Author Archives: pgfeldman

Phil 11.15.17

7:00 – 4:30 ASRC MKT

  • How A Russian Troll Fooled America Reconstructing the life of a covert Kremlin influence account (Herding behavior???)
  • Psychological targeting as an effective approach to digital mass persuasion
    • People are exposed to persuasive communication across many different contexts: Governments, companies, and political parties use persuasive appeals to encourage people to eat healthier, purchase a particular product, or vote for a specific candidate. Laboratory studies show that such persuasive appeals are more effective in influencing behavior when they are tailored to individuals’ unique psychological characteristics. However, the investigation of large-scale psychological persuasion in the real world has been hindered by the questionnaire-based nature of psychological assessment. Recent research, however, shows that people’s psychological characteristics can be accurately predicted from their digital footprints, such as their Facebook Likes or Tweets. Capitalizing on this form of psychological assessment from digital footprints, we test the effects of psychological persuasion on people’s actual behavior in an ecologically valid setting. In three field experiments that reached over 3.5 million individuals with psychologically tailored advertising, we find that matching the content of persuasive appeals to individuals’ psychological characteristics significantly altered their behavior as measured by clicks and purchases. Persuasive appeals that were matched to people’s extraversion or openness-to experience level resulted in up to 40% more clicks and up to 50% more purchases than their mismatching or unpersonalized counterparts. Our findings suggest that the application of psychological targeting makes it possible to influence the behavior of large groups of people by tailoring persuasive appeals to the psychological needs of the target audiences. We discuss both the potential benefits of this method for helping individuals make better decisions and the potential pitfalls related to manipulation and privacy
  • Wrote up notes from yesterday
  •  (MIT) is a tool that tries to engage users in constructive debate. The questions were devised by Jonathan Haidt and his team for YourMorals.org – a site that collects data on moral sense.
    • CollectiveDebate
    • CollectiveDebate2
    • After using it some, it seems awkward, and requires a good deal of busywork. Much delayed gratification, and you seem to only select the arguments that work best for you. The visualizations, based on the 5 axis are pretty cool, could be some default axis to play with.
  • Continuing with From Keyword Search to Exploration – finished. Need to get my notes over from the Kindle, which is not posting them….
  • Banging away at Angular. Basically figuring out what I did yesterday. Ok, done. I think it makes more sense now.

Phil 11.14.17

7:00 – 4:00 ASRC MKT

  • Reinforcement Learning: An Introduction (2nd Edition)
    • Richard S. Sutton (Scholar): I am seeking to identify general computational principles underlying what we mean by intelligence and goal-directed behavior. I start with the interaction between the intelligent agent and its environment. Goals, choices, and sources of information are all defined in terms of this interaction. In some sense it is the only thing that is real, and from it all our sense of the world is created. How is this done? How can interaction lead to better behavior, better perception, better models of the world? What are the computational issues in doing this efficiently and in realtime? These are the sort of questions that I ask in trying to understand what it means to be intelligent, to predict and influence the world, to learn, perceive, act, and think. In practice, I work primarily in reinforcement learning as an approach to artificial intelligence. I am exploring ways to represent a broad range of human knowledge in an empirical form–that is, in a form directly in terms of experience–and in ways of reducing the dependence on manual encoding of world state and knowledge.
    • Andrew G. Barto : Most of my recent work has been about extending reinforcement learning methods so that they can work in real-time with real experience, rather than solely with simulated experience as in many of the most impressive applications to date. Of particular interest to me at present is what psychologists call intrinsically motivated behavior, meaning behavior that is done for its own sake rather than as a step toward solving a specific problem of clear practical value. What we learn during intrinsically motivated behavior is essential for our development as competent autonomous entities able to efficiently solve a wide range of practical problems as they arise. Recent work by my colleagues and me on what we call intrinsically motivated reinforcement learning is aimed at allowing artificial agents to construct and extend hierarchies of reusable skills that form the building blocks for open-ended learning. Visit the Autonomous Learning Laboratory page for some more details.
  • There was a piece on BBC Business Daily on social network moderators. Aside from it being a horrible job, the show touched on how international criminal cases often rest on video uploaded to services like Twitter and Facebook. This process worked as long as the moderators were human and could tell the difference between criminal activity and the documentation of criminal activity, but now with ML solutions being implemented, these videos are being deleted. First, this shows how ad-hoc the usage of these networks are as a place for legal and journalistic activity. Second, it shows the need for a mechanism that is built to support these activities, where there is a more expansive role of reporter/researcher and editor. This is near the center of gravity for the TACJOUR project.
  • Flying home yesterday, I was thinking about how the maps need to get built. One way of thinking about it is that you are given a set of directions that run through a geographic area and have to build a map from that. We know the adjacencies by the sequence of the directions. It follows that we should be able to build a map by overlaying all the routes in an n-dimensional space. I was then reading Technical Perspective: Exploring a Kingdom by Geodesic Measures, and at least some of the concepts appear related. In the case of the game at least, we have the center ‘post’, which is the discussion starting point. The discussion is (can be) a random walk towards the poles created in that iteration. Multiple walks create multiple paths over this unknown Manifold.  I’m thinking that this should be enough information to build a self organizing map. This might help: Visual analysis of self-organizing maps
    • Had some discussions with Arron about this. It should be pretty straightforward to build a map, grid or hex that trajectories can be recorded from. Then the trajectories can be used to reconstruct the map. Success is evaluated by the similarity between the source map and the reconstructed one.
    • I could also add recorded trajectories to the generated spreadsheet. It could be a list of cells that the agent traverses. Comparing explore, flocking and stampede behaviors in their reconstructed maps?
  • Continuing with From Keyword Search to Exploration
    • The mSpace Browser is a multi faceted column based client for exploring large data sets in the way that makes sense to you. You decide the columns and the order that best suits your browsing needs.
    • Yippy search
    • Exalead search
    • pg 62, animation
  • Continuing along with Angular
  • Multiple discussions with Aaron about next steps, particularly for anomaly detection

