Anticipatory Intelligence Analysis with Cogent: Current Status and Future Directions Gheorghe Tecuci, Dorin Marcu, Mihai Boicu, Chirag Uttamsingh Learning Agents Center, Volgenau School of Engineering, George Mason University, Fairfax, VA 22030, USA { tecuci; dmarcu; mboicu; cuttamsi }@gmu.edu Abstract IC and DOD organizations. While praising their theoretical Anticipatory intelligence analysis is the complex task of framework and evidentiary knowledge, the analysts desired drawing defensible and persuasive conclusions about future a simplified interface and interaction, which has led to the events or states based on current information. This paper development of Cogent (Tecuci et al., 2015; 2018). presents a systematic approach to anticipatory intelligence In this paper we discuss how Cogent supports an analyst analysis with Cogent, a software cognitive assistant that enables a synergistic integration of the analyst’s imagination in performing anticipatory analysis. We start with a brief and expertise with the computer’s knowledge and critical account of the complexity of this task. Then we introduce a reasoning. It shows how current, as well as envisioned systematic approach to anticipatory analysis which is capabilities of Cogent, help alleviate many of the grounded in the science of evidence (Schum, 2009). complexities of this task. Following that we present two examples of anticipatory analysis with Cogent, and discuss how it assists the analysts Introduction in coping with their complexity. Finally we discuss future developments of Cogent. Anticipatory intelligence analysis is the complex task of drawing defensible and persuasive conclusions about future events or states based on current information of all kinds that Complexity of Anticipatory Analysis come from a variety of different sources. It addresses new and emerging trends, changing conditions, and Evidence-based Reasoning underappreciated developments (ODNI, 2019). The evidence upon which the anticipation of possible future The prevailing approach to anticipatory intelligence states or events eventually rests has five major analysis and intelligence analysis in general is the holistic characteristics that make these anticipations necessarily approach where the analysts, after reviewing large amounts probabilistic in nature (Tecuci et al., 2016a, pp.159-167). of information and performing the reasoning in their heads, The evidence is always incomplete no matter how much we reach a conclusion (Marrin, 2011). A complementary have. It is commonly inconclusive in the sense that it is approach uses structured analytic techniques, such as those consistent with more than one future state or event. Further, described by Heuer and Pherson (2011) that guide the the evidence is frequently ambiguous and, in most hypothesis generation and testing process performed by the situations, dissonant, some of it favoring one future state or analysts. Some of these methods and more advanced ones event while other evidence favoring others. Finally, the based on Bayesian probabilistic inference networks are evidence comes from sources having different levels of implemented in analytical tools, such as Netica (2019). credibility. Arguments to test the hypothesized future states We have developed a sequence of cognitive assistants or events are necessary in order to establish and defend the based on Wigmorean networks (Wigmore, 1937). The first three major credentials of evidence: its relevance, its of these systems, Disciple-LTA, integrates capabilities for credibility, and its inferential force or weight. These analytic assistance, learning, and tutoring (Tecuci et al., arguments rest upon both imaginative and critical reasoning. 2008). TIACRITIS and its subsequent version, Disciple-CD (Tecuci et al, 2016a), were developed primarily for teaching intelligence analysis and were experimentally used in many Assessing the Credibility of Evidence ______________________________________________________________________ Evidence credibility assessments form the very foundation Copyright © 2020 for this paper by its authors. Use permitted under for all arguments we make from evidence to possible future Creative Commons License Attribution 4.0 International (CC BY 4.0). states or events. The different types of tangible, testimonial, and mixed evidence have many credibility indicators and all unimportant in light of what we learn today. The would need to be assessed based on ancillary evidence in consequence is that the complex process of discovery or order to have high confidence in the accuracy of our investigation in anticipatory analysis is a ceaseless activity. anticipations (Tecuci et al., 2016a, pp. 