In the early 2000s, a group of researchers led by Steve M. Potter, in the Laboratory for Neuroengineering at Georgia Tech, developed a robot that seemed like something out of dystopian science-fiction. It was a unique blend of living neurons from a rat brain with digital parts working together and came to be called a “hybrot” (hybrid robot).
A traditional microprocessor involves an array of transistors on a semiconductor chip constructed out of silicon, which functions like tiny electronic switches. These can be turned on or off based on the voltage applied to them. As opposed to these, in a hybrot, which is a hybrid of biological and artificial parts, the commands to the switches are provided by nearly 2000 living rat neurons.
The initial idea behind the construction of this robot was to open new vistas of research into mechanisms of learning. A simple system such as this could be exploited in understanding how neuronal circuits remould themselves based on input signals. But could this cyborgian system translate into something more that finds practical applications?
The advent of hybrid intelligence systems dates back to the late 1940s, much before the modern realization of hybrots. Neurophysiologist William Grey Walter, who discovered the brain waves we detect with electroencephalograms, was one of the pioneers to propose robots with living parts. Walter’s machines were named “tortoises” (inspired from Alice in Wonderland) and they could explore their environment and in particular, were drawn toward light sources, suggesting some form of sensory awareness. Surprisingly, these robots were constructed only out of two interconnected neurons, along with circuits of sensors and motors. While the developments in robotics today are diverse, these modest contraptions motivated strong views that simple neural circuits could initiate rich behaviours in artificial systems.
Artificial Intelligence and brain functioning are achieving new milestones in today’s world by using interdisciplinary research. Over the course of several decades, we have progressed to building intricate systems that combine the best of both worlds – “semi-organic” species. The stage is set to go beyond simulating neuronal networks in computer programs alone. Actual living cells can be interfaced with artificial bodies to act as independent brains. These hybrots can be used to explore intelligence and understand how AI can find manifestations biologically. Most of these follow a simple architecture with three main parts – biological neurons, hardware and software. The neurons (and glia, the other type of cells populating our nervous system) are isolated from brains of animals such as rats, sea lamprey or molluscs, then disintegrated, and re-cultured in the lab. In this process, they branch out and establish a full-fledged network again, outside a living body. This, in a way, pays ode to the concept of tabula rasa, wherein the mind acts as a clean slate where connections develop from scratch and lead to generation of new experiences.
An essential feature of these hybrots is two-way communication between the living and non-living components using an array of electrodes. The electrode array takes up the inputs from the neurons’ firing (bursts of electrical potential) and sends them over to be processed by certain algorithms. The outputs thus generated, help in controlling the other parts of the robot which can have mobile elements or sensors. In a full circle, environmental inputs to these sensors get converted to electrical signals and are supplied to the neurons as feedback. Intuitively, we can thus understand that the “processing software” must contain both coding (to convert sensory inputs to electrical signals) and decoding (convert neural firing patterns to outputs) programs. To imbue the spirit of AI into the entire design, the software can use special “learning” codes to train the neurons in performing well.
As we begin to wrap our minds around the enticing concept of hybrots, we should also appreciate their promising potential. Unravelling the mechanism of how we learn by physical changes in our neuronal connections is made possible by this novel idea. Tweaking the algorithms being employed can enable one to monitor the changes in the biological tissue in real-time and tease out the way in which it adapts to environmental influences. This in turn, can pave the way to a better grasp on how intelligence develops. Intelligence, crudely, can simply be thought of as the ability to achieve certain goals, based on our experiences. Fundamentally, neurons are wired to do exactly this, though locally. Each neuron has the capacity to receive, process and produce information. They have machinery to mould themselves to connect to other fellow neurons. Thus, the aim of “networking” is inherent. What matters is the right kind of networking. For this, appropriate environment factors are of the essence, which is essentially what “learning” achieves.
A hybrot is thus a useful addition to a neuroscientist’s toolkit to observe local neuronal circuit changes in tissue while also studying execution of complex behaviour through the actions of the attached robot. However, pressing challenges remain. Developing robust software to allow researchers to clearly map the hybrot activity to learning can advance the resource significantly. Another constraint is the single layer culture of neuronal cells used, which is not a true representation of the three-dimensional structure of actual brains. There is scope for improvements in all three components of hybrots and this can fuel basic research in making them efficient entities. Of course, the eccentric goal of achieving eternal life by using living human brains enmeshed in artificial bodies is a path probably best left untrodden.
REFERENCES:
1] Bakkum, D. J., Shkolnik, A. C., Ben-Ary, G., Gamblen, P., DeMarse, T. B., & Potter, S. M. (2004). Removing some ‘A’from AI: Embodied cultured networks. In Embodied artificial intelligence (pp. 130-145). Springer, Berlin, Heidelberg.
2] Holland, O. (1997, July). Grey Walter: the pioneer of real artificial life. In Proceedings of the 5th international workshop on artificial life (pp. 34-44). MIT Press, Cambridge.
3] Krausová, A., & Hazan, H. (2017). Robots with biological brains: autonomy and liability of a semi-artificial life form. The Lawyer Quarterly, 7(3).