<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Spyglass - Medium]]></title>
        <description><![CDATA[Hacking for Defense team using the lean startup methodology to solve national security problems for the DOD and Intelligence Community. - Medium]]></description>
        <link>https://medium.com/spyglass?source=rss----cfbb6f8dc674---4</link>
        <image>
            <url>https://cdn-images-1.medium.com/proxy/1*TGH72Nnw24QL3iV9IOm4VA.png</url>
            <title>Spyglass - Medium</title>
            <link>https://medium.com/spyglass?source=rss----cfbb6f8dc674---4</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 11:56:35 GMT</lastBuildDate>
        <atom:link href="https://medium.com/feed/spyglass" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[“The Art of the Possible”]]></title>
            <link>https://medium.com/spyglass/the-art-of-the-possible-cc51b785cac2?source=rss----cfbb6f8dc674---4</link>
            <guid isPermaLink="false">https://medium.com/p/cc51b785cac2</guid>
            <category><![CDATA[lean-startup]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[innovation]]></category>
            <category><![CDATA[hacking-for-defense]]></category>
            <dc:creator><![CDATA[Katerina Sedova]]></dc:creator>
            <pubDate>Sun, 23 Apr 2017 21:20:42 GMT</pubDate>
            <atom:updated>2017-04-24T00:26:26.846Z</atom:updated>
            <content:encoded><![CDATA[<p>This is week 14 of the inaugural Hacking4Defense class at Georgetown. Team Spyglass is nearing the end of our journey within the framework of the class. We started down the H4D road with a broad challenge of “analyzing social media for identifying and predicting probable hotspots of social unrest” from our sponsors at the Asymmetric Warfare Group (AWG), US Army. We will be ending the class with a minimum viable product (MVP), which addresses a specific scenario within this broad challenge, a scenario where we can add value in force protection by applying image and facial recognition technology to alert troops overseas when a picture of military vehicles, military personnel, or an individual soldier’s own face is posted on social media without her knowledge, preventing a series of threats in general and crowd-sourced attacks in particular.</p><p>We have come full circle in our H4D experience. In week 3, on our very first “get-out-of-the-building” (in lean startup parlance) exercise, we visited Fort Meade to “walk in the shoes” of AWG operational advisors (OAs), to understand their workflow, pain-points, and the experience of troops they support. Little did we know that only 11 weeks later, we would be back at Fort Meade to show Spyglass in action, testing our MVP and getting feedback before our final push!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*J7u-IWqKm2DK26B408mqYg.jpeg" /><figcaption>Getting Out of The Building, one last time</figcaption></figure><p>On Friday morning, Jose, Chloe, and Katya did the first demo of Spyglass outside of class and our Georgetown location. In preparation for our visit, Jordan, who has been doing all the demoing of our MVP in class, did the knowledge transfer to Jose and Katya to ensure that we could run the MVP and troubleshoot potential issues without him in the field test, as he was not able to join us in Fort Meade. Lo and behold, of all the glitches we were anticipating, we did not expect that our conference room would have no wifi!</p><p>Jose and our AWG advisor Kyle quickly found a workaround, and we were able to present Spyglass and test it without a hitch. While Jose explained the inner workings of Spyglass and drove the live demo, Chloe and Katya conducted a loose focus group to capture the impressions of seven OAs and two of their commanders.</p><p>Overall, the feedback we received was quite positive. Our goal was to test the fundamental hypothesis: “Would our beneficiaries use this tool if they had access to it?” The answer of the AWG OAs and their commanders was a resounding yes! The next level of feedback could be summarized in the following three categories: 1) the current capabilities of the tool, 2) future capabilities/feature requests, 3) pointers on our future briefing(s) to the military leadership.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*n7W5xuYizBWBBA-KGllKJw.jpeg" /><figcaption>Testing Spyglass with AWG Operational Advisors</figcaption></figure><p>The first set of key insights focused on the current capabilities:</p><ul><li>The tool looks “tremendously useful”</li><li>For OAs, the simplicity and light weight of SMS alert is a big win, as other tools they currently use are “clunky” with sometimes too much information, making it difficult to tease out urgency</li><li>Spyglass can also be useful for the “insider threat” counterintelligence portion of force protection to identify unauthorized posts and images of military assets</li><li>Having data differentiation (vehicles vs general personnel vs specific individual) in the feed is helpful, especially if the user can customize which feeds she wants to receive.</li><li>Being able to train the neural network on photos of specific equipment, much like we do with individual, is tremendously helpful in detecting when pictures of convoys are posted on social media, while they are moving from one Forward Operating Base (FOB) to another or are on a resupply mission. Similarly, detecting pictures of physical locations (e.g. front gates of a base), would be tremendously helpful in force protection.</li></ul><p>The second category of key insights revolved around what is possible going forward and may lead to future functionality:</p><ul><li>Can Spyglass be trained to differentiate U.S. vs foreign military equipment/uniforms?</li><li>Dashboard view of the feed on a desktop would be helpful, as in the conventional army these feeds will likely be monitored by the S2/S3 intelligence or operations officer. The S-2 is also likely going to be operating in a classified environment and without a phone. So alerting would need to work through another mechanism, potentially email or visual dashboard alert.</li><li>Can bounding geographic rectangle be modified by the user to expand the area of coverage? Similarly, can geographical location be set by the user on the ground, to account for frequent short-term travel?</li><li>Can the system be trained to identify adversary drones in real time?</li><li>Counter surveillance consideration: if the phone number from which the alert is sent is always the same, it may draw attention of adversary’s surveillance. Is it possible to obfuscate the origin of the alert?</li><li>In addition to the hyperlink to the image that was posted, can alert include additional info, such as a handle from which the image was posted on social media and a map with a pin of location from where the photo was taken?</li><li>Training the tool on the photos of specific individuals assumes significant responsibility for their personally identifiable information (PII). Spyglass must have a story around storage (if any) and how it will protect against the potential cyber theft of this data.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Ktm-9nXBdj446iLMbv8b1g.jpeg" /><figcaption>Showing off facial recognition: training the model to recognize Chloe</figcaption></figure><p>The last category of feedback centered on how we speak about Spyglass in pitches and briefings to the military:</p><ul><li>Start the pitch with the threat we are trying to detect, rather than a problem statement, will help frame the conversation to the environment in which they operate</li><li>Be able to explain how this solution fits into the original problem challenge presented by the leadership</li><li>Be able to answer the question under which legal authority this tool would operate (Title 10 vs title 50)</li></ul><p>Additional interviews we have conducted this week with C., former Special Forces Green Beret and L., former Intelligence officer, US Marine Corps., have also echoed this feedback.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*c19pmXFGUoNbP3P_X35PnQ.jpeg" /><figcaption>Reflecting on the H4D experience and story-boarding our final video</figcaption></figure><p>Finally, as we near the end of our H4D journey and build our final presentation for the class, we have been looking back on the wild ride we have had with H4D. Through nearly 120 interviews of potential beneficiaries and technologists working in this space, we heard a wide spectrum of pain-points and ideas of what they think is important to glean from social media for force protection: from atmospherics to population sentiment, from social network analysis to the nuanced text processing and translation of ontologies, from geolocation and map overlays to geo-inferencing. Through beneficiary discovery, we — a group of graduate students without any military experience — learned a tremendous amount about the scope, workflow, pain-points, and the environment, in which our men and women in uniform serve around the world. Through “getting out of the building” experiences and mentorship of technology partners, we got glimpses of the magnitude of great work being done in this space, of what we didn’t know we didn’t know, and the scope of opportunity. Through our own playing with code to work around an obstacle of not having a pre-tagged dataset of threatening posts, we learned by building and ventured deeper into what is possible with image recognition.</p><p>At times what we heard in the beneficiary interviews sent us spinning in a circle of contradictions or running in opposite directions, as we tested our hypotheses, fine-tuned how to ask questions more effectively, and learned to tease out the essence of what beneficiaries actually need from what they said they needed. At times, the internal detractors — the naysayer within — almost got the better of us, overwhelmed by the complexity and nuance of our challenge. When the naysayer within says “we can’t”, this team said “we can” and it did. At times, this project eclipsed all other academic and job commitments, and yet, at times, we wished we could be working on it full time.</p><p>Our team is stronger than the sum of individuals within it, individuals who are human. Startups rarely begin with a group of people who barely know each other, who don’t already have an established dynamic of trust, communication and boundaries, coming together and hitting the ground running. As a team, we’ve had our moments when passion, personality dynamics, communication patterns, frustration with our direction or process got the better of us. That we were able to gel together and deliver the Spyglass MVP in the short 14 weeks is the testament to the professionalism, unique skill-sets, and the work ethic of the individuals within our Spyglass team, stronger together. Our skins have grown thicker through the relentlessly direct yet constructive precision questioning of our teaching team and the mentorship of our advisors.</p><p>So what’s next for our last week of H4D? We will be integrating the facial recognition into Spyglass end-to-end, continuing to reflect on the totality of our H4D experience for our final slide-deck, and building our final video. Tune in to our final presentation on May 1st as we look back on our experience and show off our MVP to the senior leadership of AWG.</p><p>The journey of Spyglass through the H4D classroom is ending. In 15 weeks, we will have shown the “art of the possible”, yet only scratching the surface. Will this journey’s end be another one’s beginning?</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=cc51b785cac2" width="1" height="1" alt=""><hr><p><a href="https://medium.com/spyglass/the-art-of-the-possible-cc51b785cac2">“The Art of the Possible”</a> was originally published in <a href="https://medium.com/spyglass">Spyglass</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Spyglass Technology Development]]></title>
            <link>https://medium.com/spyglass/spyglass-technology-development-e70c1eab7cf2?source=rss----cfbb6f8dc674---4</link>
            <guid isPermaLink="false">https://medium.com/p/e70c1eab7cf2</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[hacking-for-defense]]></category>
            <dc:creator><![CDATA[Jordan B.]]></dc:creator>
            <pubDate>Sun, 09 Apr 2017 21:26:40 GMT</pubDate>
            <atom:updated>2017-05-03T03:17:33.929Z</atom:updated>
            <content:encoded><![CDATA[<p>Understanding the inner workings of Spyglass requires understanding its supporting infrastructure. This is a description of the technology that happens behind the scenes:</p><p>We deploy our API script to an Amazon Web Services (AWS) EC2 Instance. This is easy to spin up and administer, and AWS offers affordable pricing. For storage of social media data, we use a PostgreSQL RDS instance, which is also hosted on AWS. Our EC2 micro instance also runs the <strong>Spyglass dashboard app</strong> persistently. Built in Python and Spyre, the dashboard app provides a simple interface for searching tweets in the database and displaying their image classifications in real time.</p><p>For the actual analysis of social media data, we have a Python script utilizing Tweepy client, locked into the Twitter API in precisely-specified geographic bounding boxes. We can deploy in multiple regions by creating multiple sessions on our server. This is the order of operations in which our analysis occurs when a new tweet is ingested:</p><ol><li>Spyglass parses out any media URLs (image or video).</li><li>If the text is in Arabic, Spyglass translates it into English.</li><li>If there is an image contained in the tweet, Spyglass executes the <strong>image classification</strong> <strong>algorithm </strong>and<strong> </strong>applies the resulting class label, if any, to the observation.</li><li>If untranslated Arabic text is contained, Spyglass applies a <strong>text classification algorithm</strong> to classify positive or negative sentiment, and applies the resulting label to the observation.</li><li>Once all of this is complete, Spyglass parses the text, media, coordinates, and labels, and writes all of this information to the RDS instance, where this data becomes instantaneously available in the dashboard app.</li></ol><p><strong>For our image classification algorithm, </strong>we have a pre-trained Convolutional Neural Network, trained in TensorFlow to detect military vehicles and military personnel on Twitter. TensorFlow is an open source package provided by Google for training Machine Learning models on GPUs, in a distributed format. The training procedure is arduous, so that is performed once on a dataset of images that we have aggregated. For prediction, a serialized model is deployed into the back-end of our analysis.</p><p>Convolutional Neural Networks operate by dividing up images into smaller overlapping “convolutions,” using their pixel positions and RGB values of the pixels as features. The backpropagation algorithm utilizes the training data to learn weights to apply to each of these features in order to achieve the desired outcomes. This resulting model can then be deployed on new images. Anytime we achieve a prediction with a probability higher than 0.7 for a Military Vehicle or US Soldier, we send an email alert over SMTP protocol. We utilize Twilio to send an SMS alert, which contains the detection class, image, and coordinates like so:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/966/0*85bJbcA7PNFTn64L." /></figure><p>We are continually refining the image classification feature. We have experimented with facial recognition algorithms to identify specific individuals, using Torch and CUDA to train algorithms for facial recognition on a GPU. We hope to deploy this functionality before the end of the semester.</p><p><strong>For the text classification algorithm, </strong>we first utilize Bing for translation. For the raw Arabic text we apply methods similar to what can be found in this <a href="http://ieeexplore.ieee.org/document/6716448/">paper</a>. We parse the data into bigrams and use their frequencies as features. We apply what is known as tf–idf scaling, which gives added importance to features common in a document but rare in the entire dataset. We then train a support vector machine to distinguish between positive and negative with an accuracy ~80%. The arabic training set is published in <a href="https://archive.ics.uci.edu/ml/datasets/Twitter+Data+set+for+Arabic+Sentiment+Analysis">UCI machine learning repository</a>.</p><p><strong>Implications.</strong> Our development cycle greatly enhanced our ability to iterate on Spyglass rapidly and accomplish our goals in 8 weeks. Instead of attempting to tackle a very large problem of threat detection with social media, we sought out specific ways to constrain our feature set to move the needle in precise areas of the larger problem that have not yet been tackled.</p><p>Acquiring the problem understanding necessary to put together an impactful solution is the hardest part. For weeks, we have been talking with stakeholders and potential users, referred to internally as beneficiaries, in order to understand their specific needs and pain points. Our interviews often led us in many directions due the size of the problem space, but we used our own judgment and technical understanding to arrive at a minimum viable product (MVP).</p><p>Initially we hypothesized that the real value-add of our product would be text classification for the purpose of the threat detection. The technical aspects of this challenge are rather simple, but aggregating a dataset capable of distinguishing threats from non-threats is a daunting task. We ultimately learned our stakeholders cared more about solving the problem of image recognition. To that end, we produced a dataset capable of classifying certain image classes for an MVP. We demonstrated a proof-of-concept by downloading publicly available images online.</p><p>In national security organizations, solutions often demand seven-figure bills-of-materials and years of development effort inside a non-stationary environment, in spite of their rapidly changing requirements. In true H4D spirit, we are proud of the fact that we can move the needle with a micro-solution that is under 300 lines of code, utilizes open source technologies, and meets our beneficiaries’ requirements.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e70c1eab7cf2" width="1" height="1" alt=""><hr><p><a href="https://medium.com/spyglass/spyglass-technology-development-e70c1eab7cf2">Spyglass Technology Development</a> was originally published in <a href="https://medium.com/spyglass">Spyglass</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Quickest Way to Win]]></title>
