Robots with living neurons

Walking the slippery line between bio-enriched robots and cyborgs, the work of various teams of scientists around the world is advancing faster than moral judgments and legislation.

Zhiqiang Yu’s team, from the Beijing Institute of Technology in China, have completed a review of results from studies on the use of in vitro-created biological neural networks for use in robots and other artificial systems.

Yu and his colleagues reviewed the fundamentals of intelligence in in vitro biological neural networks, such as memory and learning; how these biological neural networks can be installed in robots through bidirectional connection, forming the so-called neurorobotic systems based on biological neural networks; what preliminary intelligent behaviors these neurorobotic systems exhibit; and what current trends and future challenges face neurorobotic systems based on biological neural networks.

Our human brain is a complex biological neural network made up of billions of neurons, which gives rise not only to our intelligence but also to our consciousness. However, studying the brain as a whole is extremely difficult due to its intricate nature. By growing a portion of the brain cells in a Petri dish, simpler biological neural networks such as minibrains (brain organoids) can be formed, making it easier to observe and investigate such networks. These mini-brains could provide valuable clues to the enigmatic origins of natural consciousness and intelligence, Yu argues.

Interestingly, mini-brains are not only similar in structure to human brains, but can also learn and memorize information in a similar way, Yu points out. Specifically, these in vitro created biological neural networks share the same basic structure as biological neural networks normally formed within the framework of the natural development of living beings. In both cases, the neurons are connected by synapses and exhibit short-term memory. Furthermore, these mini-brains can perform supervised learning and be trained to respond to cues from specific stimuli. Recently, it has been shown that biological neural networks created in vitro can even perform unsupervised learning tasks, including separating mixed arriving signals.

These abilities of in vitro created biological neural networks are quite intriguing. However, it is not enough to cultivate such a mini-brain for consciousness and intelligence to arise. Our brain depends on our body to perceive, understand and adapt to the outside world, and in the same way, these mini-brains need a body to interact with their environment. A robot is an ideal candidate for this purpose, which has given rise to a flourishing interdisciplinary field at the intersection of neuroscience and robotics: that of neurorobotic systems based on biological neural networks.

A clear conclusion, as the authors of the study results review point out, is that a stable two-way connection is a prerequisite for these systems.

An artist’s recreation of the concept of a robot equipped with living neurons. (Illustration: Jorge Munnshe for NCYT from Amazings)

Yu and his colleagues have seen that the intelligent behaviors displayed by neurorobotic systems based on biological neural networks can be divided into two categories based on their dependence on computational power or on the plasticity of the network. In computationally dependent behaviors, learning is unnecessary, and the biological neural network simply operates as a data processor that generates specific neural activities in response to stimuli. However, for the second category, learning is a crucial process, since the biological neural network adapts to the stimuli and these changes are essential for the behaviors that the robot will later have or the way in which it will later perform certain tasks.

Yu’s team has determined that one of the main challenges to overcome is to be able to fabricate biological neural networks in 3D, thus making in vitro biological neural networks much more similar to their natural counterparts.

Perhaps the most difficult aspect is how to train these biological neural networks embedded within robots, especially if the bodies of these are very different from those of animals whose brain cells have been cultured.

And, of course, there is the main challenge of all: unraveling the mystery of how consciousness and intelligence arise from the network of cells in our brain. This phenomenon is still beyond the understanding of science. As Yu points out, when robots equipped with biological neural networks begin to abound, it will be possible to study the behaviors of these entities (robots or cyborgs) and detect if there are key similarities with the behaviors of intelligent living beings. Perhaps this leads to ethical dilemmas that are difficult to resolve.

The review of study results is titled “An Overview of In Vitro Biological Neural Networks for Robot Intelligence.” And it was published in the academic journal Cyborg and Bionic Systems. (Fountain: NCYT by Amazings)