Neurons as Artificial Control Circuits
When you usually envisage brain-machine interfaces, it is a picture of a human or primate skull being opened up to reveal a mass of brain tissue: billions of neurons firing in complex patterns, which gives rise to sentient, intelligent life. Into this fully functional organic computer, an array of electrodes is sent. A square grid of metal contacts pressed into its surface, reaching out and detecting signals, embedding deeper and deeper into the flesh.
That is perhaps the archetypal image of BMI. However, aside from that, and also aside from the network of electrodes placed upon the scalp covering such an organic computer, which also function in the same broad way, there is another approach to BMI, fundamentally different. In fact, the approach is diametrically opposing to the archetype of integrating with a functioning brain.
In that paradigm, the electrode arrays and the silicate computers they connect to were there, first. Organic tissue is grown on the electrodes, neuron binding to neuron as a graft atop an existing computer system.
In this way, the organic brain that forms actually becomes slaved to the computer, rather than the other way around.
Welcome to the world of animats, animal-robot hybrids, or machines driven by organic tissue.
The field of animat creation is almost as old as BMI itself. The first such having been created in the 1970s when scientists cultured heart cells on laboratory glassware embedded with electrodes that could record the cells' electrical signals.
Culturing a few heart cells is very different to growing hundreds of thousands of neurons on a single chip, and linking them into two-way traffic on a control system, yet, just 40 years on, that is the stage we are at now.
The first such breakthrough came in 2006, when scientists from the University of Florida were able to successfully grow 25,000 rat neurons in a specially designed petridish, that linked the cells up to a computer via 60 electrodes embedded in its lower surface. The cells were able to grow in the ideal environment preserved inside the dish, and at the same time, interface optimally with the electrodes, bonding to them and treating them essentially as the brain's brainstem, feeding on the sensory input that came through.
On the other end of the electrode array sat a computer, running a modified flight simulator program. The electrode input and output was linked to a simulated F-22 fighter jet.
At first, the jet crashed all the time, with the neurons having no clue what to do. But, over time, like a baby learning to walk, the scientists noticed an improvement in the behaviour of the jet. It began to tend towards a straight and level course, high above any obstructions. They then added weather simulation effects, throwing rain, hurricane wind, even tornadoes at the plane. Again, each time the neurons encountered a new hazard, they performed badly. However, again, over time they performed better, learning what to do in each case of sensory input. Also in each case, they did not do as badly for so long, when pitted against other weather hazards, once they mastered the first few. They were learning, and applying knowledge to other tasks.
By 2008, just two years later, the scene has changed drastically. The petridish is gone, and in its place, what might be described as a gel-pack; a small pot containing a pink broth of nutrients and antibiotics.
Inside those sealed pots, connected to the outside world by anywhere from 80 to 120 electrodes, anything up to 300,000 rat neurons live. The basic paradigm is still the same, of course; the electrodes are in place first and the cells are cultured around them.
Some cell gel-packs still control simulated aircraft. Others control virtual avatars. Still others are appearing as control circuitry for lab robots, wandering around, trying to avoid walls, and learning each time they do so.
The analysis capability has grown with the brain size, and several efforts are keenly monitoring the firing patterns of the neurons inside the pots, analysing how they strengthen over time, or how one strong pattern might be discarded in favour of a weaker one. There is also interest in how with these larger brains, compartmentalisation and speciality starts to occur.
Some of the interest is directed at making better animats, of course. A lot of the rest is directed at understanding how organic neural nets form and operate. They have similarities to human and animal brains, even if they are not the same, and insights gained from the study of these 'test tube brains' has real implication for our study of the workings of our own.
References & Further Reading