            <link>https://medium.com/spyglass/the-quickest-way-to-win-937bd773fee?source=rss----cfbb6f8dc674---4</link>
            <guid isPermaLink="false">https://medium.com/p/937bd773fee</guid>
            <category><![CDATA[government-contracting]]></category>
            <category><![CDATA[army]]></category>
            <category><![CDATA[innovation]]></category>
            <category><![CDATA[hacking-for-defense]]></category>
            <dc:creator><![CDATA[Max Weintraub]]></dc:creator>
            <pubDate>Sun, 26 Mar 2017 22:12:40 GMT</pubDate>
            <atom:updated>2017-03-27T02:38:21.415Z</atom:updated>
            <content:encoded><![CDATA[<p>This week, our team had the opportunity to visit Fort Meade and sit-in on a capability demo to AWG. Jose and I listened to a presentation by an early-stage technology company that uses data analytics and machine learning to counter violent extremism. Observing the representatives from Omelas present and field questions felt like an out-of-body experience because so much of the content seemed as though it was lifted straight out of our Hacking for Defense class. It felt like we were watching a version of ourselves from a few weeks ago.</p><p>It is a true testament to our development as a team and our progress through the lean startup methodology that we were able to keep up with the high-level discussion and contribute meaningful observations. The Omelas representatives were impressed to learn that none of our team members had any prior military experience. While I’m not at liberty to discuss the finer points of the presentation, many of the pieces of advice that AWG had for Omelas were relevant to us as well, particularly their advice for navigating the DOD procurement process.</p><p><strong>“The quickest win for a new thing is a successful demonstration in a training exercise.”</strong> This statement struck a chord with us as our team has begun preparations for Spyglass to be integrated in an Army training exercise next week at Fort AP Hill. This will be the first field test of Spyglass, and many eyes will be watching. Spyglass will be just one part of a larger, multi-domain exercise: there will be many moving parts and we will have minimal control over the environment. That’s why, in addition to our customer discovery efforts in validating our use case, we have spent a great deal of effort preparing for AP Hill.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*0A5qj758Wob1y80a37jwlQ.jpeg" /><figcaption>Spyglass’s activities roadmap</figcaption></figure><p>This week, we integrated all of the elements of Spyglass we’ve developed into a single tool. Jordan, our machine learning guru, got Jose’s image recognition program up and running inside Spyglass and is working to implement a real-time email alerting function. Katya and I hope to make progress on the legal and privacy front by speaking with a JAG lawyer in the coming weeks and Chloe has devised a killer strategy for getting interviews with target beneficiaries in both the conventional forces as well as special forces.</p><p>All of these tasks will be critical to the overall success of our team moving forward, but for the next few days our activities will be focused on nailing our product demonstration. Stay tuned for future updates as we will detail the results of this awesome opportunity.</p><p>Also don’t forget to RSVP for the <a href="https://www.eventbrite.com/e/the-future-of-innovation-in-national-security-tickets-33054296290">Future of Innovation in National Security </a>panel that will be replacing our regular class on April 3. Hope to see you there!</p><p><strong>Tech Discovery Summary</strong></p><p><strong>N. and M., TechChange.org, Clinic on Big Data for Foreign Policy</strong></p><p>Key takeaways:</p><ul><li>The world has collected unprecedented amount of data. Extracting insight from it is the next barrier to overcome.</li><li>Data and data scientists can’t inform the public on how to balance tradeoffs in policy. The big data revolution is too important to be left to engineers alone, so there is a growing need for policy makers to “speak data”.</li><li>CartoDB is a visualization, which provides map overlay, allows combining multiple datasets in variety of formats to demonstrate connections and relationships across datasets. Free tier at 500 mb, but may have academic license. Cool application of it was worldwide mapping of Ukraine-related hashtags during the February 2014 culmination of Euromaidan and Revolution of Dignity in Ukraine (<a href="https://srogers.carto.com/me">https://srogers.carto.com/me</a>)</li><li>Multitudes of sources for datasets exist: Data.gov, World Resources Institute (http://www.wri.org/), Humanitarian data exchange which provides open datasets that have mapping of crises, refugee flows, IOM datasets on IDPs, tracking where are people moving based on climate change and conflict (data.humdata.org)</li><li>Some possible sources of threat and social unrest datasets are: Visionofhumanity.org, Economist intelligence unit, Global peace index, Global terrorism index from Institute of Economics and Peace (http://economicsandpeace.org/reports/)</li><li>PremiseData trains a groundswell of superusers around the globe to take pictures with smartphones, “mapping reality on the ground”. Human-directed and machine-refined, Premise indexes and analyzes millions of observations, “unearthing connections that impact global decisions” (<a href="http://premise.com/">premise.com</a>)</li><li>TextIt enables creating sms campaigns through a canvas flow and send SMS alerts based on user responses. Can integrate with twilio API, which has integration and contracts with all major sms service providers in the US and some worldwide. (Textit.in)</li><li>We may be able to find a dataset at the Terrorism &amp; Preparedness Data Resource Center (TPDRC): <a href="http://www.start.umd.edu/data-tools/terrorism-preparedness-data-resource-center-tpdrc">http://www.start.umd.edu/data-tools/terrorism-preparedness-data-resource-center-tpdrc</a></li><li>“Technology is a magnifier of human intent. If you are going in the wrong direction, technology will get you there faster.”</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*99YoWg0O5oFcrSJzpsYliQ.jpeg" /><figcaption>CSIS Cyber Disrupt 2017 Keynote Panel: The Future of Security and Privacy on the Internet</figcaption></figure><p><strong>Cyber Disrupt 2017, Center for Strategic International Studies</strong><br><strong>Matt Hartley, Director, ForeScout Technologies</strong></p><ul><li>Among the emerging threats in cybersecurity is the proliferation of vulnerabilities in the “Internet Of Things” devices. There is no single point of contact to report and patch. There are more IoT devices on government networks than laptops and desktops, combined.</li></ul><p><strong>Scott Montgomery, Vice President, Chief Technical Strategist, Intel Security</strong></p><ul><li>Emerging threats include drone hijacking and using machine learning for offensive cyber operations.</li></ul><p><strong>Nuala O’Connor, President and CEO, Center for Democracy and Technology</strong></p><ul><li>Expectation of privacy: individuals have the right to know where their data is going. Triangulation of data about you when parcelled together from devices in one’s home is already too easy.</li></ul><p><strong>Michael Walker, PM, Information Innovation Office, DARPA</strong></p><ul><li>“Meet the world’s first automated cyber defense system”</li><li>Cyber Grand Challenge asked “what would it take to create an AI that would be able to make decisions entirely without human being?”</li><li>2013 Challenge to the AI: take a look at the code you’ve not seen, analyze without hints, find 0 days, and path by yourself inside a small email program with a difficult state machine — “the everest of program analysis”</li><li>AI was able to find the crashing input, to take control of the crashing system, scrolled the file system, scanned your file upload and analyzed it for vulnerabilities. One AI discovered 0 day, exploited it, while the opponent AI detected the exploit and patched it.</li><li>Caveat — this was a small binary of 50 lines of code. Nevertheless, the autonomous analysis and patching concept has been proven.<br><a href="http://repo.cybergrandchallenge.com/CFE">http://repo.cybergrandchallenge.com/CFE</a></li></ul><p><strong>Beneficiary Discovery Summary</strong></p><p><strong>K., AWG Dog Squadron</strong></p><p>Key Takeaways</p><p>· <strong>Find a partner</strong>: 95–98% of RFP’s have interested vendors. The procurement process does not favor small business. A partnership with an established entity would give us a leg up.</p><p>· <strong>Generating interest</strong>: If we build an image recognition tool that works, then people will be interested. He is not as concerned with finding an end user than demonstrating that the technology works.</p><p>· <strong>Broad customer discovery</strong>: While there are needs internal to AWG that we may be addressing, there will be many more customers from across DOD. Even if AWG doesn’t buy the product themselves they can be advocates for us, if it works.</p><p><strong>Omelas Product Demo: E. and B., Omelas and Z., M., and D., AWG Dog Squadron</strong></p><p>Key Takeaways</p><p>· <strong>Customer Bias</strong>: Army leadership is drawn from the infantry, there product focus is always on the trigger puller, AWG has begun taking those infantrymen and placing them in roles like finance and cyber because they can best understand how to use those tools to enhance the warfighter.</p><p>· <strong>Countering Violent Extremism</strong>: The field of CVE, like social media, is relatively young. There isn’t a clear methodology for proving the effectiveness of a tool in preventing extremism.</p><p>· <strong>People Problems</strong>: In theory the Army should be able to push information down from the Division level. In practice, each commander has their own preferences for how to control the flow of data. This is important in determining how to deploy Spyglass.</p><p>· <strong>Winning Quickly</strong>: The quickest way to win is to have a successful product demonstration in an Army exercise. This is how AWG gets most of their new toys.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=937bd773fee" width="1" height="1" alt=""><hr><p><a href="https://medium.com/spyglass/the-quickest-way-to-win-937bd773fee">The Quickest Way to Win</a> was originally published in <a href="https://medium.com/spyglass">Spyglass</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[“Data is the next water”]]></title>
            <link>https://medium.com/spyglass/data-is-the-next-water-5fe25cfb70b6?source=rss----cfbb6f8dc674---4</link>
            <guid isPermaLink="false">https://medium.com/p/5fe25cfb70b6</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[innovation]]></category>
            <category><![CDATA[image-recognition]]></category>
            <dc:creator><![CDATA[Katerina Sedova]]></dc:creator>
            <pubDate>Sun, 19 Mar 2017 20:00:41 GMT</pubDate>
            <atom:updated>2017-03-26T22:28:57.153Z</atom:updated>
            <content:encoded><![CDATA[<p>This week we began to peek at the horizon and get a glimpse of the future for Spyglass, beyond the H4D class. We built on our work last week exploring buy-in and the famously complex DOD procurement process. A key part of our beneficiary discovery this week was to understand how the AWG acquires technology. Our second goal was to hone in on a single use case where we can add value with a meaningful solution, while differentiating ourselves in a crowded social media data analytics space.</p><p>As one of our interviewees pointed out, “Data is the next water” and there are 3800 companies working in this space already. Throughout the H4D journey, our beneficiary and tech discovery surfaced a number of aspects that are critical to situational awareness in semi-permissive environments, from text analysis of directly threatening posts to atmospherics and general sentiment analysis. This week was no exception. Our interviews further reinforced what we have been learning: nuanced text analysis is hard, a lot of brilliant people are working on it, it is already a crowded space, and there are companies, like Babel Street, who do it exceptionally well in 250 languages. As we touched on in our last week’s post, we continue to look for points of integration to augment our current text analysis capability. We are also making progress on the threat classification front. Thanks to our new friends at Babel Street, this week we received information that we could incorporate into our engine, which allows us to continue our progress on the threat classification. We also got a lead on a list of threatening code words in English that we can translate and use to train Spyglass algorithm. We continue hunting for an expert-tagged training dataset of threatening and non-threatening tweets and tracking one down continues to be a challenge.</p><p>Moreover, we have made advances in image recognition and we believe that this analysis is one way Spyglass can differentiate itself. If we can identify pictures of US personnel posted on social media, analyze the associated text for sentiment or threat terms, evaluate the source of the tweet, and flag an alert, we could add value in the specific scenario: alerting US personnel to potentially being targets of crowd-sourced attacks. For example, we could provide the capability for AWG to train our algorithm on pictures of specific individuals who are about to deploy or already operating in the semi-permissive environments and a means to manage it as people enter and leave the service.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*OhIQx3R51XxLiWUH7qQ3xQ.png" /><figcaption>OpenFace neural network training on Steve Blank</figcaption></figure><p>To explore this hypothesis, this week Jose started playing with open source technology called OpenFace, which provides face recognition with deep neural networks. With any luck, we will have something relevant to demo in class on Monday, so tune in!</p><p>Finally, much of our learning this week came from getting out of the building to visit Haystax Technology, which specializes in monitoring security threats for large events, municipalities, and companies. Jose and Katya joined the Haystax team at their regular “Lunch and Learn” meeting, where the whole company takes a few hours over lunch to meet budding entrepreneurs, to learn about the problems they are solving, and to find ways to help. To make the most out the visit, we did a presentation to the whole Haystax team — the first such presentation for us outside of class. One of the key points of feedback echoed what we have been learning: image recognition could be a point of differentiation for us, if Spyglass could do this well.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*KKTzYVawjOxWO55xjdAM6w.jpeg" /><figcaption>With some of the executive staff at Haystax Technology</figcaption></figure><p>Bryan Ware and his executive team were also quite generous in offering us to play with their product interface, Constellation, and the Haystax threat streams application, which overlays a stream of tweets over a map in a particular location. Over the next week, we will investigate if we can use the Constellation system without going through Haystax algorithm. If this is possible, we can have a UI solution for the MVP and let Jordan do what he does best — work on the machine learning algorithm.