DOW-UAP-D130: Controlling External Devices Without Limb-Operated Interfaces - A Brain-Machine Interface Survey

Source file: DOW-UAP-D130_AAWSAP-DIRD-Technological-Approaches-to-Controlling-External-Devices-March-23-2010.pdf Originating agency: Defense Intelligence Agency (DIA), Defense Warning Office, under the AAWSA Program Document type: Defense Intelligence Reference Document (DIRD), "Acquisition Threat Support" series; control number DIA-08-1003-012 Date: 23 March 2010 (information cutoff date, ICOD: 1 December 2009) Classification: UNCLASSIFIED//FOR OFFICIAL USE ONLY (the FOR OFFICIAL USE ONLY caveat is struck through on every page; released 2026) Page count: 36 VIRIN: 260918-D-D0360-1119 PURSUE Release: 6

Summary

DOW-UAP-D130 is one of the Defense Intelligence Reference Documents (DIRDs) produced for the Defense Intelligence Agency's AAWSA Program. Its full title is "Technological Approaches to Controlling External Devices in the Absence of Limb-Operated Interfaces," and its subject is what is now usually called a brain-machine interface (BMI): operating a machine by thought, or by nerve and muscle signals, instead of by hand. The author appears only as the redacted pseudonym AAP Person 73.

It is a technology survey with a forecast attached. The paper sets a concrete performance bar, more than 5-10 bits of information per second and a response time of 300 milliseconds or less for choosing one of N commands, and measures every technique against it: scalp EEG, MEG, muscle-based EMG, fMRI, near-infrared spectroscopy (NIRS), implanted cortical electrode arrays, peripheral nerve implants, optical stimulation and neural tissue grown on chips.

The verdict has two parts. In the near term, noninvasive electrical sensors that mostly read muscle signals will dominate practical use. In the far term, true high-bandwidth control will require invasive, two-way links at the level of single neurons, in the cortex or the cerebellum. The paper does not mention UFOs or unidentified aerial phenomena (UAP) anywhere.

Research Article

A neuroscience survey inside an aerospace program

The cover identifies the document as a DIRD in the "Acquisition Threat Support" line, dated 23 March 2010, with an information cutoff date (ICOD) of 1 December 2009 and control number DIA-08-1003-012. It was "Prepared by" the Acquisition Support Division (DWO-3) of the Defense Warning Office, in the DIA's Directorate for Analysis. The administrative note on page 2 calls it "one in a series of advanced technology reports produced in FY 2009" under the AAWSA Program and directs comments and questions to AAP Person 1, the AAWSA Program Manager at DIA, at "ATTN: CLAR/DWO-3, Bldg 6000, Washington, DC." A copyright warning forbids further dissemination of the document's photographs.

The structure is conventional: a two-page introduction, background on neural signals, a survey of noninvasive and invasive technologies, a five-part discussion and a conclusion of about a page and a half, with 13 figures and 60 references. (Page numbers here are PDF page numbers; in the body of the document the printed number is six lower.) Of the twelve technical areas in the AAWSAP Statement of Objectives, the paper belongs most clearly to "human interface," with a secondary bearing on "control." Its aerospace touches are incidental: an EMG system that "could be used to untie a pilot from the cockpit" (page 15), a remark that the cold sink and vacuum of a space-based application could ease MEG's cryogenic overhead (page 14), and a hypothetical "12-direction rocket stabilization system" used to illustrate a decoding problem (page 19).

The target: 5 bits per second, 300 milliseconds

The introduction defines the endpoint: thought-based operation of remote machinery during normal human activity, with no mechanical device, no shielded room, no need to stay perfectly still and no handheld gadget with buttons. "High-bandwidth" means data rates above 5-10 bits per second (page 5).