Phil 11.9.17

Instagram, Meme Seeding, and the Truth about Facebook Manipulation, Pt. 1

  • Jonathan Albright is the Research Director at the Tow Center for Digital Journalism. Previously an assistant professor of media analytics in the school of communication at Elon University, Dr. Albright’s work focuses on the analysis of socially-mediated news events, misinformation/propaganda, and trending topics, applying a mixed-methods, investigative data-driven storytelling approach.

  • The last couple of weeks have brought us the first new major revelations about the reach and scope of the IRA media influence campaign. Yet the most important development about the ongoing Facebook investigation isn’t the tenfold increase in the company’s updated estimate of the organic reach of “ads” on its platform.

    While the estimate increasing the reach of IRA content from 10 million people to 126 million people is surely a leap, after last week’s testimony, the real question we should be asking is: how did we suddenly arrive at 150 million?

    The answer is Instagram.

Reading The Group Polarization Phenomenon working on the PolarizationGame. Some thoughts:

  • There needs a way for each player to state their support/oppose state on a slider before the debate begins. We could even color code the threads using that information, though maybe only when viewing after the debate is complete.
  • What about teams?

The Emergence of a Fovea while Learning to Attend

  • Everything is about how we deal as individuals and groups with imperfect information. Which is why a attention-based economy is crazy

Identifying Dogmatism in Social Media: Signals and Models

  • We explore linguistic and behavioral features of dogmatism in social media and construct statistical models that can identify dogmatic comments. Our model is based on a corpus of Reddit posts, collected across a diverse set of conversational topics and annotated via paid crowdsourcing. We operationalize key aspects of dogmatism described by existing psychology theories (such as over-confidence), finding they have predictive power. We also find evidence for new signals of dogmatism, such as the tendency of dogmatic posts to refrain from signaling cognitive processes. When we use our predictive model to analyze millions of other Reddit posts, we find evidence that suggests dogmatism is a deeper personality trait, present for dogmatic users across many different domains, and that users who engage on dogmatic comments tend to show increases in dogmatic posts themselves.