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*TtoslMWGa3zrnNDE9Bgxwg.png" /><figcaption>Constellation map and threat streams apps</figcaption></figure><p>In addition, we received a good summary of the Twitter terms of use and contacts at Twitter to follow-up with on GNIP data licensing. Another key takeaway from our visit to Haystax came organically as we touched on possible dual use of Spyglass. Based on our interviews at Haystax, we believe the market opportunity for a pure social media solution for corporate security customers may be shaky, as companies are typically unwilling to pay more than $10K per license annually. This confirms our direction so far to not actively seek dual use scenario as a primary goal, and focus on delivering a great solution to AWG.</p><p>The summaries below detail the additional takeaways from our interviews this week.</p><p><strong>A., Seafront Analytics</strong></p><ul><li><strong>Making technology stand out:</strong> There’s a ton of technology out there, but it takes too long to go from “zero to hero.” If we could develop a lightweight tool with a 30–60 minute learning curve that would provide value.</li><li><strong>Operational focus:</strong> Folks, especially on the conventional side don’t have a lot of money to spend, especially on non-operational things. They would rather try to get more value out of the tools they already have rather than buy new ones.</li><li><strong>Uplifting Analysts:</strong> Most clients want tools that will empower their analysts, not provide answers for them (how do we distill 200,000 posts to 200, and let the analyst provide the analytical framework on the remaining data?)</li></ul><p><strong>K., The Penrose Corporation</strong></p><ul><li><strong>Cost problems:</strong> In spite of the multitude of technology products in the OSINT space, no one introduced a cost structure that suits the DOD’s preference for enterprise solutions.</li><li><strong>Managing client expectations:</strong> Focus on what is doable given constraints (time, money, technology) and communicate that with stakeholders. Articulate what we can achieve now and where we could get given X amount of time. Get a commander comfortable with our interim solution and get to the point where they tell us “we would he a customer if.”</li><li><strong>Veracity of information:</strong> If in the conduct of normal duties, you are able to push social media information to soldiers who are in a position to corroborate that information, thus lending veracity to an account, that is a value add. What is important is that we can explain to a commander the risks and the flaws of our model, and the math that powers it will lend credibility to our efforts.</li></ul><p><strong>J., Researcher at the National Institute from Standards and Technology</strong></p><ul><li>Expert on Data mining tweets, multilingual and geographic information. Analysts won’t be able to provide real tweets, but <strong>she can recreate 25 fictitious tweets that are threatening. </strong>These tweets could be used for language analysis.</li><li><strong>Threats might use coded language</strong> so more important than language analysis is to use threat + coded language analysis: identifying in which context eggplant means a vegetable or a bomb.</li><li><strong>Some methods used for geo-inferencing are:</strong> social network analysis (who is talking to whom), looking for locations that are mentioned in tweets, particular unique locations or monuments only found in one place (pilgrimage sites), events that people are talking about, dialects spoken in different places, which can help identify the country.</li><li>If we go the route of image recognition, we could use night, day, dawn and dusk to infer the location of outdoors pictures.</li></ul><p><strong>J., Sr. Operation Specialist, and K., Sr. Military Adviser, AWG</strong></p><ul><li><strong>The Technology Readiness Level</strong> is a scale used by the DOD to categorize new technology that is not currently implemented to define whether money is going to R&amp;D or Purchases.</li><li>On levels 1 to 3, DOD usually does not invest money but expect this to be done as part of the Industry Research and Development. DOD will consider investing money after a prototype is presented through a solicitation process.</li><li><strong>There are two options for purchasing a new technology: </strong>a procurement process for wide distribution and a fielded purchase for limited edition distribution. Procurement means that not only you can buy, but also you can field it for worldwide use with safety confirmations to have soldiers use on the field. Procurement is a longer and more complicated process.</li><li><strong>If we provide a solution by the end of the class and AWG likes it, they will have to do further due diligence to decide how to proceed.</strong> If AWG commander accepts this solution, AWG could write the requirements and either sole-sourcing or open for bidding. If it is accepted for army-wide use, AWG would have to present it and require further chain of command guidance. Step 1 is presentation of the final product to the leadership.</li></ul><p><strong>Haystax Technology Site Visit</strong></p><p>B., co-founder and CEO; J., VP of Product Management; C., Director of Product Marketing; R., Director of Product Engineering; P., Solutions Architect, R., CTO; C., Executive VP of Product</p><ul><li>Haystax’ use case: <strong>“I am responsible for security of a very large place, a city or corp. I need to monitor anything that creates a problem”. </strong>Clients include major sporting events and concerts, large municipalities. This includes algorithms for insider threats: human resources, social media, network security, connecting risk into the model.</li><li>The main platform — <strong>the Constellation system</strong> — is a multi-tenant cloud-based platform. It contains a series of apps: Assets (facilities, buildings of interest); Incidents (incident log, users can subscribe to email alerts when new incidents get created); Map (shows all of the data in the system on a map, see your assets, incidents, threat streams [twitter,rss]); Threat Streams (3 column view of real-time streaming data feed analysis from Twitter &amp; RSS). Through the demo license, we can configure our own column views, add our own RSS feeds. We can also post items into the API to populate our own feed. <strong>A Spyglass tenant has been created for us, including sample asset and sample incident, which we can explore.</strong> The “how to” deep dive into the system is available upon request.</li><li><strong>Training data:</strong> look for examples of social media threats and examples of what that awareness looks like for your beneficiaries — this might uncover untold needs. Try designing 30 threatening tweets and use that as a training data set. They can give us a list of threatening code words in English, which we can potentially translate.</li><li><strong>The use case for global corporate security is there, but limited.</strong> One of their customers during Arab Spring was not getting a full picture from social media, and by the time they started the evacuation, it was under fire. Spyglass could have helped make a decision sooner. However, the commercial market for our project might be limited to big corporations, who will pay at most around $10,000 / year for the service.</li><li><strong>There are many restrictions on the use of Twitter for law enforcement or security.</strong> Use by law enforcement is not permitted except to serve public safety use cases as outlined below. Use of the product to detect potential violent threats against the public is allowable if law enforcement is acting at the behest of the event organizer to provide security. Other law enforcement uses to detect potential violent threats against the public during large public events is allowable ONLY IF: the event has broad public access and impact (e.g. large sporting events, city marathons, concerts, academic or corporate conferences) and the event is not political in nature (e.g. a protest, rally, community organizing meeting). For law enforcement, government, and public safety customers, named entities should <em>not </em>be people; sources (@handles) should <em>not </em>be reported for tweets.</li></ul><p>Overall, it has been a great week of learning. Honing in on a use case is an important milestone for us. We are doubling down on image recognition as a value add to our problem sponsor’s goals and a point of differentiation for Spyglass. Next week we will test the hypothesis that this use case is the right direction.</p><p>For now, Spyglass is a drop in the ocean of data and social media data analytics solutions. Hopefully, one day it can be a formidable tide in this ocean.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5fe25cfb70b6" width="1" height="1" alt=""><hr><p><a href="https://medium.com/spyglass/data-is-the-next-water-5fe25cfb70b6">“Data is the next water”</a> was originally published in <a href="https://medium.com/spyglass">Spyglass</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The million-dollar question: What do we want Spyglass to be?]]></title>
            <link>https://medium.com/spyglass/spyglass-on-rethinking-the-feedback-loop-abfa5e8f856f?source=rss----cfbb6f8dc674---4</link>
            <guid isPermaLink="false">https://medium.com/p/abfa5e8f856f</guid>
            <category><![CDATA[social-media]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[hacking-for-defense]]></category>
            <dc:creator><![CDATA[José Gabriel Andrade]]></dc:creator>
            <pubDate>Sun, 12 Mar 2017 22:01:15 GMT</pubDate>
            <atom:updated>2017-03-12T22:11:12.117Z</atom:updated>
            <content:encoded><![CDATA[<p>We began this week considering how to narrow the scope of our product and focus on a clear value proposition for a single beneficiary. These steps would help us prove our concepts and prototype more clearly. When it comes to open source social media, there are many options for analyzing the data collected: <strong>sentiment analysis</strong> on text in foreign languages, <strong>object recognition</strong> in images, <strong>network analysis</strong> of users and identification of trending topics, to name a few.</p><p>All of these contribute to our goal of enhancing force protection and situtational awareness, albeit in different ways. Our current challenge is to identify the minimum capabilities required to provide value to a single target beneficiary and obtain feedback from those users. We need to decide whom we want our first beta testers to be and which capability is most pressing for them. In a way, we have been trying to answer this question the entire semester, and we had hoped that our beneficiary and tech discovery would naturally lead us to an obvious conclusion. Now that we have concluded Week 7 of Hacking for Defense (H4D) without an obvious answer, it seems that we as a team might have to <strong>make some choices</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*Mx1ZJrC9uoLsdt3d." /><figcaption><em>One of the shire on Palantir offices</em></figcaption></figure><p>Our first hypothesis for this week postulated that there are training datasets with classified tweets by threat level that we could use for training a machine learning (ML) algorithm. We were unable to obtain a threat-based dataset due to clearance and procedural limitations, but we have a number of leads on alternative datasets. We looked into sentiment analysis as a proxy for threat analysis to prove the classification capability of our algoirthm. Jordan built a prototype feature of sentiment classification in Arabic.</p><p>For our starting point, we used a dataset of <a href="https://archive.ics.uci.edu/ml/datasets/Twitter+Data+set+for+Arabic+Sentiment+Analysis">sentiment-tagged Arabic tweets</a> from UC Irvine (UCI). The model was trained using a support vector. The current accuracy level is ~76%, though it is somewhat biased towards negative sentiment. According to their paper, the UCI researchers improved the accuracy of their model to 84% by incorporating a stemming algorithm that de-conjugates the Arabic text. While doable, this improvement will requires extensive Arabic knowledge and manual labor. Rather than deepen our Arabic analysis capabilities, we are considering incorporating Russian as an additional language.</p><p>Our second hypothesis stated that the primary feature of Spyglass should be a real-time alert system of unusual activity (including threats) to forces on the ground. We found it difficult to validate this hypothesis with current soldiers in the conventional military who have deployed recently. Beneficaries we spoke to either lacked experience in a combat deployment in a semi-permissive environment or had not deployed within the past 3 years. Understandably, it’s extremely difficult to get in touch with folks who are currently deployed abroad.</p><p>As a result, we are considering using AWG Operational Advisors as initial beta testers for our tool with real-time alerting capability. Even though AWG OAs are unlikely to be our target beneficiary, they may be able to provide valuable insights for Spyglass as beta testers. Furthermore, our latest H4D course lecture demonstrated the importance of winning buy-in from the organization. That could start with getting beta testers who are also in a position to demonstrate buy-in and support for Spyglass.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*lreqAp9vx90iYK5S." /><figcaption><em>The current release of Spyglass featuring Arabic sentiment analysis capabilities.</em></figcaption></figure><p>For our tech discovery, we visited one of Babel Street’s offices in Virginia and Palantir’s DC office in Georgetown. Babel Street is doing a very detailed work on understanding language differences and local ontologies. Their lexicon translation database has been trained by linguistics over several years not only on meaning but also on sentiment analysis. Palantir’s Gotham platform is an advanced secure data aggregation and analysis tool. Palantir’s strategy focuses on solving their customer’s specific problems rather than a providing generic software licenses to customers. Our key takeaways from these visits are featured in our interview summaries below.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*oNLoVJ4kV1v0fYR2." /><figcaption><em>Meeting Brendan and his team in one of Babel Street’s offices.</em></figcaption></figure><p><strong>Interview Summaries:</strong></p><p><em>Landon Bell. Deployment Strategist, Palantir. Formerly Navy Special Forces.</em></p><ul><li><strong>Getting users engaged:</strong> “The biggest problem we have with users is when they are disenfranchised from the outcome.” If we want users to buy-in, they need to feel connected to the outcomes we are trying to achieve.</li><li><strong>Improving, not replacing, humans:</strong> Their product is designed to make humans more powerful, not automate their jobs away. The higher the stakes of your mission the more you want to optimize away from the machine making a decision. Spyglass adheres to the same philosophy.</li><li><strong>Advice moving forward:</strong> Focus on capabilities, not features. The problem is almost impossible to narrow down too much, so try to create as specific a use case as possible to demonstrate the value of our product. Get it into those people’s hands and start testing ASAP.</li><li><strong>Buying in:</strong> Who needs us the most? Who is feeling the pain because they lack the capabilities we offer? Who is the decision maker who can allocate this money? We need to answer these questions if we realistically want to deploy our product.</li></ul><p><em>A., CPT. Army Cybercom, Planner for CEMA Support to Corps and Below.</em></p><ul><li><strong>Unlikely NTC can incorporate Spyglass in upcoming BCT experiments in April. </strong>We’re running cyber/social media experiments at the National Training Center (NTC) through a closed Insurgent Communications Network (ICN). The ICN runs on a closed fiber ring that applications over the open web can’t access.</li><li>During the tests, our Information Warfare cell will replicate the <strong>opposition force</strong> posting things on social media (Facebook, Twitter) and the open web (eBay, etc.). We also have neutral people replicating the population posting things on those sites.</li><li>OSINT teams scrape all of this information sent over the ICN, provide tippers to the brigade, and tell the BCTs what to be on the lookout for.</li></ul><p><em>G., CPT. Regimental Irregular Warfare Operations Officer, National Training Center (NTC).</em></p><ul><li><strong>The ICN infrastructure and hardware at NTC is clunky and not that accessible. “</strong>You have to be physically located at NTC to be in the ICN.”</li><li><strong>NTC is upgrading to a new closed network called ION over the next 6–8 months. </strong>With the NIPR backbone, ION will be remotely accessible and accessible to people outside the training centers. ION will be maintained by Army Cybercom, whereas ICN is maintained by NTC’s in-house communications shop.