Researchers have built a robot dog with a hybrid “brain”, made of silicon and human neurons grown in the laboratory. The future goal is to build machines capable of adapting to the unpredictable and continuously changing real environment, thanks to the biocomputational interface.
Scientists from the Australian company Cortical Labs have grown gods human neurons grown in the laboratory on chip of a robot dogwhich through lowercase letters electrodes (MEA) were exploited to make it move on a grid. Essentially, they used a biological neural network and alive as a computational element of the robot, connecting it to traditional IT systems. The complex approach, currently rather slow, was only used to make the robot perform simple actions, such as searching for a target within a defined area. However, the ambitious future goal, for the authors of the research, is to equip robots and autonomous systems of a “brain” hybrid based on human nerve cells And silicon to make them make decisions in real and unpredictable environments.
If in fact from the point of view of physical control robots have now achieved extraordinary capabilities, when faced with unpredictable conditions they have a certain difficulty adapt and make quick decisions. Just look at the behavior of some vehicles autonomously controlled, which launch themselves onto the fresh asphalt during work in progress or they block traffic because their AI-shaped sensors did not predict certain conditions. Equipping them with human cells would not only significantly improve theenergy efficiencybearing in mind that our brain needs much less energy than classic robot power systems, but it would also offer an edge in terms ofadaptability. We are much better than robots at adapting to situations that change, continuously and suddenly; researchers want to apply that ability to robotics.
Apparently we are faced with a revolutionary technological development for machines, thinking for example of their use in rescued. Imagine one of these robots in a burning or collapsed building, capable of tracking down and saving human lives in danger thanks to its highly sophisticated hybrid brain. However, we live in a historical moment in which AI agents continue to escape the control of their creators, furthermore, war drones commanded by artificial intelligence are already used on the battlefield. A 19 year old Ukrainian girl she would have been killed by a drone programmed to autonomously search for its targets. Now imagine these chips with human neurons applied to tools designed for kill: the idea of having armies of machines with such features it doesn’t seem that attractive. Rather, it seems like the incipit of a dystopian future that resembles the plot of Terminators. Without forgetting that the so-called Doomsday Clock continues to remind us that AI could be a significant threat to human existence.
But let’s get back to the experiment. Australian researchers at Cortical Labs, working closely with colleagues at the Honda Research Institute Europe in Germany, used these lab-grown neurons to create a platform Of processing capable of learning and solving practical tasks, with significantly lower consumption than computers based exclusively on silicon. Networks based on neuronal cultures (26 were used) were inserted into a virtual environment to make an agent move within a simple 6×6 grid. The stimulus was the smell of food. The neurons received the odor information and could make the agent move forward or turn left or right. It seems like a simple process for advanced AI systems, but it took 4,000 hours of work and over 1,000 configurations to transform the electrical stimulation in movements and interactions.
The neurons learned their task very well and the researchers transferred everything into the “brain” of a robot dog, which performed its task correctly as in the virtual environment. “The robot dog is a simple demonstration of how progress in biological systems can also be applied to the real world, and helps make this opportunity tangible,” said Dr Brett KaganCEO and co-founder of Cortical Labs. The study shows that the human neural networks they are able to learn a navigation task if connected to a virtual environment via a specific device bioelectronic interface. These are the first steps towards a platform biocomputational which could offer enormous benefits compared to traditional systems, but also monstrous consequences if transferred to war machines. Details of the research “Embodied Neurocomputation: A Framework for Interfacing Biological Neural Cultures with Scaled Task-Driven Validation” have been uploaded to ArXiv.
