For centuries, as far back as Leonardo De Vinci with his construction of a primitive robot capable of banging a drum when hand cranked, people, authors, writers, and scientists, have been interested with the idea of Artificial, or Machine, Intelligence (AI). A being, not born or grown, but manufactured and built with the capability to interact with a “living being” as if it were one. A machine capable of existing, learning, responding, and adapting to its environment the same way in which humans do. So, how does this relate to neuroscience?, you ask. Well there are a number of ways the two can be connected. As people explore the possibility and the potential outcomes from creating AI, various theoretical and fictitious approaches have yielded a number of prototypes. Of these prototypes, many have taken an approach requiring the development of an artificial brain placed in the machine to control it’s movements and behaviors, much in the same way that our brains control our bodies and are responsible for our existence, for without them we are simply sacs of meat for lack of a better phrase. Given this idea for an artificial brain, we form the bridge between neuroscience and computers. How can we create an artificial brain when we have still yet to fully understand the workings of our own?
With advancements in neural mapping and science, both neuro- and computer, in the last ten to twenty years with the uses of new technologies like fMRI and EEG, scientists have come closer to understanding the functionality of the brain and all things associated with it. From studies like the one conducted by Professors Alan S. Cowen, Marvin Chun, and Brice A. Kuhl at the Universities of Berkley, Yale, and New York, where they have taken the neural activity recorded using an fMRI and have successfully used this information to create computer representations of people’s mental images of faces. This feat brings scientists closer to the understanding of the brain and brings them another step closer to the actual neural connections and firing patterns that enable humans to use this task. With this in mind, is it possible that in the future, once total brain mapping is successful and possible, that someones neural map could be copied at a certain point and time and transferred to this theoretical brain and function as if it were the person, even down to its neural plasticity when exposed to new experiences? This is a question has come up in some of the literature concerning the ethics around AI and is argued for and against still.
Going back a little to the theoretical model of AI, is this brain model absolutely necessary for this new breed of machine to truly function and live up to the standards that scientists have envisioned for AI? This may not be absolutely necessary as some computer scientists have developed a computer program and robotic arm that functions under the same perimeters and basis that we as humans do in learning new skills. At the the Max Planck Institute for Biological Cybernetics in Germany, scientists have developed a robotic arm that when shown a new skill and left to hone the skill will attempt the task a number of times before it first completes the task on it’s own and eventually masters it, coming closer and closer to completing the task with each attempt to execute it. Video examples of the arm at work demonstrate this and show the pattern that a child might undergo when they attempt a new task. In this video the arm attempts to catch the ball on the string in the cup and over- and under- shoots it at times taking 100 trials to make it in finally. This shows a kind of plasticity that could be argued would be seen in a human brain in the hippocampus as new connections are made. However no brain was necessary for this.
These are just a few of the pieces of research currently in existence that can be applied to AI, showing that this, though not in our lifetime per se, is likely to be a reality.
Cowen, A. S., Chun, M. M., & Kuhl, B. A. (2014). Neural portraits of perception: Reconstructing face images from evoked brain activity. NeuroImage, in press, 1-11.
Kober, J., & Peters, J., (2009). Learning Motor Primitives for Robotics. Robotics and Automation, 2112-2118.