</li><li><strong>NTC replicates the environment of countries all over the world to provide a framework for its roleplaying simulations.</strong> These show how conflicts would evolve based on factors that are introduced. BCT rotations are <strong>10 days</strong> long, so sometimes developments that might take months or years in the real world are shortened.</li></ul><p><em>Michael Lumpkin, former Special Envoy and Coordinator of the Global Engagement Center (GEC) at the U.S. Department of State.</em></p><ul><li><strong>GEC’s strategy is focused on partner-driven messaging and data analytics for effective counter-messaging: </strong>GEC is an interagency entity, housed at the State Department and charged with coordinating U.S. counterterrorism messaging to foreign audiences. Primary mission is to counter messaging of violent extremism from terrorist organizations including ISIL.</li><li><strong>GEC has a chief privacy officer to carefully ensure compliance:</strong> GEC was given a “privacy carve-out” of a strict interpretation of the 1974 Privacy Act for regulating data collection through the National Defense Authorization Act of 2015–2016. GEC also established a social media fusion cell with representatives from partner nations in order to abide by privacy laws of those nations.</li><li><strong>Obstacles to effective social media counter-messaging are organizational resistance to innovation and adoption of new tools</strong>. DOD suffers from the “not invented here” bias: if it’s not created in a given organization, it’s not good enough. A way to change is through high-level buy-in from the decision makers: “breaking the rules, but having a plan.” How does Spyglass fit into the plan?</li><li><strong>Keys to effective buy-in and tool adoption: modularity, adaptability, ease of integration with existing system to decrease barrier to adoption. </strong>There are 3800 different tools focusing on social media analytics — know your unique value add and ensure that there is funding for longer-term adoption.</li></ul><p><em>Brendan Huff, Andy Pessaud, Jeff Chapman of Babel Street.</em></p><ul><li><strong>Babel Street is a powerful text analytics platform that enables search of open source intelligence</strong>, supports 55,000 news sources and blogs, 30 social media sources/channels, and the Dark Web. It breaks the language barrier by enabling search of terms in English and traversing 250 languages.</li><li><strong>Babel Street’s ontology uses</strong> <strong>an entity </strong>— not a word — <strong>as the base unit.</strong> It understands entities across languages, vernaculars, slang, and even misspellings. This level of nuance has been attained through 10 years of human translation and curation of machine learning algorithm results by linguists.</li><li><strong>Babel Street has many advanced features:</strong> granular sentiment detection, conversation volume tracking, and event mapping, to name a few. There is threat detection capability and premium alerts that can be set up for customers. Babel Street is working on building a mobile app and integration with mobile platforms.</li><li><strong>Compliance with privacy regulations and Terms of Service (ToS) contracts are paramount for tools that analyze data. </strong>Babel Street has a privacy officer to ensure full compliance with laws and ToSs of social networks and data providers, such as Twitter. Twitter is not shy about terminating access to their data to entities that violate its ToS, a fate that has affected some of Babel Street’s competitors. A GNIP license must be purchased by anyone commercializing social media data analysis — does Spyglass need this?</li><li><strong>“Data is the next water.”</strong> Everybody wants to get into the data analytics game. Few actors “own” and provide data through licensing its use to data analytics companies. Opportunities exist for Spyglass to “plug-in” to Babel Street to use their language capability, but it runs into potential data ownership problems.</li></ul><p><em>Sgt. Jon Gillis, Marine Corps Warfighting Lab.</em></p><ul><li>The biggest challenge to adoption of Spyglass will happen at handoff: can the DoD/military/etc. handle deployment of our prototype and productionizing on their own systems? As it stands today, probably not.</li><li>ATAK is not suitable for large data transfer; Nett Warrior is 3–5 years away.</li><li>One way to work around the DOD’s acquisition problems is to bundle cheap and effective solutions along with other acquisitions.</li></ul><p><em>B., SPC. U.S. Army Cyber Operations Specialist.</em></p><ul><li><strong>Cultural impediments:</strong> Cyber is sometimes viewed as “the other army.” There is a leadership culture that doesn’t always understand the mission. Developing understanding of how our product affects conventional operations is critical to getting buy-in.</li><li><strong>Not a key beneficiary: </strong>Most guys in his area will not be deployed and are assigned to support different central commands as needed. Many work in a highly-classified environment and will not likely benefit directly from our product.</li><li><strong>Geoinferencing workaround:</strong> Even if only 2% of social media is geotagged, that’s still a lot. Over time we could perform trend analysis and use those observations to inference non-geotagged tweets that fit the pattern.</li></ul><p><em>G., Special Agent, Internal Security Division, Department of Homeland Security.</em></p><ul><li><strong>Buy in HMS would be at the agency level</strong>. The Department of Homeland Security does not integrate all their agencies decision making procedures. In order to get buy in from Customs and Border Patrol, the Secret Service or the Coast Guard it will be necessary to talk to every agency independently.</li><li>Currently the Internal Security Division would be interested in a system that protects its facilities and DHS key personnel in public acts. Currently this targets are considered at low risk. They might be interesting opportunities concerning <strong>Customs and Border Patrol.</strong></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=abfa5e8f856f" width="1" height="1" alt=""><hr><p><a href="https://medium.com/spyglass/spyglass-on-rethinking-the-feedback-loop-abfa5e8f856f">The million-dollar question: What do we want Spyglass to be?</a> was originally published in <a href="https://medium.com/spyglass">Spyglass</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Spyglass is live with translation, image recognition]]></title>
            <link>https://medium.com/spyglass/spyglass-is-live-on-aws-3c2d18b65ff?source=rss----cfbb6f8dc674---4</link>
            <guid isPermaLink="false">https://medium.com/p/3c2d18b65ff</guid>
            <category><![CDATA[hacking-for-defense]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[innovation]]></category>
            <category><![CDATA[social-media]]></category>
            <category><![CDATA[image-recognition]]></category>
            <dc:creator><![CDATA[Chloe Krawczyk]]></dc:creator>
            <pubDate>Mon, 27 Feb 2017 00:11:31 GMT</pubDate>
            <atom:updated>2021-01-27T19:29:46.873Z</atom:updated>
            <content:encoded><![CDATA[<p>We’ve reached the midpoint of the semester, and Spyglass is officially live on the cloud through Amazon Web Services (AWS). We’ve implemented real-time machine translation of Arabic tweets into English and built a working image classification tool using a pre-trained <a href="http://playground.tensorflow.org/">neural network</a> in <a href="https://www.tensorflow.org/">TensorFlow</a>. Furthermore, we understand how platoon leaders would prioritize features in a tool that has the capability to index, aggregate, and analyze open source social media information.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*QWDnvJ7w7_uHeCcqzsFHRQ.png" /><figcaption>Spyglass 0.2 live on AWS with translation capability powered by Bing’s open source API.</figcaption></figure><p>We have a site visit to Babel Street planned for next week. We hope to access a tool that can plug into Spyglass to provide better translation than our current open source solution. This is especially important due to the challenges of understanding a foreign language in context given sarcasm, humor, slang, code words, and so forth.</p><p>While Jordan worked to deploy Spyglass to the cloud and implement real-time translation, Jose coded a program to classify images as a military vehicle (“MRAV”), U.S. soldier, civilian (“CIV”), or non-military vehicle (“NMV”). The neural network was trained on a dataset that Jose and Max put together by downloading public images from the Internet. Spyglass can accurately classify images into those four categories; an image of an object in one of the categories consistently showed as &gt;90% probability in the correct category.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/914/1*nwP2Zxaekt4LJ1mC0aUFgA.png" /><figcaption>Spyglass successfully identifying a picture of three U.S. soldiers facing away from the camera.</figcaption></figure><p>A potential issue to explore is the algorithm can be confused by multiple objects in different categories in the same photo. When we tested it with an image of a car and a military vehicle, the algorithm could not tell us both objects were present. Nevertheless, the current functionality serves our basic purpose by providing enhanced awareness of possible threats targeting U.S. personnel or equipment when integrated with a threat classification algorithm. Moreover, if we are able to gain access to actual pictures of U.S. personnel or equipment for a training dataset, the algorithm could be even more precise.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/909/1*C1cKVEQvmjilGdi6bCyPLQ.png" /><figcaption>Spyglass failing to identify a military vehicle when another category object (NMV) is present in the same photo.</figcaption></figure><p>At our team meeting yesterday, we came up with the product roadmap for the remainder of the semester:</p><ol><li><strong>Live Capability:</strong> Enable persistent scraping and storage in real time.</li><li><strong>Threat Classification and Real-Time Alerts:</strong> Build a machine learning algorithm on a training dataset (of tweets) already tagged by human analysts (e.g., S-2s, S-3s) as threatening or not (or whatever form of classification is in use by the DOD). This algorithm will serve as the basis for determining what real-time alerts should be sent to provide users of potential threats and unusual activity in a specific geographic area.</li><li><strong>Atmospherics Analysis: </strong>Create algorithms to analyze tweets within a specified geographic area and timeframe for trending topics relevant to atmospherics.</li><li><strong>Analyst Integration:</strong> Build a dashboard view to allow analysts (e.g., S-2s, S-3s who tend to have persistent Internet connection and sit at the battalion or above level) to plug into Spyglass. They can do searches for tweets, perform their own analysis, and push information down to platoon leaders as necessary.</li><li><strong>Customized Personas: </strong>Build out a set of filters/algorithms to present relevant and timely data to each user type on Spyglass.</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*QZLqS1TZxEiZE-jH2943YA.jpeg" /><figcaption>Jordan and Jose discussing the code behind our image recognition program.</figcaption></figure><p>We have two goals for next week. Our first hypothesis is that there is a training dataset of tweets classified by threat, which we can use to train a machine learning algorithm for Spyglass. We will talk to people in the DOD and private sector (e.g., defense contractors) to see if anyone has a training dataset available for us to use. If we cannot access a training dataset, our ability to provide real-time alerts of unusual activity will be constricted. Our second hypothesis is that the primary feature of Spyglass should be a real-time alert system of unusual activity (including threats) to forces on the ground. This workflow will be a continuation of the customer discovery from this week with contacts at lower echelons and non-commissed officers (NCOs), from whom we’d like to get feedback regarding what Spyglass’s purpose, design, and features should be.</p><p>Tomorrow, we plan to perform a live demostration of Spyglass at Georgetown Hacking for Defense class! Join us at 6:30 p.m. on February 27 in St Mary’s Hall, room 110.</p><h4><strong>Beneficiary Discovery:</strong></h4><p>This week, we interviewed four platoon leaders who deployed to Afghanistan between 2009 and 2013. We started each beneficiary interview with a standard question: <em>“What information do you want from social media?”</em> Their respective answers are in the first bullet of their interview summaries. Interestingly, three of the four beneficiaries’ immediate response was <strong>atmospherics.</strong> We also asked each person to rank five features in order of priority; their answers are in the second bullet.</p><p><strong>Atmospherics is relatively straightforward for us to implement from a technical perspective. </strong>We already have the capability to scrape geotagged tweets within a geographic area; we can perform analysis on those tweets to identify trends and important topics to the people who tweeted. Moreover, this information can be transmitted to platoons on the ground directly to immediately improve their posturing and messaging. There’s less at stake if we get something wrong in atmospherics than in threat alerts. <strong>However, we wonder if atmospherics are relevant to ground forces outside of a counterinsurgency or stability operation.</strong> It makes sense that between 2009 and 2013, when the U.S. military was “shoulder-to-shoulder” with the local populace, that understanding the population was important, but we’d like to interview more soldiers who have served in a platoon after 2014 to verify what the current priorities are.</p><p>Finally, our interviews show that soldiers on the ground always want to know about threats. However, there is some doubt as to whether Spyglass can provide accurate analysis and identification of threats through an algorithm applied to social media data. The four (former) platoon leaders and one S-2 we interviewed this week all felt that <strong>any intelligence picked up by Spyglass should be vetted by a human analyst before being transmitted to platoons</strong>. This crystallized our decision to add a dashboard layer for intel analysts to the product roadmap. If our algorithm identifies threatening tweets, alerts an S-2 for verification, and then transmits an alert to a user on the ground subject to the S-2’s approval, <strong>Spyglass could still save a significant amount of time and energy in communicating real-time intelligence to forces on the ground.</strong></p><p><strong>J., Captain, Army Platoon Leader (Afghanistan, 2009–11)</strong></p><ul><li><strong>Social media is useful for “real-time atmospherics.”</strong> Army has historical atmospherics of an area, but not the day before or day of. “As an infantry commander, I want to know dynamic of what we’re going into, what the local populace is happy or up in arms about.”</li><li><strong>Prioritization of Spyglass features: </strong>1) atmospherics; 2) image recognition of U.S. personnel and equipment; 3) real-time threat alerts; 4) identification of anomalies in tweeting traffic from baseline; 5) map view of social media activity.</li><li><strong>Threat analysis belongs with the S-2:</strong> “As someone on the ground, you obviously always want as much information about as many threats as possible. But that’s the deciphering that the S-2 does, because people aren’t going to announce real threatening intentions publicly on social media.”</li><li><strong>Wants a five-minute version of Spyglass to use on missions. </strong>“When I’m not on a mission, I might spend up to 30–45 minutes analyzing and digging into the raw data behind the analysis provided by Spyglass.”</li><li><strong>Real-time atmospherics would improve messaging and posturing of platoons that visit towns and villages.</strong> “My dream would be able to look at Spyglass, know what people are talking about, and then to walk up to a village elder and tell them, ‘Hey, I know there’s a lot of chatter around…’.”</li><li><strong>Mapping and raw tweets are less important. </strong>“Mapping is more useful for the intel analysts to understand networks of people posting on social media.” “Give the raw tweets to the analysts.”</li></ul><p><strong>K., Sergeant, Marine Corps Platoon Leader (Afghanistan, 2013)</strong></p><ul><li><strong>Social media is useful for “gathering atmospherics inside the operating bubble.”</strong> Tracking and monitoring feeds would help Marines find a “lightning strike,” or what’s trending in an area. There is a tremendous source of information on what influences a local populace that can impact what goes on on the ground.