From: 93asw@williams.edu (Andrew Wright) Newsgroups: alt.cyberpunk.tech Subject: brain-computer.short Followup-To: alt.cyberpunk.tech Date: 7 Dec 1992 02:25:24 GMT Organization: Williams College Distribution: world NNTP-Posting-Host: mac_ccc_si.cc.williams.edu On Designing a Brain-Computer Interface: After all, computers were once science fiction, too. Andrew Wright NSCI 401 Williams College 11/20/92 93asw@williams.edu awright@mindvox.com Note1: This paper is copyright 1992. I am making this paper available becuase of a large interest in the topic among people on the net. Since this is an academic paper, I am making the assumption that people out there will be ethical enough not to pass this work off as their own. To that end, feel free to disseminate this and use it in your own research, but please make sure to cite your source. Thanks - Andrew. Note2: This is a loosely edited version of this paper, with much of the technical detail left out. A full version is available on request. I would also be willing to place it on a site allowing anonymous ftp, if anyone knows of an appropriate site. A Brain-Computer interface is a staple of science fiction writing. In it's earliest incarnations no mechanism was thought necessary, as the technology seemed so far fetched that no explanation was likely. As more became known about the brain however, the possibility has become more real and the science fiction more technically sophisticated. Recently, the cyberpunk movement has adopted the idea of "jacking in", sliding "biosoft" chips into slots implanted in the skull (Gibson, W. 1984). Although such biosofts are still science fiction, there have been several recent steps toward interfacing the brain and computers. Chief among these are techniques for stimulating and recording from areas of the brain with permanently implanted electrodes and using conscious control of EEG to control computers. Some preliminary work is being done on synapsing neurons on silicon transformers and on growing neurons into neural networks on top of computer chips. The most advanced work in designing a brain-computer interface has stemmed from the evolution of traditional electrodes. There are essentially two main problems, stimulating the brain (input) and recording from the brain (output). Traditionally, both input and output were handled by electrodes pulled from metal wires and glass tubing. [] (Pickard 1979). Using conventional electrodes, multi-unit recordings can be constructed from mutlibarrelled pipettes. In addition to being fragile and bulky, the electrodes in these arrays are often too far apart, as most fine neural processes are only .1 to 2 µm apart. [] It is difficult to permanently implant such arrays, and consequently it is difficult to directly study the brain as a function of animal behavior. [] Pickard describes a new type of electrode, which circumvents many of the problems listed above. These printed circuit micro-electrodes (PCMs) are manufactured in the same manner of computer chips. A design of a chip is photoreduced to produce an image on a photosensitive glass plate. [] This is used as a mask, which covers a UV sensitive glass or plastic film. A PCM has three essential elements: 1) the tissue terminals, 2) a circuit board controlling or reading from the terminals and 3) a Input/Output controller-interpreter, such as a computer. The circuit board and computer are often located outside the skull, to minimize tissue invasion, allow for long-term implantation and permit the electrodes to be detached between trials (Kuperstein and Eichenbaum 1985). [] In addition to the ability to make multiple, closely spaced recordings, P[CMs] often outperform the traditional electrodes in a number of electronic measures (Kuperstein and Eichenbaum 1985). A further advantage of PRONG [a type of PCM] was it's continued functioning after as many as four days implantation. [] PRONG was able to simultaneously make 10-11 recordings from one side of the electrode. While it is tempting to see in PRONG the potential for permanently implanted brain recording and stimulating devices in the manner of cyberpunk fiction, more mundane if equally exciting applications of similar technology are being found now. A six channel PCM is being commercially produced for use as an implant in patients who have lost hearing but not functioning of the auditory nerve (Ineraid Multichannel Cochlear Implant). [] This device allows for hearing and speech recognition, although there are limits to the amount of information that can be extracted. In two-syllable recognition tests, scores range from 0 to 100%, with the median being 44% (Dorman et al. 1991). Interestingly, these limits may not be inherent in the cochlear device, but in the encoding algorithm. Wilson et al. (1991) have designed a new technique, CIS, which is presumably based on improved Digital Signal Processing (DSP) capabilities. [] The increase in comprehension engendered by this technique overwhelmed the sensitivity of the tests. In some cases, the subjects were well within the range of mild to moderate hearing loss. [] Another possible use for PCMs is controlling robotic prosthetics. A special type of tissue terminal, an enclosure terminal, has holes in or across conductors through which developing or regenerating neurons can grow(Pickard 1979). These are especially suitable to chronic preparation, and could be implanted in the PNS where nerve regeneration is possible. The chip could then interpret motor neuron signals for use controlling prostheses.