118-133). Limits of Individual Probability Views Anticipatory Intelligence Analysis While anticipatory intelligence analysis is probabilistic in in the Framework of the Scientific Method nature, none of the non-enumerative probability views Following the framework of the scientific method, we known to us (Subjective Bayesian, Belief Functions, model anticipatory analysis as ceaseless discovery of Baconian, and Fuzzy) can optimally cope with all the five evidence, anticipations, and arguments, in a non-stationary characteristics of evidence mentioned above (Schum, world, involving collaborative computational processes of 2001a; Tecuci et al., 2016a, pp.173-208). For example, the evidence in search of anticipations, anticipations in search conventional Subjective Bayesian view cannot cope well of evidence, and evidentiary testing of anticipations, as with ambiguities or imprecision in evidence. On the other represented in Figure 1 (Tecuci et al., 2016a). hand, the Fuzzy view can naturally cope with such First, through abductive (imaginative) reasoning that imprecisions. But neither the Bayesian view nor the Fuzzy shows that something is possibly true (Schum, 2001b), we view can account for the incompleteness of the coverage of generate alternative future events or states that may explain evidence. The only view that can account for this is the an intelligence alert. If, instead of an intelligence alert, the Baconian view where the probability of a future state or starting point is an intelligence question, the alternative event depends on how complete the evidence is, or how anticipations are the possible answers to this question. Next, many questions recognized as being relevant remain through deductive reasoning that shows that something is unanswered by the evidence we have. This is in contrast necessarily true, we use the hypothesized future with the Bayesian, Belief Functions, and Fuzzy views that states/events to generate new lines of inquiry and discover all rest on how strong is the evidence we have about the new evidence. After that, through inductive reasoning that considered future state or event. While on the Bayesian shows that something is probably true (Schum, 2001a), we probability scale “0” means disproof, on the Baconian scale, test each anticipation by developing an argumentation that “0” simply means lack of proof. A future state/event shows how the discovered evidence favors or disfavors it. currently having “0” Baconian probability can be revised As shown at the bottom of Figure 1, these are upward in value as soon as we have some evidence for it. collaborative processes that support each other in recursive calls. For example, the discovery of new evidence may lead Time Constraints to the modification of the existing hypotheses of future A major objective of anticipatory intelligence analysis is to events/states or the generation of new ones that, in turn, lead help insure that the policies and decisions reached by the to the search and discovery of new evidence. Also, governmental and military leaders, at all levels, are well inconclusive testing of the considered anticipations requires informed. In many cases analyses are required to answer the discovery of additional evidence. The next sections questions that are of immediate interest and that do not allow illustrates this process using Cogent in two analyses. analysts time for extensive research and deliberation on Alternative Future Events or States Probabilities of Anticipations available evidence regarding the questions being asked. Non-Stationary World As outlined above, anticipatory intelligence analysis has many inherent difficulties, but none seem more difficult than the fact that analysts must assess future states or events Alert or Question New Evidence in a non-stationary world that keeps changing as analysts are Evidence in search Anticipations in Evidentiary testing of anticipations search of evidence of anticipations trying to understand it. As a result, we have continuing Abduction Deduction Induction streams of new information, some items of which are A or Q  possibly H H  necessarily E E  probably H relevant evidence regarding our anticipations. An Figure 1: A framework for anticipatory analysis. explanation for some pattern of past events analysts have previously regarded as correct may now seem incorrect in light of new evidence just discovered today. A future event Alert-Driven Anticipatory Analysis regarded as highly likely today may be overtaken by events we will learn about tomorrow. In fact, the very questions we The following example of anticipatory analysis with Cogent asked yesterday may need to be revised or may even seem shows how