must decide where to put power-efficient way. It opens the door to a world field can deliver benefits like greater power the intelligence. The machines need to be able of robotics and edge computing that we can see efficiency, speed, and accuracy in more to execute a set of instructions, but also adapt, from afar but have yet to attain. complex problems. For instance, one technology react, and learn about their environment. One at the forefront of biomimicry is neuromorphic option is to put AI models at the edge, but then computing. Neuromorphic chips, like Intel’s the algorithm likely needs to run on extremely Loihi, have introduced a brand-new design to power-intensive GPUs. With the current computer chips: They are modeled after the limitations of batteries, power consumption 178 becomes a significant design challenge, not just human brain. The chips use artificial neurons Introduction // WebMe // Programmable World // The Unreal // Computing the Impossible 80