Douglas J. Bakkum, 
"Dynamics of Embodied Dissociated Cortical Cultures for the Control of Hybrid Biological Robots" 
2007 (Georgia Tech PhD dissertation)

SUMMARY

One currency of the brain is action potentials. They relay sensations of the world into commands for muscles to move the body and produce speech, and their manipulation is responsible for adaptive behavior and cognition. However, much remains unknown about the fundamental rules governing such neural information processing in the brain. By growing networks of cortical neurons and glia over multi-electrode arrays (MEA), which can be used to both stimulate and record multiple neurons in parallel over durations up to months, a 2-way communication with neuronal network activity becomes feasible. In particular, I was interested in embodying these networks with robotics to study the importance of environmental interaction, or behavioral feedback, in neural processing. Here, the recorded activity of the neurons was transformed into movements in an artificial environment, and sensory feedback was transformed into electrical stimulation on multiple electrodes. Stimulation influences neural activity and in turn the subsequent movements, creating a closed-loop system we call a neurally-controlled animat.
My ultimate goal was to develop animats that could learn something about the environment and/or body given to them. Upon entering the lab, the technology to culture neurons for long durations on MEAs, robustly record neural activity, and stimulate an MEA’s electrodes was recently achieved. However, the crucial ability to induce and detect neural plasticity was missing: methods were needed to determine appropriate sensory-motor mappings and training algorithms in order to produce any kind of adaptive behavior. I took a step back to first determine, in open-loop experiments, what types of stimulation could induce plasticity and what kinds of activity statistics could identify plasticity. This knowledge was then applied to construct embodied systems.
To paraphrase the results, most any stimulation could induce neural plasticity, while the inclusion of temporal and/or spatial information about neural activity was needed to identify plasticity. Following a tangent from one of the open-loop experiments, the plasticity of action potential propagation was observed. This is a notion counter to the dominant theories of neuronal plasticity that focus on synaptic efficacies and is suggestive of a vast and novel computational mechanism for learning and memory in the brain.
The results from the open-loop experiments were next used to develop animats that achieved adaptive goal-directed behavior. The feedback of patterned training stimuli, contingent on behavioral performance, was found able to sculpt the network activity into desired states. Network plasticity was not just induced, but could be customized, suggesting a potential role for the rehabilitation of neural pathologies. Furthermore, in collaboration with artists from SymbioticA at the University of Western Australia, neurons were embodied with a robotic drawing machine and exhibited at galleries throughout the world. This provided a platform to educate the public and initiate critical discussions of biotechnology.