evidence about a missing cesium-137 canister leads to anticipating that a dirty bomb will be set off in the cesium-137 canister was stolen. Then it is possible that it Washington, D.C., area (Tecuci et al., 1016a). Note that this was stolen by a terrorist organization, or by a competitor of scenario and all the entities involved are fictitious. XYZ, or by an employee. Upper level hypotheses concern Mavis, a counterterrorism analyst, reads in today’s possible future events, such as, the terrorist organization will Washington Post that a canister containing cesium-137 is use the cesium-137 canister to build a dirty bomb, the dirty missing from the warehouse of the XYZ Company in bomb will be set off in the Washington, D.C., area, or it will Maryland (see E* at the bottom of Figure 2). The question be set off in the New York area. is: What hypothesis would explain this observation? The analyst and Cogent need to assess each of these Through abductive (imaginative) reasoning, Mavis infers competing hypotheses, and determine which of them are that a dirty bomb will be set off in the Washington, D.C., likely. Starting from bottom-up, each hypothesis is put to area (see H5 at the top of Figure 2). However, no matter how work to guide the collection of additional evidence: imaginative or important this future event is, no one will Assuming that the cesium-137 canister is indeed missing, take it seriously unless Mavis and her cognitive assistant, what other things should be observable? What are the Cogent, are able to justify it. So they develop the chain of necessary conditions for an object to be missing from a abductive inferences shown in the left side of Figure 2: warehouse? It was in the warehouse, it is no longer there, We have evidence that the cesium-137 canister is missing. and no one has checked it out. Therefore it is possible that it is indeed missing. It is possible As a result, the analyst contacts Ralph, the supervisor of that it was stolen. It is possible that it was stolen by a the warehouse, who reports that the cesium-137 canister is terrorist organization. It is possible that the terrorist registered as being in the warehouse, that no one at the XYZ organization will use the cesium-137 canister to build a Company had checked it out, but it is not located anywhere dirty bomb. It is possible that the dirty bomb will be set off in the hazardous materials locker. He also indicates that the in the Washington, D.C., area. lock on the hazardous materials locker appears to have been But these are not the only hypotheses that may explain the forced. Ralph’s testimony provides several items of relevant evidence. Just because there is evidence that the cesium-137 evidence, and the question is: What is the probability that canister is missing does not mean that it is indeed missing. the cesium-137 canister is missing, based on this evidence? At issue here is the credibility of the source of this To answer this question, Mavis and Cogent build the information. Thus an alternative hypothesis is that the Wigmorean probabilistic inference network from Figure 3 cesium-137 canister is not missing. But let us assume that it (Wigmore, 1937; Tecuci et al., 2018). It integrates logic and is missing. Then it is possible that it was stolen, but it is also Baconian probabilities with Fuzzy qualifiers, and uses the possible that it was misplaced, or maybe it was used in a min/max probability combination rules common to the project at the XYZ Company. Now let us assume that the Baconian and Fuzzy views of probability (Cohen, 1977; 1989; Zadeh, 1983). That is, the probability of a conjunction H5: A dirty H’5: A dirty of hypotheses is the minimum of their probabilities, and the bomb will be bomb will probability of a disjunction of hypotheses is the maximum Where? set off in the be set off in the New of their probabilities. Washington, D.C., area York area First Mavis and Cogent have to assess the probabilities of the bottom hypotheses in Figure 3, based on the Why? H4: build dirty H’4: build corresponding relevant evidence. Then these probabilities bomb unshielded are composed to produce the probability of the top radiation sources hypothesis. The probability of a hypothesis, like the one H3: stolen H”3: stolen from the bottom left of Figure 3, shown also at the top of H’3: stolen by Who? by terrorist by employee Figure 4, is assessed based on the three credentials of competitor organization evidence: credibility, relevance, and inferential force (Tecuci et al., 2016a, pp. 62-73). They are assessed by using H”2: used the ordered symbolic probability scale from the upper right How? H2: stolen H’2:misplaced in project of Figure 3. As in the Baconian