 

Phil 11.8.17

ASRC MKT 7:00 – 5:00, with about two hours for personal time

  • After the fall of DNAinfo, it’s time to stop hoping local news will scale
    • I think people understand that this sensation of unreality has a lot to do with the platforms that deliver our news, because Facebook and Google package journalism and bullshit identically. But I’d argue that it also has a lot to do with the death of local news to a degree few of us recognize.
    • This is not unheard of in digital local news: People pay to drink with the investigative reporters at The Lens in New Orleans and to watch Steelers games with the staff of The Incline in Pittsburgh.
  • And as a counterbalance: Weaken from Within
    • The turtle didn’t know and never will, that information warfare — it is the purposeful training of an enemy on how to remove its own shell.
  • Rescuing Collective Wisdom when the Average Group Opinion Is Wrong
    • Yet the collective knowledge will remain inaccessible to us unless we are able to find efficient knowledge aggregation methods that produce reliable decisions based on the behavior or opinions of the collective’s members.
    • Our analysis indicates that in the ideal case, there should be a matching between the aggregation procedure and the nature of the knowledge distribution, correlations, and associated error costs. This leads us to explore how machine learning techniques can be used to extract near-optimal decision rules in a data-driven manner.
  • Inferring Relations in Knowledge Graphs with Tensor Decompositions
    • Ankur Padia
    • Multi-relational data, like knowledge graphs, are generated from multiple data sources by extracting entities and their relationships. We often want to include inferred, implicit or likely relationships that are not explicitly stated, which can be viewed as link-prediction in a graph. Tensor decomposition models have been shown to produce state-of-the-art results in link-prediction tasks. We describe a simple but novel extension to an existing tensor decomposition model to predict missing links using similarity among tensor slices, as opposed to an existing tensor decomposition models which assumes each slice to contribute equally in predicting links. Our extended model performs better than the original tensor decomposition and the non-negative tensor decomposition variant of it in an evaluation on several datasets.
  • From today’s Pulse of the Planet episode:
    • Colin Ellard is a cognitive neuroscientist and the author of Places of the Heart: the Psychogeography of Everyday Life. He says that the choices we make in siting a house or even where we choose to sit in a crowded room give us clues about the way humans have evolved.  The idea of prospect and refuge is an inherently biological idea. It goes back through the history of human beings. In fact for any kind of animal selecting a habitat, kind of the holy grail of good habitat choice can be summed up by the principal of seeing but not being seen.
      Ideally what we want is a set of circumstances where we have some protection, visual protection, in the sense of not being able to be easily located ourselves, and that’s Refuge. But we also want to be able to know what’s going on around us. We need to be able to see out from wherever that refuge is. And that’s Prospect. The operation of our preference for situations that are high in both refuge and prospect is something that cuts across everything we build or everywhere we find ourselves.
  • So, prospect-refuge theory sounds interesting. It seems to come from psychology rather than ecology-related fields. Still, it’s a discussion of affordances. Searching around, I found this: Methodological characteristics of research testing prospect–refuge theory: a comparative analysis. Couldn’t get it directly, so I’m trying ILL.
    • Prospect–refuge theory proposes that environments which offer both outlook and enclosure provoke not only feelings of safety but also of spatially derived pleasure. This theory, which was adopted in environmental psychology, led Hildebrand to argue for its relevance to architecture and interior design. Hildebrand added further spatial qualities to this theory – including complexity and order – as key measures of the environmental aesthetics of space. Since that time, prospect–refuge theory has been associated with a growing number of works by renowned architects, but so far there is only limited empirical evidence to substantiate the theory. This paper analyses and compares the methods used in 30 quantitative attempts to examine the validity of prospect–refuge theory. Its purpose is not to review the findings of these studies, but to examine their methodological bases and biases and comment on their relevance for future research in this field.
    • This is the book by Hildebrand: The Wright Space: Patterns and Meaning in Frank Lloyd Wright’s Houses. Ordered.
  • Ok, back to Angular2
    • Done with chapter 3.

