Duke University

Brain-Computer Interfaces

Paper 1

Benjamin C Lawrence

Math 89S: Mathematics of the Universe

Professor Hubert Bray

September 27, 2016

Introduction

From sci-fi movies like theTerminator trilogy andI, Robot to children’s shows such as Treasure Planet, fictional characters exist in modern cinema that are a mixture of both organic humans and inorganic computers. However, these cyborgs might not be fiction for much longer. In the last 10-15 years, scientists have been developing methods to integrate computers and living creatures in the hope of both helping humans and controlling animals. For example, BrainGate is a research initiative that is studying how computer implants can be placed in a tetraplegic’s[1] brain so that one can control a computer mouse with his mind, and scientists such as Dr. Zhaohui Wu are working on placing computer chips into “roborats” in order to remotely control them.

Mind Over Matter

In 2002, a medical company called Cyberkinetics was formed to help raise money for research involving brain-computer interfaces (BCIs) and in doing so began the BrainGate project. BrainGate’s main purpose is to help tetraplegics use a computer to interpret neural signals which are then sent to an electronic device, such as a computer mouse or a prosthetic limb, to perform an action (“BrainGate”, 2016).

Although a tetraplegic is not capable of moving his limbs, he is still able to think about moving an arm, leg, or any other part of his body. As a result, scientists can observe what happens in the brain when a subject thinks he is moving his arm. Then theycan then program computers to perform an action when the subject recreates the same neural signals. Using this method, BCIs have engendered the control of cursors on a computer screen, robotic arms, and a few other devices by tetraplegics (Jarosiewicz, Sarma, Bacher, & Masse, 2015).

While this sounds amazing at first, researchers must overcome one significant obstacle before BCIs can be of any real use: calibration. The BrainGate team soon found that the neural signals for an action such as moving a cursor are different from person to person and can even change for one given person over time. As a result, two experiments formed: closed-loop and open-loop. A closed-loop trial is one in which the subject is not actually controlling the mouse cursor. Instead, the computer analyzes what the brain is doing while the cursor is moving. In open-loop, on the other hand, the mouse cursor is moving in real time to the subject’s thoughts. Originally, scientists would have a subject in a closed-loop environment until they had enough data to allow the subject a few minutes of open-loop time[1]. However, the BrainGate team recently published an article on how retrospective target inference (RTI) can be used to keep a computer calibrated to a subject’s thoughts while the subject is in an open-loop trial (Jarosiewicz et al., 2015).

While in an open-loop experiment, the computer must find two directions for the mouse cursor. The first is the direction read from the brain and the second is the actual direction the subject wants the cursor to move after computer error, also called bias, has been dealt with. Figure A shows the results from an experiment the BrainGate team performed on a subject with amyotrophic lateral sclerosis (ALS). The black dots represent the bias recorded over time by the computer and the red arrow is the average bias recorded. Figure B shows how the computer accounts for bias by removing the average bias from the velocity vector[1]that it detected as the direction the brain wanted the cursor to go. RTI attempts to make the process


even more efficient by analyzing past targets in order to predict a patient’s future targets. In contrast, in a traditional experiment a dot or some other object appears on a screen and the subject tries to move the cursor towards the image with his mind.The problem with this method is that the computer can calculate for bias very accurately because it already knows which direction the object on the screen actually is relative to the cursor. As a result, the computer has an advantage in figuring out which direction the subject wants the cursor to move. To overcome this built-in advantage, the RTI system is used during heuristic experiments in which the subject moves the cursor wherever he wishes. This method is much more realistic since in the real world there are a large number of places that users may go to click on an everyday computer screen. This technique forces the computer to use past data to see how the subject has been moving the cursor to try and predict how he will move it in the future (Jarosiewicz et al., 2015).

Although BrainGate technology is making significant advances, running experiments on human brains has a few drawbacks. Any experiment that involves cutting open a human’s head in order to place electronic implants must first have numerous trials on nonhuman primates (NHPs). But there are a number of issues with this situation. First, the intracranial space of a human is much larger than an NHP. This means that human brains are more likely to move within the skull, giving the computer a much harder time reading neural signals. Second, experiments performed on NHPs are in controlled, noise-reducing laboratories, while those on tetraplegics are often done in the person’s own home[1]. Therefore, an experiment that might have worked very well in a lab on a monkey might not work at all in an average tetraplegic’s living room (Jarosiewicz et al., 2015).

Matter over Mind

While some scientists have tried to help disabled people by developing means for the control of computers with a human mind, others have been working on the exact opposite: controlling the mind of animalswith a computer. In the article “The Amazing Adventures of Roborat,” Dr. Miguel A. Nicolelis from Duke University described how electrical brain stimulation[1] can be used as a form of operant conditioning[2] to control rats.His paper also discussed how this research can be used for biomimetics, a general description for developing engineering processes or systems that mimic biology (Paulson, Linda, 2004), and for the use of rats (instead of robots) in search and rescue operations (Nicolelis, Miguel, 2002).

Neurons and effectors are electrical components of the body which can be physically hacked like any regular computer. Thus, in Dr. Wu’s paper “Maze Learning by a Hybrid Brain-Computer System,” he and his team show exactly how a BCI can control a rat and how this affects its ability to perform certain tasks. They accomplished this “mind control” by inserting electrodes into the bilateral medial forebrain bundles (MFB)[3] in order to control of the rat’s serotonin and dopamine levels. Then the computer can “influence” the rat’s decision by encouraging it to go a certain direction and then rewarding it for doing what the computer said (Wu, Zheng, Zhang, Zheng, & Gao, 2016).