system, “lacking support” for a hypothesis means that we currently have no basis to H’1: not consider that the hypothesis might be true. However, we What? H1: missing missing may later find evidence to infer that the hypothesis is, for instance, “likely.” Figure 3 shows a favoring argument for E*: Article on the top hypothesis and therefore it appears under the left cesium-137 canister missing (green) square. Disfavoring arguments (if any) appear under the right (pink) square. Figure 2: Multi-step abduction and competing explanations. The credibility of the evidence item E2 in Figure 4 is Figure 3: Wigmorean probabilistic inference network. assessed as very likely because its source, Ralph, has access the cesium-137 canister was stolen with a U-Haul truck. to the reported information and has a reputation for honesty. Having concluded that the cesium-137 canister is The relevance of E2 is assessed as almost certain because missing, Mavis and Cogent have now to establish whether it the records of the XYZ Company are almost certainly was stolen with a truck, it was misplaced, or it was used in correct. Consistent with both the Baconian and the Fuzzy a project at the XYZ Company. Each of these hypotheses is min/max probability combination rules, the inferential force put to work to guide the collection of evidence for assessing of E2 on the hypothesis H is determined as the minimum it: If the cesium-137 canister was stolen with a truck, what between the credibility of E2 (very likely) and the relevance other things should be observable? of E2 (almost certain). Thus, the inferential force of E2 on Based on the current evidence, Mavis imagines the H is very likely. Obviously, an irrelevant item of evidence following scenario on how the cesium-137 canister might will have no inferential force, and will not convince us that have been stolen: The truck entered the company, the the hypothesis is true. An item of evidence that is not canister was stolen from the locker, the canister was loaded credible will have no inferential force either. Only an item into the truck, and the truck left with the canister. of evidence that is both relevant and credible supports the Such scenarios have enormous heuristic value in truthfulness of a hypothesis. advancing the investigation because they consist of mixtures Because in the argumentation from Figure 4 there is only of what is taken to be factual and what is conjectural. one item of favoring evidence, E2, its inferential force on Conjecture is necessary in order to fill in natural gaps left by the hypothesis is also the probability of the hypothesis. In the absence of existing evidence. Each such conjecture, general, however, the probability of the hypothesis would be however, opens up new avenues of investigation, and the the result of the balance of probabilities between the combined inferential force of the favoring evidence items very likely Probability of H and the combined inferential force of the disfavoring items. H: The canister was in the warehouse As shown at the top of Figure 3, it is very likely that the very likely Inferential force: Probability of H based cesium-137 canister is missing, this being the minimum only on E2 between the probabilities of the three sub-hypotheses and almost certain Relevance: Probability of H2 assuming the relevance of their conjunctive argument. that E2 is true Some of the newly discovered evidence may trigger new very likely Credibility: Probability that E2 is true hypotheses or the refinement of the current hypotheses. For E2 Canister registered: Ralph, who has example, during her initial investigation, Mavis discovered a reputation for honesty, reports that a video segment from a security camera at the warehouse the cesium-137 canister is registered as showing a person loading a container into a U-Haul truck, being in the warehouse. leading her to refine the “stolen” hypothesis to indicate that Figure 4: Evidence credentials. discovery of additional evidence, if the scenario turns out to Jihad Bis Sayf will set off a dirty bomb in the Washington, DC, area. be true. This scenario, for instance, leads Mavis to check the records of the security guard and they show that a panel & truck bearing Maryland license plate number MDC-578 was in the XYZ parking area on the day before the discovery that reasons desire capability the cesium-137 canister was missing. Fusing all the discovered evidence, Mavis and Cogent … & & conclude that it is very likely that the cesium-137 canister was stolen with the MDC-578 truck. After further Jihad Bis Sayf is A dirty bomb set ability to ability to set off investigation, they also conclude that the two competing a terrorist off in