Phil 11.7.17

7:00 – 6:00 ASRC MKT

  • Renting a spec Miata at Summit Point 
  • This is really good: The Human Strategy A Conversation With Alex “Sandy” Pentland [10.30.17]
    • Human behavior is determined as much by the patterns of our culture as by rational, individual thinking. These patterns can be described mathematically, and used to make accurate predictions. We’ve taken this new science of “social physics” and expanded upon it, making it accessible and actionable by developing a predictive platform that uses big data to build a predictive, computational theory of human behavior.
  • Rerunning the DTW with the selected agent weight being the specified weight rather than scaled by the distance from the angle so that it matches better the RANDOM_AGENT and the RANDOM_AGENTS settings.
  • Ok, here’s the results. The relationships between the populations appears more consistent, but that could be normal variability. Time for some true statistics to see if these are actually distinct populations. I can also increase power by doing more runs. Possibly also increasing the population size, though there might be confounding effects. DTWEqualWeight
  • Pandas can read in a specific Excel sheet and numpy can run bootstrap on DataFrames, so I can automate the analysis. Going to talk to Aaron first, since he might be the one to go down this road.
  • I think the next step is to start on the UI for the polarization game. Angular?
      • Installing NodeJS
      • npm install -g @angular/cli -> added 968 packages in 56.599s. That is a lot of packages. The IntelliJ plugin seems to be working, the @angular/cli package is visible: NodeNPM
      • Creating a new project is reasonable NewAngularProject
      • Once the project is running, the way to compile and run seems to be to run ng serve –open in the IntelliJ terminal (Note: When running as non-admin, do this in a terminal with admin privileges). It then does a whole bunch of things when a code change is made: 
        ** NG Live Development Server is listening on localhost:4200, open your browser on http://localhost:4200/ **
 10% building modules 8/10 modules 2 active ...\PolarizationGameOneUI\src\styles.csswebpack: wait until bundle finished: /                                                              Date: 2017-11-07T15:50:25.164Z
Hash: b3174f5198d14bdc05ac
Time: 4708ms
chunk {inline} inline.bundle.js (inline) 5.79 kB [entry] [rendered]
chunk {main} main.bundle.js (main) 20.8 kB [initial] [rendered]
chunk {polyfills} polyfills.bundle.js (polyfills) 553 kB [initial] [rendered]
chunk {styles} styles.bundle.js (styles) 33.8 kB [initial] [rendered]
chunk {vendor} vendor.bundle.js (vendor) 7.02 MB [initial] [rendered]

webpack: Compiled successfully.
webpack: Compiling...
Date: 2017-11-07T15:51:07.132Z
Hash: 7b89b5a301e4a411e92d
Time: 703ms
      • Everything is then sent to localhost:4200/, so all the browser debuggers are available
      • RunningAngular
      • And you can change the picture in the app.component.html file. re-renders on the fly. Pretty nifty. Yep verified:The ng serve command builds the app, starts the development server, watches the source files, and rebuilds the app as you make changes to those files.The --open flag opens a browser to http://localhost:4200/.
      • Pleasantly, if the install fails, ng serve –open will complete the install nd then start the server.
      • Added the ‘heroes’ component: AngularCLI AngularComponent
      • Then I got this error message: 
        ERROR in src/app/heroes/heroes.component.ts(7,18): error TS2304: Cannot find name 'ViewEncapsulation'.
      • Turns out that I had to add ViewEncapsulation to the imports in heroes.components: 
        import {Component, OnInit, ViewEncapsulation} from '@angular/core';

@Component({
  selector: 'app-heroes',
  templateUrl: './heroes.component.html',
  styleUrls: ['./heroes.component.css'],
  encapsulation: ViewEncapsulation.None
})
export class HeroesComponent implements OnInit {
  constructor() { }
  ngOnInit() {
  }
}

        Once added in, the rebuild happened and everything functioned normally. Correct error message in the IDE and everything!

  • Talked to Aaron about next steps with the herding data. We need to do something with NNs, and this could be a good fit
  • And now I have a nice little certificate of candidacy!

Phil 11.6.17

7:00 – 4:00 ASRC MKT

  • Going to try a batch job that runs the sim on a single population with a .2 radius and see if I can see a difference between the behaviors using DTW.
  • I had created a few bugs with changing the names of the flocks to Red and Green. Also, I had never run in batch mode with StorageAndRetreival. And calculations for an average center don’t work when there are no members of your flock. So fixing bugs.
  • First set of outputs from the batch jobs. Here’s the headings: HerdingHeadings
  • And here’s the DTW for the same settings (smaller stage though for proportionally greater differences): HerdingDTW
  • The first really obvious thing it that NoHerding is distinct from the other settings, which are more like echo chambers. Groupings tighten up as the radius increases, and the average heading approach may be statistically better than the random agents, but not by much. Lastly, RANDOM_AGENTS and RANDOM_AGENT lie on a continuum. As the switch between each agent takes longer, the more AGENTS will start to look like AGENT.