[] [PCMs] may even be useful in administering micro-doses of ionotophoretic drugs (Pickard 1979). A fundamentally different approach to interpreting output from the brain is the use of EEGs. According to Wolpaw et al. (1991), "in theory [the] brain's intentions should be discernible in the spontaneous EEG." However, the vast number of neurons and the complex structure of the brain make such interpretation difficult if not impossible. Therefore, efforts have focused on training people to produce desired EEGs through biofeedback mechanisms. [] An immediate use for such a system can be seen in providing a mechanism for communication between paralyzed patients and the outside world through the computer. The possibilities of interpreting EEG data and using it to control computers have been brought to the consumer electronics front by the IBVA, or Interactive Video Brainwave Analyzer (Nathan 1992). A headband with four adhesive electrodes sends data through a radio transmitter to a port on a Macintosh personal computer. The EEG is the filtered and run through a fast fourier transform before being displayed as a three dimensional graphic. The data can then be piped into MIDI compatible music programs. Furthermore, MIDI can be adjusted to control other external processes, such as robotics. The level of control provided by IBVA is limited at best and the software does not actually interpret the brain's impulses. Instead, the user must program the software to interpret consciously determined gross changes in the EEG. The interface between the brain and computers, either through interpreting EEGs or through recording directly through PCMs is currently limited by computing strength. Conventional computers are well suited to processing linear data, but only have limited application to more distributed processes such as pattern recognition. In order to address these problems, neural net computers are modeled after the brain's complex system of weighted synapses. The strength of these neural nets can be considered a function of the number of connections made between functional units. Computers are hampered by the limited number of connections imposed by the constraints of processing time and memory space. To circumvent this, Masuo Aizawa is working on growing neurons into neural net computers (Freedman 1992). A neuron is capable of processing many more inputs and outputs than a transistor, and is obviously uniquely suited to neural net computing. [] It is currently possible to grow strips of interconnected neurons on [a semi-conductor chip] the oxide []. Of course, input to and output from the "chip" is highly problematic. Aizawa's solution is to divide the underlying semiconductor into electrodes that can stimulate or record from the cells, however he projects that this may take several years to develop (Freedman 1992). Other researchers are working on ways to interface neurons directly into silicon chips as well. In a "first step toward multiple recording in neurons and neural nets and toward the development of neural biosensors and neuroelectronic circuits", Fromherz et al. grew neurons into silicon insulated-gate Field Effect Transformers (FETs) (1991). [] According to Fromherz et al., the neuron-Si junction, outperforms metallic electrodes, because of the capacitive coupling at high seal resistance as attained by adhesion of the neuron to the gate without a metallic conductor. It is possible to construct patterns of such silicon FETs on a single chip, allowing multiple recordings in a cell culture. There is an added advantage of the possibility of long term recordings at high resolution and high signal-to-noise ratios (Fromherz et al. 1991). If this type of neuron-Si junction could be integrated into printed circuit micro-electrodes, a extremely versatile and functional interface until could be constructed, an improvement over metallic electrode PCMs. If incorporated into enclosure type tissue terminals, researchers and clinicians would have access to an interface between neurons and electronic computers with high efficiency, conservation of information and one-to-one neuron-electrode junctions. Immediate applications of this include increasing the functionality of cochlear implants. By increasing the number of electrodes, the efficiency and specificity of the electrode-neuron interaction and the computational power driving the electrodes, speech recognition could be enhanced even further than currently possible. This prefigures applications in which PCMs are used to input data from computers directly into normal people's auditory nerve, bypassing the actual production of sound by the computer. Such functionality would be relatively easy to add by including an input jack on the DSP chip of the implant. A similar application could be found in encoding visual stimuli directly into the optic nerve of blind people. This could then be adapted to present computer generated visual signals divorced from real world input. The auditory and optic nerves are perhaps the most accessible methods of input to the brain. Memory structures in the brain itself are considerably less well understood. However it is not outside of the realm of possibility to directly include artificial memories through the use of advanced PCMs implanted in the brain. The problem of encoding information in a manner recognizable to the human brain would be difficult to surmount, especially given the possibility that data structures may be encoded differently by each person. [] It may be possible, then, to tailor such neural nets to output data in a manner specific to one person's internal representations of memories. Although current technology is probably incapable of such a feat, because of the incredible amount of processing required, the combination of increased silicon chip speed and the use of complex neural nets built from neurons could make this possible. The current approaches to interfacing the brain and computers described above offer very concrete examples of utility. Indeed, the potential for even currently available systems is still unrealized. However, it could be argued that they are hampered by inherent limitations. EEG based systems have no possibility of input to the brain, and full comprehension of the human EEG may be out of reach by even the most advanced imaginable computers. Similarly, PCM recording or stimulating devices may be limited by both the size of the electrode arrays, the difficulty of implantation and by the complexity of the brain. Despite optimistic projections, the brain's complexity may make it impossible to directly access from a computer. Even if these problems could be overcome, there are still problems involved with the very idea of implanting electrodes into the brain. The potential damage to biological structures may outweigh the tentative benefits of a direct computer interface, especially in higher cortical structures. Despite these conceptual difficulties, the prospect of extending the inherent capabilities of the organic brain allows the consideration of transcending the limitations imposed by the corporeal body. After all, computers were once science fiction, too.