Washington, obtain a its dirty bomb in organization D.C., the capital of dirty bomb the Washington, hypotheses, “misplaced” and “used in a project,” lack opposed to the the United States, D.C., area. evidentiary support. United States. would have a very high impact. … Continuing the investigation with the rental company … owning the truck, it is discovered that Omar al-Massari … Jihad Bis Sayf will Jihad Bis Sayf has a build a dirty bomb. rented the MDC-578 truck giving his alias, Omer Riley, and presence in the a false address, and that the truck is now contaminated Washington, D.C., area. & because cesium-137 is radioactive. These lead to the … conclusion that Omar used the truck to steal the cesium-137 has has has expertise has secure canister. It is further discovered that Omar al-Massari has radioactive explosive to build a place to build ties with terrorist organizations, and that he has given the material material dirty bomb a dirty bomb cesium-137 canister to Saeed al-Nami, alias Kenny … … … Derwish, who is a member of the terrorist organization Jihad & Bis Sayf. These discoveries lead to the specializations of the hypotheses from Figure 2 as shown in Figure 5. cesium-137 canister The cesium-137 Figure 6 shows the analysis of the top anticipatory was stolen by Jihad canister contains hypothesis “Jihad Bis Sayf will set off a dirty bomb in the Bis Sayf, a terrorist enough cesium-137 to organization. build a dirty bomb. Washington, DC, area.” It shows that Jihad Bis Sayf has reasons, desire, and capability to set off the dirty bomb. It … … has reasons because it is a terrorist organization opposed to Figure 6: Anticipatory analysis. the United States, it has a presence in the Washington, DC, area, and a dirty bomb in this area would have a very high Through such spiral hybrid reasoning, where abductions, impact. Furthermore, Jihad Bis Sayf has both the ability to deductions, and inductions feed on each other in recursive build the bomb and to set it off. In particular, it has the calls, Mavis and Cogent continuously generate and update radioactive material from the stolen cesium-137 canister, intermediate alternative hypotheses, use these hypotheses to and further investigation has determined that it has both the guide the collection of relevant evidence, and use the necessary explosive material (Saeed al-Nami has stolen 2 evidence to test these hypotheses, until the probability of the pounds of RDX explosive) and expertise to build the bomb top-level hypotheses are assessed, ultimately anticipating (Saeed al-Nami has expertise in explosives and has received that Jihad Bis Sayf will likely set off a dirty bomb in the training in the building of dirty bombs). Washington, D.C., area. Note, however, that performing this analysis is not as H5: Jihad Bis Sayf will set off a dirty simple as one may infer from this presentation. It is the Where? bomb in the Washington, D.C., area methodology from Figure 1 and Cogent that guide the Why? H4: Jihad Bis Sayf will build a dirty bomb analyst and simplify it. Many things can and will indeed go wrong. But Cogent provides the means to deal with them. H3: cesium-137 canister stolen by Based on evidence, you come up with some hypotheses, but Who? Jihad Bis Sayf, a terrorist organization then you cannot find more evidence to support any of them. So you need to come up with other hypotheses, and you H2: cesium-137 canister stolen by How? should always consider alternative hypotheses. The Omar al-Massari with MDC-578 truck deduction-based decomposition approach guides you on What? H1: cesium-137 canister missing how to look for evidence, but your knowledge and imagination also play a crucial role. As illustrated here, E*: Article on cesium-137 canister missing Mavis imagined a scenario where the cesium-137 canister was stolen with a truck. But let us now assume that she did Figure 5: Evidence-based hypothesis specialization. not find supporting evidence for this scenario. Therefore, Mavis has to imagine other scenarios. Maybe the cesium (likely). They combine into an overall desire of likely. canister was stolen by someone working at the XYZ Finally, almost certainly the United States have the required Company, or maybe it was stolen by Ralph, the capabilities that consist of required homegrown scientific administrator of the warehouse. The important thing is that knowledge, technical knowledge, economic resources, and each such scenario opens up a new line of investigation. natural resources. Question-Driven Anticipatory Analysis Current Cognitive Assistance The previous section illustrated a situation where the Consider again the process described in Figure 1. With the anticipatory analysis was driven