Phil 11.5.17

It’s a rainy day, so research.

Google TF Collaboratory 

I’m still looking at the Antifa/Nov search. Here’s what it looks like on Nov 5: AntifaNov5

To me, that looks a lot like the echo chamber hitting a respawn wall: EchoChamberAndTracesIn the above pictures, the recovery from the wall hit isn’t shown. I’m currently working on adding herding pieces, so I’ll the full chart later.

Got the random agents and random agent incorporated

Phil 11.3.17

7:00 – ASRC MKT

  • Good comments from Cindy on yesterday’s work
  • Facebook’s 2016 Election Team Gave Advertisers A Blueprint To A Divided US
  • Some flocking activity? AntifaNov4
  • I realized that I had not added the herding variables to the Excel output. Fixed.
  • DINH Q. LÊ: South China Sea Pishkun
    • In his new work, South China Sea Pishkun, Dinh Q. Lê references the horrifying events that occurred on April 30th 1975 (the day Saigon fell) as hundreds of thousands of people tried to flee Saigon from the encroaching North Vietnamese Army and Viet Cong. The mass exodus was a “Pishkun” a term used to describe the way in which the Blackfoot American Indians would drive roaming buffalo off cliffs in what is known as a buffalo jump.
  • Back to writing – got some done, mostly editing.
  • Stochastic gradient descent with momentum
  • Referred to in this: There’s No Fire Alarm for Artificial General Intelligence
    •  AlphaGo did look like a product of relatively general insights and techniques being turned on the special case of Go, in a way that Deep Blue wasn’t. I also updated significantly on “The general learning capabilities of the human cortical algorithm are less impressive, less difficult to capture with a ton of gradient descent and a zillion GPUs, than I thought,” because if there were anywhere we expected an impressive hard-to-match highly-natural-selected but-still-general cortical algorithm to come into play, it would be in humans playing Go.
  • In another article: The AI Alignment Problem: Why It’s Hard, and Where to Start
    • This is where we are on most of the AI alignment problems, like if I ask you, “How do you build a friendly AI?” What stops you is not that you don’t have enough computing power. What stops you is that even if I handed you a hypercomputer, you still couldn’t write the Python program that if we just gave it enough memory would be a nice AI.
    • I think this is where models of flocking and “healthy group behaviors” matters. Explore in small numbers is healthy – it defines the bounds of the problem space. Flocking is a good way to balance bounded trust and balanced awareness. Runaway echo chambers are very bad. These patterns are recognizable, regardless of whether they come from human, machine, or bison.
  • Added contacts and invites. I think the DB is ready: polarizationgameone
  • While out riding, I realized what I can do to show results in the herding paper. There are at least three ways to herd:
    1. No herding
    2. Take the average of the herd
    3. Weight a random agent
    4. Weight random agents (randomly select an agent and leave it that way for a few cycles, then switch
  • Look at the times it takes for these to converge and see which one is best. Also look at the DTW to see if they would be different populations.
  • Then re-do the above for the two populations inverted case (max polarization)
  • Started to put in the code changes for the above. There is now a combobox for herding with the above options.

Phil 11.2.17

ASRC MKT 7:00 – 4:30

  • Add a switch to the GPM that makes the adversarial herders point in opposite directions, based on this: Russia organized 2 sides of a Texas protest and encouraged ‘both sides to battle in the streets’
  • It’s in and running. Here’s a screenshot: 2017-11-02 There are some interesting things to note. First, the vector is derived from the average heading of the largest group (green in this case). This explains why the green agents are more tightly clustered than the red ones. In the green case, the alignment is intrinsic. In the red case, it’s extrinsic. What this says to me is that although adversarial herding works well when amplifying the heading already present, it is not as effective when enforcing a heading that does not already predominant. That being said, when we have groups existing in opposition to each other, that is a tragically easy thing to enhance.
  • Hierarchical Representations for Efficient Architecture Search
    • We explore efficient neural architecture search methods and present a simple yet powerful evolutionary algorithm that can discover new architectures achieving state of the art results. Our approach combines a novel hierarchical genetic representation scheme that imitates the modularized design pattern commonly adopted by human experts, and an expressive search space that supports complex topologies. Our algorithm efficiently discovers architectures that outperform a large number of manually designed models for image classification, obtaining top-1 error of 3.6% on CIFAR-10 and 20.3% when transferred to ImageNet, which is competitive with the best existing neural architecture search approaches and represents the new state of the art for evolutionary strategies on this task. We also present results using random search, achieving 0.3% less top-1 accuracy on CIFAR-10 and 0.1% less on ImageNet whilst reducing the architecture search time from 36 hours down to 1 hour.
  • Continuing with the schema. Here’s where we are today: polarizationgameone

