“NEURINT” and Neuroweapons - Neurotechnology in National Security and Defense - page 83

detection, processing, and/or integration. Neurotechnologic innovations that are
capable of processing high volume, complex data sets are based upon computational
biology as physiomimetic computing hardware (NAS 2008). Such hardware lever-
ages analog, rather than digital components, with “a continuous set of values and
a complex set of connections,” based on an understanding of neural networks as
more than mere binary switches. An analog circuit approach would address cur-
rent “modeling and simulation challenges,” be smaller and “. . . easy for the US—
and its adversaries—to construct” (NAS 2008). As well, given the analog nature of
the magnetic fields used for real-time computing, a small, portable, physiomimetic
computer of this type might be uniquely valuable for applications of high-density
information processing (Watson 1997; Konar 1999; von Neumann 2000; Giles 2001;
Arbib 2003; Siegelmann 2003; Schemmel et al. 2004; Trautteur and Tamburrini
2007). The Systems of Neuromorphic and Adaptable Plastic Scalable Electronics
(SyNAPSE) program at the Defense Advanced Research Projects Agency (DARPA)
has explored these possibilities (Pearn 1999). Table 7.1 provides an overview of sev-
eral other such NSID research programs with a neuroscientific focus. (Information
used in the discussion of DARPA/IARPA programs was derived from the respective
program websites, which are listed in the “Online sources” section of the references.)
Information systems could conceivably be conjoined so that neural mechanisms for
assigning and/or detecting salience (i.e., processes involving cortical and limbic net-
works) may be either augmented or modeled into neurotechnologic devices for rapid
and accurate detection of valid (i.e., signal vs. noise) information within visual (e.g., field
sensor, satellite and unmanned aerial vehicle [UAV]-obtained images) and/or auditory
aspects (e.g., narratives, codes) of human intelligence (HUMINT) or signal intelligence
(SIGINT). Formulating and testing credible hypotheses while monitoring large amounts
of information could be accomplished by computational cognitive frameworks that are
capable of both self-instruction (e.g., using the Internet as a “training environment”) and
learning from experience (e.g., via direct access to the operational environment). This
articulates a form of artificial intelligence (AI) that functions to mimic human neural
systems in cognition. The 2008 NAS ad hoc committee identified such technology as a
potential threat, but one that remains largely theoretical—at least at present (NAS 2008).
Nevertheless, efforts are already underway with these aims (Table 7.1). For exam-
ple, the Mind's Eye Program (MEP) at DARPA is already attempting to build a “smart
camera with machine-based visual intelligence,” capable of learning “generally
applicable and generative representations of action between objects in a scene”. For
text-based environments, DARPA's Machine Reading Program seeks to replace
“expert and associated knowledge engineers with unsupervised or self-supervised
learning systems that can “read” natural text and insert it into AI knowledge bases
(i.e., data stores especially encoded to support subsequent machine reasoning).”
The Intelligence Advanced Research Project Activity's (IARPA) program titled
Knowledge Representation by Neural Systems (KRNS) is aimed at “understanding
how the brain represents conceptual knowledge to lead to building new analysis
tools that acquire, organize, and yield information with unprecedented proficiency”.
Another IARPA program, Integrated Cognitive Neuroscience Architecture for
Understanding Sensemaking (ICArUS), seeks to understand how humans conduct
sense-making under various conditions and how bias affects computational models