by an intelligence alert. current version of Cogent, the analyst has to imagine the This section illustrates a situation where the analysis is possible future states or events. However, Cogent helps with driven by the intelligence question: Who will be the world developing argumentations that lay out the underlying leader in wind power within the next decade? analytical framework for every anticipation, including the The top level of the corresponding anticipatory analysis connection between the evidence and various intermediate is shown in Figure 7. It is likely that the United States will hypotheses in the analysis, the evaluation of the credibility be the world leader in wind power within the next decade of evidence and its strength in supporting a hypothesis, and because almost certainly they have reasons, likely they have the role of assumptions in addressing missing information. the desire, and almost certainly they have the necessary Anticipatory analysis may be affected by the analyst’s capability. A reason is that significant production of wind biases. Cogent can detect several of them, such as the power will reduce the current need of the United States to confirmation bias (building an argumentation and/or only consume huge quantities of oil that represent a danger to the searching for evidence that confirms the analyst’s beliefs environment. The desire of the United States, which is a while dismissing or ignoring evidence to the contrary), the representative democracy, is determined by the desire of the satisficing bias (choosing the first hypothesis that appears people (almost certain), the desire of the major political good enough, rather than carefully identifying all possible parties (very likely), and the desire of the energy industries hypotheses and determining which one is the most consistent with the evidence), and potential absence of evidence bias (failure to consider the degree of completeness of the available evidence). Many other biases are avoided because explicit argumentations are developed that employ an intuitive system of symbolic probabilities. Additionally, Cogent facilitates the analysis of what-if scenarios, where the analyst may make various assumptions and Cogent automatically determines their influence on the analytic conclusion. It also automatically updates the computed probabilities based on new or revised evidence. Once the analysis is finalized, Cogent generates a structured report that the analyst then transforms into a more understandable and persuasive report that includes argumentation fragments and evidence, can be shared with other analysts, subjected to critical analysis, and correspondingly improved. The hierarchical structure of the Wigmorean argumentation enables the analyst and Cogent to perform the analysis at different levels of abstraction. Moreover, the analyst may drill down on selected sub-hypotheses as much as allowed by the available time and evidence. Consider, for example, the dirty bomb anticipated event from Figure 6. In time-limiting situations the analyst may assume that Jihad Bis Sayf has the reasons and desire to set off the dirty bomb, and focus the investigation on its capability. Note that an anticipated future state made at a given moment in time may change afterwards because many of its indicators are dynamic, such as those that determine the Figure 7: Another example of anticipatory analysis. desire of the United States in the wind power evidence scenario (see Figure 7). Therefore evidence of these indicators needs to be continuously credibility of image monitored and updated. In the current version of tangible testimonial missing authoritative Cogent, this monitoring has to be done by the evidence evidence evidence record analyst who also needs to insert the new authenticity reliability evidence in the analysis. After that Cogent accuracy automatically updates the analysis. However, as demonstrative real unequivocal equivocal discussed in the next section, continuous tangible tangible testimonial testimonial monitoring and updating of evidence can also be evidence evidence evidence evidence automated. credibility of human source Cogent has a knowledge base that includes an unequivocal testimonial testimonial probabilistically ontology of evidence and rules for assessing its evidence based upon evidence based equivocal testimonial direct observation on opinion evidence credibility. Figure 8 shows a fragment of this accuracy ontology (Schum et al., 2009). For each type of competence unequivocal testimonial veracity completely equivocal evidence from this ontology, Cogent has a evidence obtained at testimonial evidence second hand procedure for assessing its