</li><li><strong>Prioritization of Spyglass features: </strong>1) real-time threat alerts (but only if vetted properly); 2) image recognition of U.S. personnel and equipment; 3) atmospherics; 4) identification of anomalies in tweeting traffic from baseline; 5) map view of social media activity.</li><li><strong>Information about threats is not useful if it’s not vetted by intel analysts. </strong>“I don’t think a social media tool can bypass a human intel cell that ensures the information from the algorithm is accurate, and a perceived threat is actually a threat.”</li><li><strong>To implement real-time threat alerts, Spyglass’s priority should be figuring out how to vet intelligence more quickly and efficiently. </strong>Capability to identify threats from social media already exists in the Marines, but the vetting process could be speeded up. “We have complete units inside Marine Corps intelligence community where all they do is monitor social media, video and audio feeds (cellphones, landlines), iCom chatter (short-band radio waves), etc.”</li></ul><p><strong>A., Captain, Rifle Platoon Leader (Afghanistan, 2011–12)</strong></p><ul><li><strong>Social media is useful for “atmospherics.” </strong>Gauging the population’s reaction to operations is useful wherever a soldier goes. “If I had access to social media [in 2011], I would’ve wanted to know positives and negative indicators from the population — e.g., I want to know if there are tweets like, ‘There’s a firefight outside my door; I hate the Americans’ or ‘I love the Americans.’”</li><li><strong>Prioritization of Spyglass features: </strong>1) atmospherics <em>(“I can do a lot with that”)</em>; 2) image recognition of U.S. personnel and equipment; 3) real-time threat alerts; 4) identification of anomalies in tweeting traffic from baseline; 5) map view of social media activity.</li><li><strong>Threat analysis from Spyglass is not as useful because it’s less reliable:</strong> “You can put some stock in threat alerts, but I don’t think it’s the most important thing, because not everyone is going to tweet that they’re about to blow something up. They generally tweet it afterwards.”</li><li><strong>Wants a 15-minute version of Spyglass; if time, would spend up to an hour on Spyglass: </strong>“I could probably spend an hour looking at Spyglass at the end of each day to see what was trending and understand our impact of the operations we were doing that day. In a more kinetic environment, I might only look at Spyglass if I had 15 minutes until the next patrol. But I don’t think social media should be designed for a kinetic fight.”</li><li><strong>Atmospherics are useful in a counterinsurgency </strong>“when you have to win over the population’s trust and instill faith in their local government. You’ll be more effective if you can monitor their social media and see what people are complaining about and tease out trends about what’s pissing people off.”</li></ul><p><strong>D., Lieutenant, Army Combat Engineering Platoon Leader (Afghanistan, 2010–11)</strong></p><ul><li><strong>Social media is useful for situational awareness. </strong>“Platoon commanders will benefit the most from the real-time social media data if it can help answer the question: Are we in the wrong area? Did we target the wrong house based on the intelligence? Need this tools to help validate what they are seeing around them.”</li><li>Platoon commanders are empowered to execute the mission set forth by the company commander, but have command and control in the absence of different orders to react to the situation for defensive purposes. For offensive operations, the intel provided by the social media tool would need to be validated by S-2 intel cell.</li><li><strong>NCOs are not the ultimate beneficiary of social media information. </strong>Knowledge management and data management occurs in the operation center. It can occur at the company level if they are remote, but predominantly at the battalion and brigade level. S-2s would be tasked with intel analysis, while platoon leader would call back to them for support if they see any suspicious activity.</li><li><strong>Primary beneficiary should be the intel chief at the operation center who consolidate the intel from different sources and push it down to the platoon level.</strong> Secondary beneficiary could be civil affairs units (e.g., s-7s, S-9s) who track how U.S. actions are playing out in the local population.</li></ul><p><strong>A., Captain, Army Cyber Command (ARCYBER), CEMA Support to Corps and Below (CSCB)</strong></p><ul><li><strong>The CEMA (Cyber Electromagnetic Activities) cell sits at the brigade level but works to bring everything down to the ground fighter.</strong> “Right now everything we do is at the brigade level; they task organize to the battalion, and then they disseminate info as needed to each company based on mission.”</li><li><strong>Threat alerting might be more useful for present/future Army priorities. </strong>A tool that provides atmospherics sounds most useful in a counterinsurgency fight where there are stability operations and key leader engagements (KLEs). My impression is that the Army combat training centers have shifted from counterinsurgency “back to the basics” of conventional warfare.</li><li>However, concerned about “the ability of someone down at the platoon to digest that info [threat alerts] in meaningful manner for it to be actionable” without passing it through the upper echelons.</li><li><strong>CEMA is exploring how to support Open Source Analysts at the brigade and division level; Spyglass could make a difference in their workflow. </strong>“They go through and scrape social media and output reports. What they do is fairly manual.”</li></ul><p><strong>A., S-2, Army (Afghanistan)</strong></p><ul><li>Commanders don’t really need access to raw social media data unfiltered by S-2 analysts. In fact, it could be dangerous to provide access to this kind of information to soldiers on the ground who are not trained in social media analysis. Building a “commander-centric” tool does not necessarily mean that the tool should be used by commanders on the ground.</li><li>S-2s don’t need a new tool; there are plenty of great tools out there. S-2s need the improved capability to use existing tools in classified networks more easily.</li></ul><h4><strong>Tech Discovery:</strong></h4><p><strong>J., Sergeant, and D., Staff Sergeant, Marine Corps Integrated Task Force</strong></p><ul><li><strong>ATAK and Nett Warrior are potential conduits for Spyglass. </strong>ATAK is low latency and can only transmit text alerts rapidly. Nett Warrior can accomdate more but is 3–5 years away.</li><li><strong>Ground forces need timely alerts when something threatening happens. </strong>They would likely be inundated if they had to look through dashboards and raw tweets.</li><li>A tweet scraper, classifier, and alerting tool will solve the issues of ground forces and can have low bandwidth. A dashboard layer on top of that can serve the commanders and S-2s.</li><li>Analysts already have solutions; ground forces are missing theirs.</li></ul><p><strong>C., Quantitative Crisis Response (QCR), DARPA</strong></p><ul><li>QCR provides the “population sentiment” heatmap solution to the pain-point of needing to understand how US personnel in advise and assist roles are impacting the sentiment of the local populace.</li><li>QCR works on different aspects of social media analysis for information operations and operational security. Project is on a rapid transition cycle — built in 1.5 years and transitioned to DoD — and they are soliciting real-time feedback from users in the field as development is proceeding.</li><li>Goal of QCR is to measure the impact of information operations through defined research questions. QCR collects data from Twitter, Facebook, Telegram (with YouTube and VK in progress), extracts data in the “who did what to whom” format, and codifies data from competitive to friendly.</li><li>QCR allows for the tracking of information campaigns and measures performance and effect, through time series view of sympathy vs the baseline, visual representation of inferred emotion and tone, volume, heatmap level of the conversation (detecting spikes in conversation from the usual user base with new accounts joining)</li><li>QCR exhibits the feedback loop capability to answer “What affect did I have?” question, rather than merely “Did I have an effect?”, “What audience did I reach?” vs “did I reach my target audience?”</li></ul><p><strong>J., Private Sector (The Penrose Corporation)</strong></p><ul><li>The ability to triage data without sending it on to a linguist is absolutely significant. Babel Street is close to this, but the user often has to leave the secure platform if s/he wants to inspect a document more closely.</li><li>Boolean filters of key words are a critical component of his workflow, but it is a labor-intensive process. The ability to automate those tasks would be a huge value add.</li><li>Platoon leaders care about the tactical situation and want a granular level of information, but the are burdened with many other responsibilities. On the other hand, analysts don’t often have an operational background, so how do we structure Spyglass to leverage the skills of both without overburdening the end user?</li><li>If Spyglass can be built to be one percent better than the Army’s current capabilities as an alert platform/social media gathering tool and demonstrate a roadmap for improving the product in just 10 weeks, that would be a tremendous success.</li></ul><p><strong>In summary, through next week’s interviews, we hope to understand:</strong> 1) whether atmospherics or real-time threat alerts are a greater priority for current soldiers deployed in semi-permissive environments, and 2) whether we can train an algorithm on a social media dataset that has been analyzed by S-2s or S-3s. If the latter is not possible, it would be significantly harder for us to build a tool this semester that accurately identifies threats and unusual activity from social media data. In that case, we may have to ask whether we can build our own training dataset or to pivot to a different feature (e.g., atmospherics) as the primary use case of Spyglass.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3c2d18b65ff" width="1" height="1" alt=""><hr><p><a href="https://medium.com/spyglass/spyglass-is-live-on-aws-3c2d18b65ff">Spyglass is live with translation, image recognition</a> was originally published in <a href="https://medium.com/spyglass">Spyglass</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Learning by building]]></title>
            <link>https://medium.com/spyglass/learning-by-building-e548bae50361?source=rss----cfbb6f8dc674---4</link>
            <guid isPermaLink="false">https://medium.com/p/e548bae50361</guid>
            <category><![CDATA[hacking-for-defense]]></category>
            <dc:creator><![CDATA[Max Weintraub]]></dc:creator>
            <pubDate>Sun, 12 Feb 2017 21:32:10 GMT</pubDate>
            <atom:updated>2017-02-27T02:27:27.965Z</atom:updated>
            <content:encoded><![CDATA[<p>Building off our momentum from last week, we continued to conduct beneficiary interviews with the goal of understanding the needs of company commanders and platoon leaders for situational awareness. We also continued technology discovery to understand existing capabilities on the market, including a trip to the Voyager Labs offices in Reston, VA for a live product demonstration. Most exciting, our tech whiz, Jordan, has built the first prototype of the Spyglass platform: sneak preview to follow.</p><p>The true power of lean startup methodology is to distill the needs of beneficiaries and end-users to determine our priorities for features required to make an impact in their workflows. We brainstormed ideas in the morning…</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*fHyMTk1eH3LRo50TyGao0g.jpeg" /><figcaption>Team Spyglass brainstorming session.</figcaption></figure><p>…and built a working prototype of a Twitter scraper in time for dinner.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*22tDicH9ewqPaxNl4TLphQ.png" /><figcaption>This is what raw tweets look like!</figcaption></figure><p>Over the course of the next 12 weeks, we hope to transform the ideas generated on that whiteboard into a product that can be deployed to the field and refined based on the feedback from actual soldiers. Each week, we create a list of hypotheses to get a better sense of which features should be prioritized. <strong>This week our hypotheses were:</strong></p><ol><li>Company commanders or platoon leaders want information about specific events from social media.</li><li>Company commanders or platoon leaders care more about the content of tweets and the time/location it was posted than trends or interactions.</li><li>Company commanders or platoon leaders don’t need tweets mapped in a MVP.</li><li>Company commanders or platoon leaders use ATAK/iTAK as a mapping tool on their phone when they’re on mission or patrol.</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/955/1*mnawHo50nKYrO7a69yqfZQ.png" /><figcaption>Our current MVP: Spyglass 1.0.</figcaption></figure><h4><strong>Beneficiary Discovery Key Takeaways</strong></h4><ol><li><strong>N., S-3, U.S. Army (Iraq, Central Africa)</strong></li></ol><ul><li><strong>Translation and Bureaucratic Roadblocks: </strong>In my experience with social media (SM) analysis, key issues were <strong>language translation</strong> and <strong>getting approval</strong> from the military IT system <strong>to use paid software</strong> like Echo-C or HootSuite <strong>or open-source tools</strong>. Most OS tools were blocked by firewall filters, even a tool like TweetDeck that was made by Twitter.</li><li><strong>Multiple End-Users for Spyglass: </strong>Spyglass could have interfaces for S-2s and commanders. The S-2 could push down a set of queries for each commander based on what they think the company or platoon should know, but the commander could also perform and save his/her own searches. Spyglass could be <strong>a routine part of the checklist </strong>for a commander before a mission.</li><li><strong>SM is Ad-Hoc:</strong> At a company level, there might be one person who happens to be Twitter-savvy and thinks to look at it, so s/he’ll set up something to look at social media.</li></ul><p><strong>2. Anonymous, J-2, U.S. Army</strong></p><p>Based on what Spyglass wants to accomplish, there are three things to consider when building the tool:</p><ul><li><strong>Control for false positives:</strong> Bot activity and zombie networks may be inflating the incidence or importance of a tweet. “Peer pressure” behavioral psychology (i.e., people are braver and louder online on SM than in person) may make a tweet seem more threatening than it is.</li><li><strong>Control for false negatives:</strong> Poor local wifi connectivity may be suppressing social media activity.</li><li><strong>Efficiency in handling translation</strong> is important for making SM data actionable in real time. Sheer volume of translation needs is already a problem in S-2 battalions; a tool that increases SM analysis adds to that.</li></ul><p><strong>3. R., Senior Intelligence Officer, Army Special Forces</strong></p><ul><li><strong>Real-time importance: </strong>Our enemies have faster networks for putting information on social media. A social media tool can have a large impact if it can monitor in as close to real-time as possible.</li><li><strong>Possible conflicts over access to raw data: </strong>Many intel analysts sit behind a desk in an office. The field intel officer doesn’t want them tampering with the raw social media data because s/he feels s/he is in the best position to determine what is useful in the area of operations.</li><li><strong>Device fatigue is pervasive: </strong>Devices are abandoned because they lose support, aren’t updated, don’t have a clear end user. Nett Warrior could be hardware Spyglass could integrate with.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/640/1*7YWc-51TebSzhVkJreQv3A.jpeg" /><figcaption>The U.S. Army’s Nett Warrior system</figcaption></figure><p><strong>4. C., Captain, U.S. Army (Military Intelligence Company, Afghanistan)</strong></p><ul><li><strong>Possible Spyglass 1.0 MVP implementation:</strong> Platoon leaders consult with a battalion S-2 during the mission planning process, typically the 35D S-2. 35Ds provide daily intel updates and a last-minute briefing to the platoon leader an hour or so before mission. Right now, the S-2 shop could use Spyglass to speed up social media searching and the 35D could more include more SM insights in their briefings to the platoon leader.</li><li><strong>Information Spyglass should filter for:</strong> Indications and warnings (INWs) of two types of attacks: <strong>kinetic</strong>, such as roadside bomb on U.S. forces or ambush, AK-47 rounds on dismounted soldiers talking to a group of civilians, and other force-on-force attacks; and <strong>intimidation</strong>, such as roadblocks, kidnappings and beheadings of villagers, and anything that could be a threat influencing the local population.