In an age where we can construct complex electronic chips with feature sizes in the single-nanometer range, it’s easy to forget that we still don’t know everything about how the human brain works. The exact nature of the brain’s signal transfers to other body parts via the central nervous system is a code we have yet to crack. But artificial-intelligence researchers are on the case.
A U.S. research project at Brown University and Intel that kicked off last month aims to reconnect parts of the spine in patients with spinal cord injuries. Using advanced implantable-electrical–interface technology, the researchers will listen to signals sent down the spinal cord from the brain and try to decode them before relaying them back into the spine at a point below the patient’s injury.
These signals are electrical and can be directly measured as voltages. Typically, different frequency bands in the signal represent different neuronal processes. But the researchers won’t know exactly which neurons they are listening to. The contacts on the interface are around a millimeter square, while each neuron is around 20 microns. The signals received from each electrode will therefore be the superimposed signals from hundreds of thousands of neurons.
AI is the perfect tool to help decode these signals, because the data is extremely complex and there is little obvious correlation between the physical signals recorded and the response in the muscles. The researchers will use a specially designed neural network with a “biologically inspired” architecture in a bid to understand the data sufficiently to produce signals that will stimulate the spine in the right way. The eventual aim is to enable lower-limb movement and bladder control in spinal-injury patients.
The neural network will also handle mapping between the two electrode arrays, another complex yet critical task. It’s unlikely, given the way the system will directly stimulate many neurons at a time, that direct mapping from neuron to neuron will be achieved.
But here, the project will use a crucial function of the nervous system: its own ability to learn. Neuronal pathways in the brain are formed and reinforced through repetition (“neurons that fire together, wire together”), so the hope is that the nervous system will learn to interpret the team’s imperfect signals correctly over time.
In other research findings published last month, French biomedical research center Clinatec reported using AI to decode brain activity via a minimally invasive wireless device implanted in a tetraplegic patient’s sensorimotor cortex. Electrocorticograms from the implant are decoded in real time using highly sophisticated AI models that predict the deliberate movement the patient imagines. The information controls a proof-of-concept exoskeleton that lets the patient move all four limbs and even take several successive steps.
AI is the perfect tool to help us decode the brain’s signals, and the results of these two projects could change the lives of spinal-injury patients. Projects like these illustrate AI’s promise as a source for good. Rather than fear the implications of advanced AI, we should work toward its responsible use, harnessing its potential to help solve some of humanity’s most complex problems. ■