credibility. For example, as illustrated in the left hand side of Figure 8: Evidence ontology and credibility patterns. Figure 8, the credibility of an item of demonstrative tangible evidence (e.g., a map, a sound time, Cogent will learn domain analysis rules from the recording, or a satellite image) depends on its authenticity, contributions of the Analyst, through the employment of the its accuracy, and the reliability of the instrument the Disciple-EBR multistrategy learning approach, which produced it. The credibility of a human source depends on integrates learning from examples, learning from the source’s competence, veracity, and accuracy. These explanations, and learning by analogy and experimentation, indicators depend on lower level indicators. For example, in a mixed-initiative interaction with the expert. Successive the competence depends on the source’s access and versions of this learning approach are presented in (Tecuci expertise, while the accuracy depends on the source’s 1998; Tecuci et al. 2002; 2005; 2008; 2016b). Cogent will objectivity and observational sensitivity. also facilitate the development of analyses in collaboration As discussed in the next section, future research will with other analysts, each contributing sub-arguments based address the problem of learning general analysis rules from on their expertise, reviewing and commenting on each- an expert analyst, which will speed-up and improve the other’s contributions, and sharing previously learned development of new analyses. Note, for example, that the domain analysis rules, sources, and evidence. analyses from the previous sections, although very different, (2) Automatic Analysis Updating, shown in the bottom both make use of the following reasoning pattern: An actor part of Figure 9, where Cogent (through its collaborative will perform a certain action or achieve a certain state if it autonomous agents) continuously monitors the Multi-INT has reasons, desire, and capability. The rules learned from Environment and updates the performed analysis based on the analysis in Figure 7 will enable Cogent to automatically the newly discovered evidence. These agents are copies of generate analyses of future states such as: China will be the the corresponding modules of the Learning and Reasoning world leader in solar energy within the next decade. Anticipations Learning to Probabilities discover and Mixed-Initiative monitor evidence Mixed-Initiative and Automatic Learning and Reasoning Assistant Learning to Learning to generate test Anticipatory Analysis Alert or Question hypotheses Collection Requests/Evidence hypotheses Reference Repository Management Service We plan to significantly extend Cogent with Alert Surveillance Knowledge Base (KB) knowledge-based reasoning and learning capabilities, Agent Manager Surveillance Surveillance effectively evolving it into the multi-agent architecture Hypothesis Surveillance Hypothesis Agent Agent Generation KB Queue Agents from Figure 9, similar to that described in (Tecuci et al., Generation Agent Hypothesis Analysis Multi-INT Environment 2019). It will have two complementary functions: KB Queue Hypothesis (Sensors and Sources) Hypothesis Hypothesis (1) Analysis and Learning, shown in the upper part Analysis Agent Collection Evidence Collection Analysis Agent Collection Collection Analysis Agent Agent of Figure 9, where the Analyst and Cogent (through its Agent Agents KB Queue Evidence Collection and Mixed-Initiative Learning and Reasoning component) User Review Monitoring KB Queue Agent Manager will rapidly develop complex, logical, and compelling argumentations in a transparent manner. At the same Figure 9: Envisioned extended architecture of Cogent. Assistant except that they are configured to run Heuer, R. J., and Pherson, R. H. 2011. Structured Analytic autonomously and communicate by developing and Techniques for Intelligence Analysis, CQ Press, Washington, DC. exchanging Knowledge Bases. The component agents of Marrin, S. 2011. Improving Intelligence Analysis: Bridging the gap between scholarship and practice. Routlege, London & New York. Cogent will be connected to the application environment through a Surveillance Manager and a Collection and Netica, 2019. https://www.norsys.com/ Monitoring Manager, the latter continuously monitoring the ODNI, 2019. The National Intelligence Strategy of the United States of America, Office of the Director of National Intelligence. results returned by the Collection Agents that operate on the Multi-INT Environment. Once a new or updated evidence Schum, D. A. 2009. Science of Evidence: Contributions from Law and Probability. Law Probab Risk 8 197–231. item is detected, it is introduced