</li><li><strong>HUMINT Collectors are tangential beneficiaries: </strong>Platoons in MICs could benefit from Spyglass — particularly HUMINT collectors who are often in the field. HUMINT collectors may be able to verify intel from their sources on the spot with Spyglass.</li><li><strong>SM is Ad-Hoc: </strong>There is no formalized, dedicated role for OSINT collection until you get to the <strong>division</strong> level, and sometimes at the <strong>brigade</strong> level. Current efforts at battalion and company level are ad-hoc.</li><li><strong>MICs are the intelligence collectors of a BCT. </strong>Military intelligence companies (MICs) are composed of platoons that specialize in human intelligence (HUMINT), counterintelligence (CI) or signal intelligence (SIGINT). Each platoon produces intelligence summaries that are sent up to the brigade level to be turned into <strong>intelligence products</strong>.</li></ul><p><strong>5. N., Captain, U.S. Army. Formerly Platoon Leader (Iraq, 2008 and Afghanistan, 2010 and 2011)</strong></p><ul><li><strong>Information Spyglass should filter for: </strong>people collecting on U.S. and reporting U.S. activities; attacks on a camp; anything that would affect local sentiment (construction project, local government initiative); effectiveness of civil-military operations; “identifying what local populations in our AO care about to help shape our posture and presence when we go on patrol.”</li><li><strong>Preferred view of Spyglass for a commander </strong>is a list of tweets with content (ideally translated) and timestamp, with the ability to do keyword searches.</li><li><strong>Commanders and platoon leaders could use Spyglass to assess effectiveness</strong> <strong>of his/her operations through social media. </strong>Build a section of Spyglass for users to input observations and reflections during or after missions, especially in response to intel from Spyglass. Normally, commanders have to rely on the S-2 shop for this assessment.</li><li><strong>Security</strong> of Spyglass is paramount. It could be a huge OPSEC vulnerability if an enemy could collect info on what keywords a platoon leader is searching for.</li><li><strong>Mapping</strong> is not a priority: Mapping could be useful to understand neighborhood differences in a big urban center like Kabul. A commander might not have time to scrutinize maps, and delegate to support staff.</li></ul><p><strong>6. N., Lieutenant, U.S. Marine Corps (Afghanistan). Formerly Platoon Leader (Iraq, 2009)</strong></p><ul><li><strong>Pain-point: “We never knew if a genuine relationship is being built.” </strong>Social media could give insight into understanding local sentiment.</li><li><strong>Everyone has cellphones: </strong>Locally purchased “drop” phones are used to communicate with local populace who provide tips.</li><li><strong>Blue Force Tracking:</strong> Once away from base of operations, situational awareness updates could be sent to the Blue Force Tracking system in vehicles, connected to satellite.</li><li><strong>Detection of pictures</strong> or video of forward operating bases on social media is a key aspect of force protection.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*f9gD5AewyWiflL6a75YCYA.jpeg" /><figcaption>Team Spyglass visits Voyager Labs</figcaption></figure><h4>Technology Discovery Key Takeaways</h4><p><strong>7. D., Geospatial Information Officer, Office of the Secretary of Defense</strong></p><ul><li>For military-specific mapping capabilities, I’d consult the Army Geospatial Center (AGC) and National Geospatial Agency (NGA).</li><li>OSD has policy oversight which includes how GIS is used on military servers, but specific implementation and capabilities in semi-permissive areas is deferred to other organizations.</li><li>Most geospatial work is done on ESRI’s <strong>ArcGIS</strong>. Real-time feeds can be mapped on ArcGIS.</li><li>Geospatial information can be used in the following formats: KMZ (open standard); Shapefile (open standard); KML (Google Earth); ESRI’s proprietary format.</li></ul><p><strong>8. H., Voyager Labs (Site Visit)</strong></p><ul><li>Voyager’s speciality is in <strong>network topology</strong> and understanding strength of relationships between individuals and groups. Understanding networks behind influencers can help measure impact and reach of statements.</li><li>Proprietary algorithm is based on the interdisciplinary work of behavioral psychologists and cognitive scientists that understand how people make connections. Voyager addresses the pain point of being the outsider in the community when engaging with host nation partners.</li><li><strong>Topic search function has powerful language support </strong>for 52 languages with possibility of enabling 103 more, across publicly posted information on Instagram, Vimeo, YouTube, Facebook, Twitter, Medium, Tumblr, Pinterest, VK, G+.</li><li><strong>Geolocation</strong> is a hard problem to solve as most social networks don’t provide that data in the open API, unless user “checks in.” Collections can be acquired and stored on a client/laptop, so persistent connection to the cloud is not necessary to use Voyager.</li></ul><p><strong>9. C., IST Research, </strong>working with DARPA on threat monitoring and detection via social media</p><ul><li><strong>Geolocation is a big obstacle to real-time capability.</strong> Only 2% of open source social media is immediately geolocatable, so the result is “this is the conversation around me from the 2% that is geolocatable.” To get past it, must use geo-inferencing.</li><li><strong>Geo-inferencing/geo-sensing</strong> is a hard problem and requires data correlation from both classified and public sources of intelligence.</li><li><strong>Key aspects of force protecton </strong>include real-time alerting, geo-inferencing, and image detection of a known physical location (base) or personnel.</li><li>IST is a technology company that works with DARPA on developing capability of monitoring and detection of threats to operational security (OPSEC) through social media. Focused on Twitter, Facebook and Telegram collection, as well as friendly forces leaking information.</li><li>IST is part of a larger effort by DARPA’s Quantitative Crisis Response (QCR) program, which is persona-based and powered by machine translation, analysis and scoring. Not real-time.</li><li>DOD is focused on Twitter, but complexity starts with Telegram or WhatsApp or Viber — where communications actually occur.</li></ul><p><strong>10. T., Talon Intelligence, </strong>expert on Telegram and open source research on extremists</p><ul><li><strong>Telegram</strong> consists of Groups and Channels. <strong>Groups</strong> work like typical group messages (think Whatsapp). <strong>Channels</strong> allow one user to broadcast content to many.</li><li><strong>Security</strong> is provided by 1) hosting servers in many different countries, and 2) storing the decryption codes in different countries so that serving warrants requires unprecedented cooperation between many nations.</li><li>ISIS typically opens a channel publicly for a very brief period to allow people to join and then closes it to avoid detection.</li></ul><h4>Hypothesis Evaluation</h4><ol><li>Company commanders or platoon leaders want information about <strong>specific events</strong> from social media. <em>(Validated.)</em></li><li>Company commanders or platoon leaders care more about the <strong>content of tweets and the time/location</strong> it was posted than trends or interactions. <em>(Partially validated; should ask more beneficiaries.)</em></li><li>Company commanders or platoon leaders don’t need tweets <strong>mapped</strong> in a MVP. <em>(Partially validated; should ask more beneficiaries.)</em></li><li>Company commanders or platoon leaders use <strong>ATAK/iTAK</strong> as a mapping tool on their phone when they’re on mission or patrol. <em>(Partially invalidated; ATAK/iTAK not consistently used in conventional Army.)</em></li></ol><p>For every question we answer, it seems that two new ones emerge in its stead. While tactical commanders at the company level and platoon leaders feel that greater access to social media would be beneficial, there is not a consensus as to what that access should look like:</p><ul><li><em>Do platoon leaders need access to raw social media data when out on patrol, or would it be more effective to disseminate that information through the battalion S-2?</em></li><li><em>What types of filters are required to help commanders identify relevant threats to their forces?</em></li><li><em>Does the Army have existing hardware on which we can run our platform, and is their a reliable connection for soldiers out in the field to receive real-time updates?</em></li></ul><p>Furthermore, we’re looking forward to talks with private sector companies such as BabelStreet to see if Spyglass can be integrated with their existing capabilities. As we’ve heard many times, language translation is a huge sticking point when it comes to social media analysis.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e548bae50361" width="1" height="1" alt=""><hr><p><a href="https://medium.com/spyglass/learning-by-building-e548bae50361">Learning by building</a> was originally published in <a href="https://medium.com/spyglass">Spyglass</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Team Spyglass clears its objective lens]]></title>
            <link>https://medium.com/spyglass/spyglass-clears-its-objective-lens-587382f472e3?source=rss----cfbb6f8dc674---4</link>
            <guid isPermaLink="false">https://medium.com/p/587382f472e3</guid>
            <category><![CDATA[hacking-for-defense]]></category>
            <dc:creator><![CDATA[Jordan B.]]></dc:creator>
            <pubDate>Sun, 05 Feb 2017 22:49:41 GMT</pubDate>
            <atom:updated>2017-02-06T04:48:31.353Z</atom:updated>
            <content:encoded><![CDATA[<p>For week 4, we continued <strong>beneficiary discovery </strong>focusing on line battalion commanders and lower echelons, and <strong>tech discovery</strong> focusing on people who have spent significant time analyzing social media data to produce insights for national security.</p><p>Regarding beneficiaries, we aimed to continue talking to line battalion commanders and drill lower down the chain of command. We learned from AWG that folks at lower echelons, along with S-2 officers working with open source intelligence (OSINT), are their conceptions of ideal end users of our product.</p><p>Technology discovery allows us to better understand the technical and analytical hurdles we will have to overcome in order to build a product that delivers value as stated. This also informs us of what tools, both enterprise and open source, are already available off-the-shelf to augment what we seek to accomplish with Spyglass.</p><p>This week, we also sought to better understand the differences between operating environments and how the different operating environments affect beneficiary interaction with social media data. We believe that Spyglass will be deployed in semi-permissive environments, which are environments where a host nation is supportive but not in total control. While entry operations may be supported, operations could be opposed at anytime by insurgencies or other forces outside of a host nations’s control. We are still working to understand the different environments but based on our beneficiary interactions we get the picture that reality is much more fluid than what can be captured by discerete definitions of operating environments.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Adz-AX5g735kplFQZ0AWBA.jpeg" /><figcaption>Team Spyglass, hard at work.</figcaption></figure><p><strong>This week the team sought to test the following hypotheses:</strong></p><ol><li>Semi-permissive environments have different OSINT (open source intelligence) requirements than non-permissive environments.</li><li>Information Operations Officers don’t have a tool to measure the effectiveness of their operations.</li><li>Battalion commanders need direct access to social media information (not only intel briefings).</li><li>Intelligence Officers want to get more insights from social media.</li><li>Social media information is required in real time. There is hardware already in place to present real-time OSINT data.</li></ol><p>Our interview summaries below highlight our key takeaways pertaining to these hypotheses:</p><p><strong>M., Lt. Col, Battalion Commander and Company Commander (Afghanistan and Iraq)</strong></p><ul><li><strong>As an outsider in a culture or community,</strong> a tactical commander in “advise, train, and assist” role needs cultural context, open source and social media validation of population sentiment and intelligence on “who is sitting across the table”, to whom he is connected and how.</li><li><strong>OSINT and social media is more actionable at the battalion level</strong> where the information can be vetted and synthesized. “At the battalion level, I could feed that info to the lower level digitally, but I have analysis done, so that company commander doesn’t have to think.”</li><li><strong>A company-level commander has more immediate priorities </strong>of physical security, fire systems, communicating with platoon leaders, understanding where they are on a map. “As a company leader, I am using my eyes, I am doing everything to defeat the enemy kinetically, not how to pick up their signals or what he is thinking”.</li><li>Definition of semi-permissive vs non-permissive environment depends on “the amount of time you have to assess the situation.”</li><li><strong>Capability for OSINT exists, but major issues with limited bandwidth and “cross domain” issue of how to get “OSINT into the closed loop”</strong>. In their intel team, there is a role called the “swivel chair.” This is the human in the loop, whose job is to transfer info from unclassified system to the classified network mediums manually, swiveling from one desk to the other (E.g. “Palantir” to D6GSS). It is time-consuming and labor intensive.</li><li>The current reliable communications system to talk to captains on the ground is radio.</li></ul><p><strong>Hypotheses:</strong></p><blockquote>H1: Validated. Semi-permissive environments take more into account than population factor than permissive.</blockquote><blockquote>H3: Validated. He tried to use OSINT at the battalion level.</blockquote><blockquote>H5: Partially validated. No real-time requirement. There are important limitations in connectivity when talking to captains on the ground and the amount of information they can consume.</blockquote><p><strong>K., J-3, U.S. Army (Iraq, Afghanistan)</strong></p><ul><li><strong>S-3 is operations; their job is to coordinate all operations: fire, lethal and non-lethal.</strong> “As a J-3 IO, I’m in charge of all non-lethal effects. Any effect that makes someone do something (e.g. influence operations) but doesn’t kill them.”</li><li><strong>IO is a long-term planning process. </strong>“It’s important for me to stay ahead as much as I can. A lot of my work is self-driven; I look at the commander’s target set for the next few months and plan my own operations to support those targeting efforts.”</li><li><strong>In conventional Army, at all levels — brigade, division and corps — there is lack of IO manpower. </strong>S-3 monitor all intel for threats when teams are deployed on missions as much as possible to help them gain situational awareness, but “it’s difficult for us to keep track of every battalion or every company. There’s not enough IO officers to do that.”</li><li><strong>Special operations probably has the most tools for situational awareness. People in the military are monitoring everything because it’s imperative that mission stays quiet. </strong>“There might be an entire cell dedicated to that mission that reports to me for any posting relevant to the special op.”</li></ul><blockquote>H1: Invalidated. No clear and consistent distinguishing factors between semi-permissive and non-permissive when it comes to OSINT. “Not something you should be focusing on, in my opinion.”</blockquote><blockquote>H2: Partially invalidated. Depends heavily on what type of operations you are referring to and what you mean by a “tool.” For many operations, it’s clear from the mission or key engagement (KOE) outcome or from population sentiment whether we’ve succeeded.</blockquote><blockquote>H3: Partially validated. From my perspective, it would save me a lot of time if a commander could get direct social media analysis. But the reason I’m here is because I’m trained to analyze OSINT or I might be better at analyzing it than a commander is.</blockquote><blockquote>H5: Partially validated. Real-time is cool, but there’s value in OSINT that isn’t real-time as well. Don’t let lack of real-time capability deter you.</blockquote><p><strong>J., Army Special Forces Officer (Afghanistan)</strong></p><ul><li><strong>A lot of information and intel comes from the bottom-up.