The background explains why this is hard. A neuron's action potential lasts 1-10 milliseconds, and neurons fire at most at 100 Hz to 1 kHz; metallic circuits, by contrast, are "about a million times faster" than the biological system, a mismatch that has so far prevented a direct connection between the two (pages 7-8). The human neocortex is about 2.5 mm thick, and even an eye blink involves signalling through a million neurons (page 9). Noninvasive methods see only large groups of neurons firing together. The much-studied P300 signal appears about 300 milliseconds after a stimulus, and the blood-oxygen (BOLD) response measured by fMRI and NIRS peaks 4-6 seconds after an event, too slow for the paper's targets (pages 9-10).

The paper also supplies benchmarks: finger pointing conveys about 14 bits/s, a mouse about 8 bits/s, and stylus-tapping a soft QWERTY keyboard on a PDA about 5 bits/s (page 12). It raises a practical question that is often skipped: for a healthy person a hand movement is usually simpler, so a BMI earns its place mainly when the hands and feet are already busy. It notes the cognitive limit on supervising several semi-autonomous systems at once, and speculates that over-training through pathways in the cerebellum could make control nearly automatic (page 12).

Noninvasive methods: EEG, MEG, EMG, fMRI and NIRS

EEG, first described in 1929, records the summed activity of nominally 50,000 neurons (pages 12-13). The best results come from ALS patients, whose paralysis removes muscle noise, and even there a command takes seconds. The paper is openly sceptical of the consumer headsets of 2009: the NeuroSky and Emotiv systems lack peer-reviewed support, raising doubt "about whether neural signals are being measured at all," while OCZ's NIA admits it reads a mix of EEG, EMG and eye-muscle signals (page 13). Figure 3 shows the NeuroSky MindSet and the OCZ nia. A peer-reviewed system with six dry electrodes reached 90 percent accuracy on a one-dimensional cursor, and twelve electrodes could cut the error rate to about 5 percent (page 14). Elsewhere the paper puts EEG at 20-30 bits per minute, under 0.5 bits/s (page 20).

MEG is ruled out for field use: brain fields of about 100 femtotesla are "about 100 million times smaller than the Earth's magnetic field" (page 14). EMG, which uses the muscles as "a biological amplifier of neural signals," reached 1-2 bits/s with minimal training, about four times a comparison EEG forehead sensor (page 15). fMRI offers unmatched 3-D resolution among noninvasive methods, but its signal lags by seconds; the paper sees combined low-field MRI and MEG as "a long term prospect, likely in the 20- to 40-year timeframe" (page 16). Single-trial NIRS decisions reached only 80 percent accuracy (page 16).

Invasive methods: arrays, neurochips, lampreys and human volunteers

The invasive survey starts from the cochlear implant (22 electrodes, about 100,000 users, more than 30 years of development) and moves to cortical electrode arrays (pages 16-17). Open-loop decoding runs into a basic problem: algorithms learn from how the brain moves a real arm, so decoding commands for a device "the brain has never controlled before" is much harder (page 19). Closed-loop work is illustrated by the CB-i humanoid robot, which reproduced in real time the walking movements of a monkey on a treadmill at Duke University after the decoded data were relayed to Advanced Telecommunication Research in Japan (Figure 4, pages 19-20).

The paper reviews a 56-gram, battery-powered "Neurochip" with 12 microwire electrodes in a monkey's motor cortex (page 21); experiments showing that the lower layers of the cortex encode directional information better (pages 22-23); MRI-compatible microwire arrays that recorded activity on 12 of 16 electrodes over six weeks without significant tissue damage (pages 25-26); and hybrid systems in which the brainstem of a lamprey, a primitive jawless vertebrate, steered a small wheeled robot toward light (Figure 12, pages 26-27).

On human trials the paper is direct about the ethics: volunteers "would likely have difficulty understanding all of the risks involved." It cites the recipient of a second-generation artificial vision system, spelled "Jans Naumann" in the document, who now admits he did not understand the risks and is considering having the implants removed; a footnote says the procedures were performed in Lisbon "to avoid U.S. prohibitions against implant surgery" (page 28). Epilepsy patients with temporary ECoG grids moved a cursor in 1-2 seconds with up to 75 percent accuracy; amputees who had undergone nerve reinnervation selected one of 10 motions in under 170 milliseconds; and a healthy self-experimenter used a median-nerve implant for three months in 2002 (pages 28-29).