in the analysis by the Schum, D. A. 2001a. The Evidential Foundations of Probabilistic Evidence Agent, and the analysis is updated by the Reasoning. Northwestern University Press. Hypothesis Analysis Agent. Schum, D. A. 2001b. Species of Abductive Reasoning in Fact The automatic hypothesis generation is performed by the Investigation in Law, Cardozo Law Review, 22 (5-6):1645–1681. Alert Agent in collaboration with the Hypothesis Generation Schum, D., Tecuci, G., Boicu, M., Marcu, D., Substance-Blind Agent. Next, the hypothesis-driven evidence discovery is Classification of Evidence for Intelligence Analysis, in performed by Hypothesis Analysis Agents in collaboration Proceedings of the Conference “Ontology for the Intelligence with the Evidence Agent. Then evidence requests are issued Community,” Fairfax, Virginia, 20-22 October 2009. to collection agents through the Collection and Monitoring Tecuci, G. 1998. Building Intelligent Agents: An Apprenticeship Manager. After that, evidence-based hypothesis testing is Multistrategy Learning Theory, Methodology, Tool and Case performed by the Evidence Agent in collaboration with the Studies, San Diego: Academic Press. Hypothesis Analysis Agents. Tecuci, G., Boicu, M., Marcu, D., Stanescu, B., Boicu, C., Comello, J., Lopez, A., Donlon, J., Cleckner W. 2002. Development and Deployment of a Disciple Agent for Center of Conclusions Gravity Analysis. Proceedings of the Eighteenth National Conference of Artificial Intelligence and the Fourteenth After reviewing several of the complexities of anticipatory Conference on Innovative Applications of Artificial Intelligence, intelligence analysis, this paper illustrated how they can be 853-860. Edmonton, Alberta: AAAI Press. alleviated through the use of the Cogent cognitive assistant Tecuci, G., Boicu, M., Boicu, C., Marcu, D., Stanescu, B., within a systematic approach based on the science of Barbulescu, M. 2005. The Disciple-RKF Learning and Reasoning evidence and the scientific method. Key to overcoming Agent. Computational Intelligence, 21 462-479. these complexities and performing more accurate Tecuci, G., Boicu, M., Marcu, D., Boicu, C., Barbulescu, M. 2008. Disciple-LTA: Learning, Tutoring and Analytic anticipatory analyses based on imperfect information in a Assistance. Journal of Intelligence Community Research and dynamic world, is the synergistic integration of the analyst’s Development, July. imagination and expertise with the computer’s knowledge Tecuci, G., Marcu, D., Boicu, M., Schum, D. A. 2015. Cogent: and critical reasoning. Cognitive Agent for Cogent Analysis, in Proceedings of the 2015 AAAI Fall Symposium “Cognitive Assistance in Government and Public Sector Applications,” 58-65, Arlington, VA, Technical Acknowledgements Report FS-15-02, AAAI Press, Palo Alto, CA. David Schum has significantly influenced this research that Tecuci, G., Schum, D. A., Marcu, D., Boicu, M. 2016a. was supported in part by NSF under grant number 1611742, Intelligence Analysis as Discovery of Evidence, Hypotheses, and Arguments: Connecting the Dots, Cambridge University Press. by AFRL under contract number FA8750-17-C-0002, by IARPA under contract number 2017-16112300009, and by Tecuci, G., Marcu, D., Boicu, M., Schum, D. A. 2016b. Knowledge Engineering: Building Cognitive Assistants for Evidence-based George Mason University. The views and conclusions Reasoning, Cambridge University Press. contained herein are those of the authors and should not be Tecuci, G., Kaiser, L., Marcu, D., Uttamsingh, C., Boicu, M. 2018. interpreted as necessarily representing the official policies, Evidence-based Reasoning in Intelligence Analysis: Structured either expressed or implied, of any organization of the U.S. Methodology and System, Computing in Science and Engineering, Government. The U.S. Government is authorized to 20(6) 9-21, November/December. reproduce and distribute reprints for governmental purposes Tecuci, G., Meckl, S., Marcu, D., Boicu, M. 2019. Instructable notwithstanding any copyright annotation therein. Cognitive Agents for Autonomous Evidence-Based Reasoning. Seventh Annual Conference on Advances in Cognitive Systems, August 2-5, MIT, Cambridge, MA. References Wigmore, J. H. 1937. The Science of Judicial Proof: As Given by Cohen, L. J. 1977. The Probable and the Provable, Clarendon Logic, Psychology, and General Experience and Illustrated in Press, Oxford. Judicial Trials, 3rd edition, Little, Brown & Co, Boston, MA. Cohen, L. J. 1989. An Introduction to the Philosophy of Induction Zadeh, L. 1983. The Role of Fuzzy Logic in the Management of and Probability, Clarendon Press, Oxford. Uncertainty in Expert Systems, Fuzzy Sets and Systems, 11 199- 227.