</strong> People on the ground collecting information and feeding it up. Intel is generated from the guy doing the missions.</li><li><strong>For situational awareness, we had a continuous ISR</strong> [intelligence, surveillance, reconnaissance] <strong>(think: a big camera in the air)</strong> that would send info back to a workstation at our base. Someone would be monitoring that feed while we were on missions, and would radio us if there was something we needed to know.</li><li>I typically had a E-7 Intelligence Sergeant with me who would be trained to manage the team’s intelligence. Generally he would receive the radio reports from the ISR workstation if something was urgent.</li><li><strong>Most people don’t use government issued phones overseas. </strong>And most people didn’t bring personal phones because they were a security liability. “We had texting capability on our vehicles, but usually we communicated over radio.”</li></ul><blockquote>H3: Partially invalidated</blockquote><blockquote>H4: Partially invalidated</blockquote><p><strong>Anonymous, Army Special Forces Team Captain, Counter-ISIL Operations (Iraq and Syria)</strong></p><ul><li><strong>The SOF unit Anonymous is in is very linear/non-hierarchical. Responsibilities are spread evenly among the 10 team members, including intelligence.</strong> “We do all the intelligence collection operations ourselves. We did have higher echelon support, but we probably had more of a grasp of the intelligence picture on the team than up high.”</li><li><strong>OSINT doesn’t integrate well with other intelligence due to classification differences. </strong>“I don’t use social media much because I can’t integrate my open source intel with other intel to easily give us a complete picture of what’s going on.”</li><li><strong>An Android app can be integrated with other software that we use, such as ATAK</strong> (<a href="https://en.wikipedia.org/wiki/Android_Tactical_Assault_Kit">Android Tactical Assault Kit</a>), a geomapping program that “allows me to see missions, essential info out there, where my team members are. ATAK pushes all the info to an open VPN that we use.” ATAK does not have a real-time social media feed; only classified info.</li><li><strong>In semi-permissive areas, you’ll usually have connection to 2G/3G/4G data.</strong> I carried both Android and iPhone; people can carry multiple phones with multiple carriers.</li></ul><p><em>Real-time OSINT that might be useful for me are (MVP ideas):</em></p><ul><li>Sound signals of gunshots, helicopters, someone following me etc. in my location</li><li>Any type of link association. For example, if I’m targeting someone and someone they are associated with posts something geotagged in my location, that would be useful. Or if I’m talking to someone who might be associated with my target and I could input his Twitter handle or screenname into an app that would output his links and associates, that would be helpful.</li><li>Filters for criteria that I specify. If I’m doing surveillance, I’d want to hear about protests and roads. If I’m doing general stuff, I want to know most information.</li></ul><blockquote>H1: Partially invalidated. OSINT requirements depend more on the mission, area, team, etc.</blockquote><blockquote>H4: Partially invalidated.</blockquote><blockquote>H5: Partially invalidated.</blockquote><p><strong>L., Marine Corps Special Operations, S-2 Captain</strong></p><ul><li>In Special Operations, the S2 captain works with his team in a classified room where classified information resided and can be discussed.</li><li><strong>The intelligence offices use a system of predefined queries to be alerted of relevant incoming intel reports on real time. </strong>Ideal solution for analysts at this level is a system allowing to upload requirements based on search logic and get a tailored report incorporating finished intel reporting and the unstructured social media data</li><li><strong>Social media intelligence should go to the lowest level commander. </strong>Threat information most useful at the watch officer level.</li><li>Data collection for intelligence is limited by authorization and legal boundaries. Intelligence analysts at company level are limited by regulations around what can be done from USG-issued hardware to only public social media data, public broadcasting feeds, announcements of rallies for situational awareness.</li></ul><blockquote>H3: Invalidated. Social media information is only one part of a billion things a commander requires. S2s serve them with the information required.</blockquote><blockquote>H4: Validated. There is a need to a clear conduit to collect and use data that now is out of reach.</blockquote><blockquote>H5: Validated. The S2s have the hardware to work with it. Real-time data might be relevant if the information is relevant.</blockquote><p><strong>M., Lt. Colonel, U.S. Army (Bagram Airfield)</strong></p><ul><li><strong>The commander has competing demands and limited assets.</strong> When on an advise-and-assist mission you need to determine how much to share with the host nation and agree on what to prioritize.</li><li><strong>A commander is always working with his staff to form hypotheses. He puts collection assets out in the field to test the hypothesis.</strong> The S3 create requests for information (RFI’s) for the patrols to collect information on.</li><li><strong>The S2 and the S3 work extremely closely:</strong> the 2 is the subject matter expert and the 3 is putting together the operation. The commander is assessing the risk. “I needed to weigh whether I would assume more risk by not patrolling than patrolling”</li><li><strong>Risk tolerance is highly dependent on geographical location and mission structure.</strong> “I had colleagues in hostile sectors with higher threat targets but because of the strategic nature of my location I had a greater risk profile.”</li></ul><blockquote>H1: Validated. When dealing with the population more you need to know their sentiment.</blockquote><blockquote>H3: Partially validated. The commander wants access to anything that helps him understand his environment.</blockquote><p><strong>C., Lt. Colonel, U.S. Army</strong></p><ul><li><strong>“Intelligence drives maneuvers.”</strong> The more you understand your environment and the enemy, the greater fidelity you have in your operations.</li><li><strong>They have different systems with unique capabilities, but they are unable to share information. That process is done manually </strong>(which sounds similar to the “swivel chair” problem). “The worst thing a tool can do is have a niche capability but can’t communicate with anything else.”</li><li><strong>“I want to be able to visualize the information through a graphic depiction of the environment.” </strong>Tools are operated inside the intel team: “they add some ‘magic’ to it and put it on a PowerPoint slide so the commander can understand it.”</li><li><strong>“If you are ever going to sell it to the military, you need to understand where it fits doctrinally.”</strong> Explaining the pain point you are solving with military doctrine is how you get buy-in.</li></ul><blockquote>H1: Partially validated, though it is more on the basis of infrastructure.</blockquote><blockquote>H2: Partially validated. Information operations officers below the brigade level lack training and split their duties.</blockquote><blockquote>H3: Invalidated. While not every piece of information requires and action, the commander has too many responsibilities to handle social media directly.</blockquote><blockquote>H5: Partially validated. The quicker he can receive actionable intelligence the better, but there are technical and personnel impediments to that.</blockquote><p><strong>A., Captain, U.S. Army</strong></p><ul><li><strong>When a captain goes on patrol, he is acting as the eyes and ears for the battalion</strong>. During his tour he was doing advise-and-assist so he often worked with the Iraqi army.</li><li><strong>Intelligence was sometimes lacking in depth. </strong>Once, on a patrol during a holiday: <strong>“We knew there was a holiday because of intel, but we didn’t know what it would actually look like.”</strong> People were burning cars and basically rioting.</li><li><strong>When on patrol you are most concerned about the physical threats posed by your immediate environment. </strong>An ideal tool might be a battalion level asset that can be pushed down to companies when necessary.</li></ul><blockquote>H1: Partially validated, but it is not so much based on the permissiveness of the environment but the goals of the mission.</blockquote><blockquote>H5: Partially invalidated. ACaptain has limited bandwidth and is too focused on physical threats to have time to ingest real time information, but he want to know about potential threats from social media.</blockquote><p><strong>N., Captain, U.S. Army (Iraq)</strong></p><ul><li>Intelligence was provided at battalion and brigade level, but <strong>as battalions and companies deployed on their own, they need to stand up their own intel, OSINT/social media units.</strong></li><li>D6 is a battalion level asset, but companies are starting to get their own. Primary purpose of company one is to receive data feeds from battalion one, focused on situational awareness and hostile action, maps with markers for any kind of threat.</li><li>Connectivity is satellite-based with limited bandwidth, unless they are in physical proximity and can “pull the cable across” to sync the data feeds between company and battalion systems.</li><li><strong>The system isn’t usable while it is syncing data, so “real-time” is an issue. Radio is the most reliable means of communication for immediate threat.</strong></li></ul><blockquote>H1: Validated</blockquote><blockquote>H3: Validated</blockquote><blockquote>H5: Partially validated; current capability is limited.</blockquote><h4>Tech Discovery Highlights:</h4><p><strong>E., Researcher, Rand Corp.</strong></p><ul><li>There are advanced methods to research networks connections and lexicon on social media data. There is no need to create this again.</li><li>Most of the research is done on Twitter historical data. There is no need to adapt research to a certain language.</li><li>It is difficult to work across different platforms. RAND had to develop their own data aggregator in house.</li></ul><p><strong>Hendrik van der Meulen, Voyager Labs</strong></p><ul><li>Underlying algorithm and clustering organized to show social network topology</li><li>Interoperable with FB, Twitter, VKontakte, Instagram, YouTube, Pinterest, Google+, ability to plug in others.</li><li>Substantial language support</li><li>Open source collection only: astounding amount of information can be gleaned from public profiles and posts without need to create false profiles or hacking.</li><li>Deep topic search as well as individual at the center of network</li><li>Relationship strengths determined based on algorithm and intuition, not just the frequency of contact and engagement</li><li>Ability to identify trolls on the basis of the shallow profile/depth of connections</li><li>No “real-time” capability — connectivity is key</li><li>No push notifications, but possible feature request</li><li>Predictive capability unclear</li></ul><p><strong>K., OSINT Center, CIA</strong></p><ul><li>Bottom-line advice: look at the whole spectrum of the problem</li><li>Understand the categories of capacity you need to build</li><li>Look at commercial solutions like Babel Street, Creative Radicals, Dataminr, others</li><li>Could the data be carried via sat? Is it a “new tool problem” or a “connectivity” problem?</li><li>Do solutions exist “state side”, but not in the field? Is it the “hardware pipes” vs “solution doesn’t exist” issue?</li></ul><p>As we wrap up the week, a few things are clear: we have more customers to interview, more “off-the-shelf” tools to understand, more questions to answer.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=587382f472e3" width="1" height="1" alt=""><hr><p><a href="https://medium.com/spyglass/spyglass-clears-its-objective-lens-587382f472e3">Team Spyglass clears its objective lens</a> was originally published in <a href="https://medium.com/spyglass">Spyglass</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Team Spyglass goes to Fort Meade]]></title>
            <link>https://medium.com/spyglass/team-spyglass-goes-to-fort-meade-5e56eff2f735?source=rss----cfbb6f8dc674---4</link>
            <guid isPermaLink="false">https://medium.com/p/5e56eff2f735</guid>
            <category><![CDATA[hacking-for-defense]]></category>
            <dc:creator><![CDATA[Katerina Sedova]]></dc:creator>
            <pubDate>Sun, 29 Jan 2017 23:33:49 GMT</pubDate>
            <atom:updated>2017-01-30T16:34:49.030Z</atom:updated>
            <content:encoded><![CDATA[<p>On Friday morning, armed with coffee and laptops, we piled into a van and hit the road to Fort Meade, where Asymmetric Warfare Group (AWG) calls home. Our goal was to interview as many AWG operational advisers and operational support staff as possible, or in lean-startup-methodology speak — to get out of the building.</p><p>We centered our customer discovery around the hypothesis that U.S. in-country personnel feel that they lack an awareness of threats that could be identified through social media. More importantly, we wanted to hear what they felt were the pain points of their work and to get a visceral understanding of their experience on the job. <em>What is their typical day here and in-country? What happens as they get ready for deployment, when they step foot on foreign soil, and at every stage of their experience? </em>We wanted to “walk in their boots” and “feel” their environment to the extent that we could, being civilian graduate students removed from the theater of operations around the world where U.S. service members find themselves.</p><p>Our visit to AWG headquarters did not disappoint! We left with the impression that AWG is a special beast within the U.S. Army with the mission to identify and solve problems for the Army, as management consultants may be hired to solve problems for a private company. They are the Mr. Wolf (“Pulp Fiction”) of sorts. To that end, the men and women serving at AWG are chosen for their significant experience and, among other qualities, the ability to identify and pinpoint the heart of the matter in any given situation where they are serving in the “advise and assist” capacity. Every advise mission is different. Every “customer” varies, as do the problems AWG is tasked to help solve.</p><p>We started our visit at the range, <a href="https://www.dvidshub.net/video/335510/pittman-memorialized-new-awg-indoor-range">dedicated in honor</a> of a fallen soldier, Master Sergeant Robert Pittman, who was killed in Afghanistan while serving as an Operational Advisor. MSG Pittman was the only person AWG ever lost on the job in its 10-year history. We were carefully instructed as we fired an M4 rifle at paper targets. It was the first time firing a standard issue Army rifle for all of us. Some of us were better than others at marksmanship — time spent playing video games may have strongly correlated with our accuracy. From the difficulty of aiming in standing position, to the kickback, the smell, and the heatwave hitting one’s face, this exercise was a powerful glimpse for us into the everyday experience of U.S. Army personnel.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*QltMIK8yL0ju1dszPXYwmQ.jpeg" /><figcaption>Memorial for MSG Pittman at Ft. Meade.</figcaption></figure><p>After the range, we settled into a conference room for our customer discovery interviews. Throughout the day, we spoke with 8 members of the AWG serving in different roles, from Operational Advisors (OA) to Operational Support (OS) such as communications specialists (S-6) and intelligence analysts (S-2). OS work with OAs before, during, and after their deployment to help with their missions. Our summaries below highlight the differences in roles, pain points, and takeaways we gleaned through our conversations.</p><ol><li><strong>J., Operational Advisor (Asia)</strong></li></ol><ul><li>There is a need to use social media at the OA level to gain situational awareness in the area they are in. But the OA is able to obtain information through searching social media by him/herself.</li><li>Twitter is not a useful social media in Asia (Southeast mainly). Facebook and Instagram are more useful.</li><li>On mission in country, he already has a routine to ensure situational awareness that includes social media search. However, it is time-limited by a wifi connection, so usually he does not have access to up-to-date data during the day.</li><li>Hypothesis validated regarding the need for gleaning information from social media for situational awareness and self-protection; however, OAs may not be our target beneficiary because they seem to have a comparatively less need for a new solution.