The forecast: 5 years, 20-40 years, 30-40 years

The discussion groups everything into five areas. The near-term call is commercial: mainly EMG-based sensors, driven by the gaming market, "in the next 5 years" (page 30). EEG-only devices for healthy people are "unlikely in the next 5 years," and the paper proposes a test of progress: noninvasive detection of a coherent group of fewer than 10,000 neurons (pages 30-31). MEG would need portable shielding that attenuates background "nearly 10 orders of magnitude" (page 32). Using the cochlear implant as a yardstick, the metal-electrode path is 30-40 years away barring a disruptive advance (page 32).

The long-term bets are optical stimulation, a noncontact "full-duplex information channel" to single neurons, and electrode chips encased in neural tissue grown ex vivo, so that the implant's contact is "biological to biological" (pages 32-33). The conclusions set the far-term goal at 5-20 bits/s and point to the cerebellum as a connection point that would make the external device, as far as the brain is concerned, part of the body (page 34).

What the document does not say

There is no reference to UFOs, UAP, unconventional craft or reverse engineering, and nothing about interfacing with non-human technology. There is also no classified data: every source is open literature from 1929 to 2009, and web sources were "last accessed 12 May 2009." Readers should note some loose ends. The bandwidth target shifts between 5-10 bits/s (introduction), "the target 5 bits/sec" (discussion) and 5-20 bits/s (conclusions). The lamprey studies are cited as References 28-30, but those numbers belong to EEG and MEG papers in the bibliography, apparently a citation error. And the prediction that a handful of dry sensors "should surpass" traditional interfaces within five years sits uneasily with the paper's own figures: 1-2 bits/s for EMG against about 8 bits/s for a mouse.

Significance

D130 shows how broadly AAWSAP defined its "human interface" area: a DIA aerospace program commissioned a literature review in neuroscience. Its value to the archive is as a snapshot of mainstream BMI research in 2009, with its benchmarks and caveats, and as evidence that much of the DIRD series had nothing to do with unidentified phenomena. The timelines it commits to, five years for commercial progress, 20-40 years for combined MRI and MEG, and 30-40 years for electrode interfaces, give later readers a dated yardstick against which to measure how the field actually developed.

Key People

Locations

Key Concepts

Notable Quotes

"The technology endpoint this paper seeks is thought-based operation of remote machinery during normal human activities without mechanical device interaction..." -- page 5

"The paper argues, mainly by eliminating other approaches, that the most probable technology in the long term involves invasive single-neuron-based direct cortical connections to form a network of high-bandwidth duplex communication pathways." -- page 5

"The Neurosky and Emotive systems claim proprietary processing algorithms, lack peer-reviewed literature supporting their claims, and even discuss using facial muscle movement to send signals (References 26, 27), raising doubt about whether neural signals are being measured at all." -- page 13

"...the fact that 100 fT is about 100 million times smaller than the Earth's magnetic field will prove an insurmountable barrier to sifting signal from noise in anything but a heavily-shielded, metal-free environment." -- page 14

"Such a system could be used to untie a pilot from the cockpit." -- page 15

"It will thus be quite an issue in an open-loop design to develop an algorithm that can decode control signals for a system the brain has never controlled before, say a 12-direction rocket stabilization system." -- page 19

"If the 30-year advancement of cochlear is a realistic guide of innovation, the proximal electrical technology path to success is 30-40 years off save for a disruptive advance in the development of penetrating electrodes." -- page 32

"The cerebellar connection point is extremely interesting to perhaps provide a connection that would effectively make the external device a part of the body as far as the brain is concerned." -- page 34