In Dr. Wu’s experiments there were two different types of ratbots. The first ratbot, as shown in Figure A, had a backpack stimulator and a bird’s-eye camera placed above a maze that

the ratbotwas to navigate. The Second ratbothad a backpack stimulator as well, but this time a camera was attached to the backpack instead of hanging above the entire maze. The maze was constructed such that the walls could be moved around, allowing for a variety of setups. The one common aspect was that at each junction there was exactly one correct way to go and one incorrect path. Dr. Wu’s team placed an indicator to point in the correct direction and measured how long the rats took to understand the purpose of the indicator. At the end of the maze the rats received water as a reward. After they found the water, the maze would be set to a new configuration and the test run again (Wu et al., 2016).

Dr. Wu discovered that the control rats took much longer than the ratbots to solve each maze. When the camera, either on the rat’s head or in the sky, saw that a ratbot was at a junction, it gave the rat a stimulus of MFB when facing toward the correct path. As the ratbot got closer to the finish, the stimulus became stronger. As a result, the ratbots learned very quickly to use their newly enhanced BCI brains to solve mazes in half the trial numbers than the control rats and had a stable performance of an 80% success rate (Wu et al., 2016).

Similar to using BCIs in humans to control computers, there are a number of drawbacks to installing BCIs to control rats. First, stimulating a rat’s brain to make it “feel good” to go in a certain direction does not actually force the rat to move that way. Therefore, saying that scientists have “control” over rats is a slight exaggeration. In reality, scientists have achieved a very strong influence over how a rat behaves. Second, experiments thus far have been performed only on rats, not any other type of animal. As discussed earlier, nonhuman primates and humans have different brain signals and, as a result, require different means of calibrating. In the same way, a rat and a dog could require different types of brain stimulation, severely diminishing the applications of studying rats.

Future Applications

Overall, BCIs could have very serious implications in our society’s future. For example, they couldenable disabled people not only to control computers with their minds, but also entire exoskeletons. If so, this application would be a great step toward curing the world of physical disabilities. Moreover, BCIs that control animals could one day be used as a form of therapy, allowing for the treatment of the mentally disabled in addition to the physically disabled. Therefore, BCIs are an important area of research that will hopefully one day make the world a better place.

Works Cited

  1. Beata Jarosiewicz, Anish A. Sarma, and Daniel Bacher. “Virtual Typing by People with Tetraplegia Using a Self-Calibrating Intracortical Brain-Computer Interface”. Science Translation Medicine. 17September, 2016. /313ra179
  2. Medical-Dictionary. Web. 23 September 2016.
  3. “BrainGate.” Wikipedia. Web. 23 September 2016.
  4. Merriam-Webster Dictionary. Web. 24 September 2016.
  5. Nicolelis, Miguel. “The Amazing Adventures of Roborat.” Science Direct. Web. 20 September 2016.
  6. “Electrical Brain Stimulation.” Science Direct. Web. 22 September 2016.
  7. “Operant Conditioning.” Science Direct. Web. 23 September 2016.
  8. “Electrophysiological.” Medical-Dictionary. Web. 22 September 2016
  9. “Postcentral Gyrus.” Wikipedia. Web. 24 September 2016.
  10. “Medial Forebrain Bundle.” Science Direct. Web. 23 September 2016.
  11. Science Direct. Web. 22 September 2016.
  12. Paulson, Linda. “Biomimetics Robots.” Web. 25 September 2016.
  13. Zhaohui Wu, Nenggan Zheng, Shaowu Zhang, Xiaoxiang Zheng, Liqiang Gao, Lijuan Su. “Maze Learning by a hybrid brain-computer system.” Nature. Web. 22 September 2016.
  14. Reuters. “Scientists control mouse brain by remote control.” Science Daily. Web. 15 September 2016
  15. “BrainGate.” BrainGate. Web. 20 September 2016.

1

[1]Tetraplegia (aka quadriplegia): paralysis of all four limbs, motor, and/or sensory function in the neck; part of the spine is impaired or lost due to damage of that part of the spinal cord.

[1] In monkeys, long periods of neural recordings are possible, but human’s neural activity can change over time because of physiological and/or recording nonstationarities (i.e., neural signals that vary over time) causing open-loop trials to be very short.

[1] Note that the term “velocity”describes the cursor motion because it has both a direction and a magnitude that must be measuredaccurately.

[1] The BrainGate team performs experiments in people’s homes deliberately because their long term goal is for BCIs to be used in more ordinary environments than laboratories.

[1]Electrical brain stimulation: electrical stimulation of the brain through chronically implanted electrodes.

[2]Operant Conditioning (aka The Law of Effect): behaviors followed by positive consequences are strengthened, while behaviors followed by negative consequences are weakened.

[3]Bilateral Medial Forebrain Bundles: a fiber system coursing longitudinally through the lateral zone of the hypothalamus that also carries fibers from cell groups containing norepinephrine and serotonin in the brainstem to the hypothalamus and cerebral cortex, as well as dopamine-carrying fibers from the substantial Nivea to the caudate nucleus and putamen. The hypothalamus: the part of the brain that regulates body temperature, certain metabolic processes, and other autonomic activities