</li></ul><p><strong>2. T., Communications Analyst</strong></p><ul><li>There is a redundancy plan for communications mechanisms, which includes in order of preference Primary, Alternate, Contingent (local cellular infrastructure), and Emergency (PACE) to help minimize communication failure.</li><li>Local cell phone infrastructure is the preferred mechanism in some cases because it has more bandwidth. But no classified information goes through a cell-phone.</li><li>Satellite communications are available in most areas, but it is considered as a tool of last resort.</li><li>Hypothesis: N/A</li></ul><p><strong>3. B., Operational Advisor (Middle East)</strong></p><ul><li>Major pain points of my job would be: getting past cultural barriers working with local staff. E.g., when working with or talking to a senior Iraqi officer or his troops, trying to get the officer to understand your viewpoint of the situation and intentions without offending them.</li><li>Getting our own US coalition and advisory forces to take the same advice when working with Iraqi officers or their troops. If something’s not working in their work with Iraqi forces, they should look at the problem from a different angle.</li><li>Finally, getting our US coalition and advisory forces to get further down to the tactical level (e.g., at the platoon or company level) when working with local</li><li>Hypothesis partially validated</li></ul><p><strong>4. R., Operational Advisor (Middle East)</strong></p><ul><li>Every mission and every problem is different, so there’s no typical problem. “It depends” was the most frequently used phrase.</li><li>Operational advisors know a great deal about operational security and the digital signatures they are leaving, but conventional forces don’t think like that and many commanders don’t understand the problem. A little tidbit on social media can have huge political ramifications.</li><li>Open source intelligence (OSINT) tells you more about what people are thinking. When cross-referenced with traditional intelligence, you could make predictive threat assessments. How do you take an aggregate of all that information and synthesize it?</li><li>The environment drives military operations, a commander needs to know how to deploy troops and navigate that environment. Intelligence drives our knowledge of the environment, and open source is a key component of that.</li><li>Hypothesis validated</li></ul><p><strong>5. T., Operational Advisor (Middle East)</strong></p><ul><li>A conventional force likes to bring all of its gadgets, all of which leave signatures behind. Soldiers may not be cognizant of the fact that U.S. rules for monitoring own personnel don’t apply to other nations. Your local SIM card is transmitting everything to the Iranian telecom company.</li><li>When activity occurs, we may not be able to observe it explicitly, but we can observe increases in electronic activity to give us an indication that something is happening. (This may dependent on cultural factors as the example was of Iraq which is a “talkative nation.”)</li><li>A problem for commanders is that there is a wealth of information out there, but they don’t have the tools to gather that information and they don’t have the personnel to analyze that information. There is no central repository of open source intelligence. You could be pulling data at a regional level and feeding it up to a higher level.</li><li>Hypothesis partially validated</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*5kigkGqvaNnLAfXG2_3tWQ.jpeg" /><figcaption>Team Spyglass at the shooting range.</figcaption></figure><p><strong>6. S., Lead Intelligence Analyst</strong></p><p><strong>7. E., Intelligence Analyst</strong></p><p><strong>8. C., Human Intelligence Collector</strong></p><p>Key takeaways:</p><ul><li>We’re in a multi-domain battle environment and intelligence sources, but we don’t have all the tools to aggregate, overlay, and interpret this data in one medium to allow for different functions (electronic warfare, Geoint, cyber, HumInt, etc) to understand and relate to one another.</li><li>Lack of tooling to enable zooming in/out to analyze and report information at appropriate levels for appropriate audience (e.g., strategic, tactical, ground level)</li><li>In general too much information and different tools, but right now can data be mined after the fact, not in a timely predictive way. Want to tap into machine learning to allow to make more nuanced connections, decipher misinformation and doctored imagery, relate intel from different sources for more context and process. Takes too long to make those connections now. One tool to connect them all and overlay information over a table-top map/ virtual earth to visualize and make sense of all the information in one medium</li><li>Open Source Social Media is 80% of intelligence, but the allocation of OSINT resources within in-country conventional forces is frequently spontaneous, at the discretion of the commander, with resources thrown into the social media space without preparation or tooling in an ad hoc way to fill immediate need. Social media scraping at a basic level, sometimes with linguistic support from local staff.</li><li>Technical obstacles to having persistent connectivity to the headquarters, if the solution for in-country is cloud-based. Text capability over primary means of communication, some text and code over radio</li><li>Hypothesis validated, but at the HQ Analyst level. The level of analysis, the goal, and the degree of zoom diverges from the needs of line troops.</li></ul><p>Our AWG site visit reinforced the themes we heard in the few interviews we conducted earlier in the week, with the following key insights:</p><p><strong>9. M., Lieutenant Colonel, U.S. Army</strong></p><ul><li>Information has the power to mitigate risk and exploit opportunities. A tactical commander needs the ability to aggregate information and the personnel to make sense of the information in a timely enough manner to do either of those things.</li><li>Social media monitoring was performed by local nationals. The commander’s concern was focused on general population sentiment, what are they saying about my forces, any negative events I should know about) and was mainly used to facilitate communication with the local populace.</li><li>Social media’s greatest utility for force protection comes from the ability to alter patrol routes. Are people posting images of troop routes? Has general sentiment toward the U.S. military worsened in a particular region? Commanders want to know where not to send their guys.</li><li>Hypothesis validated</li></ul><p><strong>10. A., Private Sector, formerly of Information Operations Command</strong></p><ul><li>Biggest difference between private sector and military: “I don’t have to get anyone’s permission to do open source research”</li><li>In the military there are different procedures for getting authorization that are dependent on the purpose of the research. A big problem is that the people who make the rules in the military are too far from the end user.</li><li>As an Information Operations Officer, he was always assigned multiple duties (understaffed). His commander tasked him with getting general atmospherics. A positive of social media analysis is that you can use the people’s own words to engender good will. There is a problem with our own units posting information online.</li><li>Most people at higher levels of command (06 level) do not understand the risk of cyber threats. The biggest challenge is tying a vulnerability to a specific threat to make them understand the problem, but it is challenging to prove the effectiveness of your efforts.</li><li>Hypothesis validated</li></ul><p><strong>11. J., Special Forces (Afghanistan)</strong></p><ul><li>He does not require access to social media. The commander on the ground requires only the information that is relevant for him.</li><li>On high profile missions there was an intel team assigned to assist on real time. They checked social media and were able to identify postings that might be relevant within a one-hour time frame.</li><li>Social media capabilities are more useful to manage perception by the population. Counter-insurgency can not be effective without the population support.</li><li>Hypothesis partially validated. Social media is useful, but some capabilities are already in place. Social media is very useful for strategic information operations to influence the population.</li></ul><p><strong>12. J., former Team Leader of a Military Transition Team (MTT) in Afghanistan</strong></p><ul><li>Force protection factors for the commander in-country rest on the fundamental question: are there specific threats I must consider today that could change the current situation, while my troops are outside of the safe zone?</li><li>General awareness of being in hostile territory, which the intelligence briefing provides, is helpful. But need a real-time situational awareness updates as movement proceeds and situation changes while you are away from base of operation.</li><li>Need real-time social media awareness of overtly hostile or potentially destabilizing messages and their reach (how many followers are seeing the message, how many retweets); indicators of destabilizing events (civilian casualties of US actions, misconduct by troops) that could change the population sentiment while en route; indicators of compromise of operational security.</li><li>Factors of concern can be categorized as time, location, content, and persona.</li><li>Hypothesis validated</li></ul><p><strong>Despite the difference of missions </strong>(“It depends!”)<strong>, experience and roles, some common threads emerged across our customer discovery interviews this week:</strong></p><ol><li>Social media is valuable to developing situational awareness and as a source of intelligence to inform commanders’ tactical decisions.</li><li>Commanders want to understand when social media indicates changes in population sentiment that may jeopardize the safety of U.S. personnel, as well as signal opportunities for greater tactical engagement.</li><li>Real-time, predictive capability is key, but achieving this in the conditions of non-persistent or limited connectivity will be a technical hurdle.</li><li>The needs of the in-country personnel in social media understanding are a specific subset of the needs of intelligence analysts, which need a capability to zoom in to the tactical in-country level or zoom out to the strategic level and integrate social media data with other sources of intelligence.</li></ol><p>It was a busy and exciting week 3 for Team Spyglass. We honed in on our original problem set and made progress on envisioning a MVP. We also began to resolve what we perceived last week as a contradiction between the needs of in-country personnel (specific) and intelligence analyst level (scaling up, down, and across). Our goal for next week is to speak to beneficiaries in the Army, e.g. commanders, soldiers, and intelligence analysts in line battalions, to understand their needs and threat environment directly. Stay tuned!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5e56eff2f735" width="1" height="1" alt=""><hr><p><a href="https://medium.com/spyglass/team-spyglass-goes-to-fort-meade-5e56eff2f735">Team Spyglass goes to Fort Meade</a> was originally published in <a href="https://medium.com/spyglass">Spyglass</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A first dive into our problem statement]]></title>
            <link>https://medium.com/spyglass/a-first-dive-into-our-problem-statement-daeb0ffe54ed?source=rss----cfbb6f8dc674---4</link>
            <guid isPermaLink="false">https://medium.com/p/daeb0ffe54ed</guid>
            <category><![CDATA[lean-startup]]></category>
            <category><![CDATA[customer-interview]]></category>
            <category><![CDATA[problem-solving]]></category>
            <category><![CDATA[pivoting]]></category>
            <category><![CDATA[hacking-for-defense]]></category>
            <dc:creator><![CDATA[Chloe Krawczyk]]></dc:creator>
            <pubDate>Sun, 22 Jan 2017 23:32:13 GMT</pubDate>
            <atom:updated>2017-01-23T05:34:33.604Z</atom:updated>
            <content:encoded><![CDATA[<p>As is typical of startup life, <strong>Week 2</strong> of our Hacking for Defense class did not unfold as we had expected. Our original goal for this week was to speak with 10 beneficiaries from the Asymmetric Warfare Group (AWG) to understand their daily workflow and to uncover pain points they were facing in their work. Unfortunately, it turned out that we were unable to connect with more than one beneficiary from AWG, forcing us to question whether we want to pivot our target customer for feasibility reasons.</p><p>Instead, we interviewed 12 potential beneficiaries from a variety of organizations who gave us useful insight into their current work, past experiences, and perspectives on our problem statement as written. These individuals were affiliated with organizations ranging from Army Special Forces to the Navy Cyber Defense Operations Command (NCDOC) to the CIA Open Source Center (OSC).</p><p>Undoubtedly, our 12 interviewees had differing ideas of what were important problems and potential solutions related to social media information gathering and threat identification. However, we picked up on <strong>10</strong> <strong>key themes</strong> that emerged from our interviews which will inform our work going forward:</p><ol><li>We can use social media data to understand <strong>population sentiment</strong>.</li><li>We cannot use social media data to <strong>identify immediate threats </strong>to force protection like improvised explosive devices (IEDs) or indirect fire (INFs). Many intents to plan such attacks are communicated on the dark net or other encrypted channels, not through social media.</li><li>We may want to build a solution that incorporates data from <strong>multiple domains</strong> (i.e., beyond social media exclusively).</li><li>We may want to build a tool that uses <strong>image recognition</strong> technology rather than language recognition or natural language processing (NLP) technology. The latter runs up against more foreign language/culture obstacles.</li><li>There were conflicting opinions on whether we should build a tool for <strong>an</strong> <strong>analyst at headquarters </strong>or for <strong>someone in the field</strong> with mission control (i.e., decision making capability). If we build a tool for analysts, whether we want to help analysts filter out the noise or whether they want all the info so they can do the filtering themselves is up for debate.</li><li>People in the field do not have reliable Internet access outside of hotspots in vehicles. People in the field on an active mission only have access to satellite radio. Satellite communications are unencrypted and the assumption is that they can be intercepted and decoded.</li><li>The problem we tackle may require not a technical solution, but rather advocacy for a <strong>policy or structural change</strong> within military leadership.</li><li>There is no centralized way to access social media in the Army or centralized repository of social media data.</li><li><strong>Our own soldiers’ activities</strong> on the Internet can increase their vulnerability to reverse image searches and geotagging analysis by adversaries.</li><li><strong>Information warfare</strong> and influence operations may be more important problems to tackle than predicting threats from social media.</li></ol><p><strong>None of these themes are firm conclusions. </strong>Depending on the actual target customer we choose to hone in on creating a solution for, we will dive deeper into some of these themes in future weeks as explore their specific problem, key hypotheses, and potential solutions.</p><p>As of now, we have also identified five trajectories related to our written problem statement that we could pursue in subsequent weeks:</p><ul><li>Understanding population sentiment</li><li>Identification of active threats</li><li>Monitoring own troops’ Internet activity for vulnerabilities</li><li>Aggregation and integration of multiple domains</li><li>Countering informational warfare and influence operations</li></ul><p><strong>All of these problems are exciting and challenging in their own way. </strong>Our main goal for next week (Week 3) is to connect with our H4D instructors and problem sponsor to 1) determine the likelihood of speaking with 10 AWG beneficiaries by the end of the week and 2) decide whether we should pivot to a different target customer. Based on the outcome of those discussions, we will come up with a list of hypotheses and key beneficiaries we’d like to interview.</p><p>Stay tuned for developments from Team Spyglass.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=daeb0ffe54ed" width="1" height="1" alt=""><hr><p><a href="https://medium.com/spyglass/a-first-dive-into-our-problem-statement-daeb0ffe54ed">A first dive into our problem statement</a> was originally published in <a href="https://medium.com/spyglass">Spyglass</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>