New AI Design is Based on Microscopic Worms

As a way to decrease power consumptions, increase learning efficiency and transparency MIT’s start-up, Liquid AI has recently unveiled its new neural network, taking investment from companies like Samsung and Shopify.

The most fascinating fact about the innovation of new AI is the inspiration which is drawn from a millimeter-long worm found in rotting vegetation or moist soil. The worm has limited no. of neurons, say 100 which are mapped throughout the body making a highly complex system. Likewise, the liquid networks are also designed using few neurons which are capable of carrying out complex tasks. For example, a network with only 19 neurons and 253 synapsis can control a self-driving car.

In liquid neural network, each neuron is assigned an equation which defines its behavior rather than the static weights assigned in case of conventional neural networks. With time, the network as a whole matures by solving a cascade of linked equations allowing it to learn even after training. Another difference from the existing system is the associated transparency which also makes the inspection and research easy.

Currently, their new models can be used for only certain type of tasks which use temporal data such as detecting financial fraud, controlling self-driving cars and analysis of genetic data, but efforts are being carried out to make the AI more advanced and dynamic with time.