The development of the next generation of robots must engage and answer satisfactorily the central question behind the field of artificial intelligence: how to equip machines with the ability to learn and acquire new information independently? Even now, the most sophisticated humanoid robot in the world—ASIMO—is limited by the fact that it has to be routinely programmed to perform its complex kinematic functions. The greatest challenge to the advancement of the next generations of intelligent robots has to do with how to develop a synthetic architecture that can make independent decisions; express basic cognitive skills through emotive, intelligent, and linguistic expressions; and is capable of freely responding to external stimuli.
If we view the human brain and its complexity from a neuroscientific and information processing perspective, then the biomimetic (human-made processes, substances, devices, or systems that imitate nature) obstacles here are extremely formidable. The vast challenges in this area derive primarily from the inability of current technologies to replicate the function of the human brain on even a modest scale.
Consider the complexity of the human brain. While it weighs a mere three pounds, the brain is comprised of 100 billion neurons, which send and receive electro-chemical pulses to and from the brain and nervous system. These electro-chemical charges are called synapses and are produced and propagated by sodium and potassium ions that send these charges along the neuron in a process called ‘synaptic transmission’ or ‘neurotransmission.’ As one source describes, this process of synaptic transmission has been likened to a binary circuit (though it’s significantly more complex than this): “There are trillions of synaptic junctions in the human brain. Learning occurs at least in part by changes in the number, strength and kind of synaptic connections. Early studies of neurons focused on the on-off characteristic of action potentials and a misleading comparison has been made with the transistor binary switch in digital circuits.”
A major technological step forward towards emulating the functionality and circuitry of the brain was announced recently through a research study conducted at the University of Southern California. Funded by the National Science Foundation, the “synthetic cortex” relies on the possibility of building synthetic neurons from carbon nanotubes in order to replicate human brain functions. Alice Parker, Professor of Electrical Engineering, is leading a team of researchers at USC in the quest to introduce a new approach to replicating human cortical functions, well beyond the scope of previous attempts in this field. In a Science Daily article last month, Parker points out that “the brain is kind of like a biochemical factory, operating in a sphere that you can’t stretch out on integrated circuits and circuit boards in order to emulate all of its electrical activity. The connectivity is too great and too many delays are introduced. We had to turn to nanotechnology to build something three-dimensionally, so that eventually we’ll be able to emulate how the neurons fire and activate others along a specific path within that sphere.”
Until quite recently, the size and cost of conventional electronics would have made construction of complex brain-like structures totally impractical, if not impossible. The ability to emulate basic brain functions and neuroplasticity —which is the major characteristic of the brain that enables it to organize new neural pathways through learned experience—simply could not be developed through conventional hardware. Even though the brain construction represents what Parker calls a completely “non-linear phenomenon,” her research team is proceeding with an attempt to computationally map out the complex interactions that occur between each neuron in the cortex and the tens of thousands of other neurons that are attached to it. The researchers involved with the “synthetic cortex” study have shown that carbon nanotube circuit models can be used to actually model portions of a brain neuron. Carbon nanotubes are molecular-scale tubes of graphitic carbon, which are among the stiffest and strongest fibers known, and have remarkable electronic properties and many other unique characteristics. According to another source, “Parker and her co-researcher, Chongwu Zhou, are in the process of combining these circuit models of neurons to create a functional carbon nanotube circuit model of a small network of neurons. This small network of interconnected neurons will be simulated using the carbon nanotube models.” Parker believes carbon nanotubes are an ideally suited method to emulate brain functions because their 3-D structure allows connectivity in all directions on all planes—similar to what occurs within the neuronal connections of the human brain.
The prospects for this technology have wide-ranging applications for the future in a variety of scientific fields. As Parker points out, “Researchers have already built experimental cochlear implants that are able to restore some hearing in the deaf and new vision systems that can restore some sight to the blind, but what we’re working on now is what you’ll see 30 years in the future. This is work that could revolutionize neural prosthetics, for one thing, and give us some pretty amazing biomimetic devices.”
Finally, to this statement should be appended the word “humanoid robots.” The development of synthetic brain neurons constructed from carbon nanotubes has potentially significant applications to the development of robots in general and humanoid robots in particular. The capability of constructing 3-dimensional circuits at the nanoscale mean will not only result in orders of magnitudes increases in computing power, but will eventually result in scalable applications to the development of robots with human-like levels of cognitive sophistication and agility.
[The animation below shows an artist's conception of a carbon nanotube synapse. The orange nanotubes are PMOS transistors (metal-oxide-field effect transistors with material containing an excess of holes) and the green nanotubes are NMOS transistors (metal-oxide-field effect transistors with material containing an excess of electrons). The red features are metallic interconnections and transistor gates, and the blue features are metallic interconnections. Credit: Khushnood Irani and Alice Parker, University of Southern California]
Tags: carbon nanotubes, human robot interaction, humanoid robot, nanocircuits, nanoelectronics, neuroscience, robotics
Thanks for discussing our work. I should point out that it will be many years before we can build the neural structures we are simulating. Lab work on some simple synapses is just starting!