Phil 11.1.17

Phil 7:00 – ASRC MKT

    • The identity of the machine is just as important as the identity of the human, argues Jeff Hudson.
    • Agent-based simulation for economics: The Tool Central Bankers Need Most Now
    • Introducing Vega-Lite 2.0 (from MIT Interactive Data Lab)
      • Vega-Lite enables concise descriptions of visualizations as a set of encodings that map data fields to the properties of graphical marks. Vega-Lite uses a portable JSON format that compiles to full specifications in the larger Vega language. Vega-Lite includes support for data transformations such as aggregation, binning, filtering, and sorting, as well as visual transformations such as stacking and faceting into small multiples.
    • Wayne says ‘awareness’ is too overloaded, at least in CSCW where it means ‘a shared awareness’. What about alertness, cognition, or perception?
    • Started Simulating Flocking and Herding in Belief Space. Shared with Wayne, Aaron and Cindy
    • Yay, finally got the array problems solved. The problem is that a PHP array is actually a set. But you can convert any set into a zero-indexed array using array_values(). So now all my arrays begin at zero, as God intended.
    • Meeting with the lads. Some really good stuff.
      • Add tmanage
        • dungeon_master
        • game
        • scenario
        • min_players
        • max_players
        • time_to_live
        • state (waiting, running, timeout, terminated, success)
        • open (true/false)
        • visible
      • Add trating
        • target_message
        • relevance
        • quality
        • vote
        • rating_player
      • Add ttopics
        • title
        • description
        • parent
      • Add tplayerstate
        • player
        • game
        • state (waiting, playing, finished, terminated)
      • Add tcontact
        • player
        • name
        • email
        • facebook (oAuth)
        • google (oAuth)
      • Add tinvite
        • contact
        • game
        • player

 

  • Humans + Machines (CNAS livestream)
    12:30 – 1:35 PM
    Dr. Jeff Clune, Assistant Professor of Computer Science, University of Wyoming
    Kimberly Jackson Ryan, Senior Human Systems Engineer, Draper Laboratory
    Dr. John Hawley, Engineering Psychologist, Army Research Laboratory
    Dr. Caitlin Surakitbanharn, Research Scientist, Purdue University
    Dan Lamothe, National Security Writer, The Washington Post (moderator)

Phil 10.31.17

7:00 – 4:30 ASRC MKT

    • Wrote up notes from yesterday’s meeting
    • Look for JCMC requirements
    • Change the rest of the “we” to “I” in the DC, then submit. Done, did a spell check because I had forgotten to integrate a spell checker!
    • Saw this today on the Google Research Blog: Closing the Simulation-to-Reality Gap for Deep Robotic Learning. In it they show how simulation can be used to improve deep learning because of the vast increase in conditions that can be simulated rather than found or built in the real world. The reason that it’s important in my work is that the simulation can feed and support the training of the classifiers once the simulation becomes sufficiently realistic.
    • Because I can’t stop reading horrible things, ordered Totalitarianism, Terrorism and Supreme Values: History and Theory, by  Peter Bernholz
    • Not the most exciting thing, but yay! 
      ID	posted		message					playerID	parentID
1	1509458541	message 0 of 20 by Abbe, Karleen	5	6	
2	1509458541	message 1 of 20 by Abbey, Abbi	7	6	
3	1509458541	message 2 of 20 by Abbey, Abbi, responding to message 1	7	6	2
4	1509458542	message 3 of 20 by Abbe, Karleen, responding to message 2	5	6	3
5	1509458542	message 4 of 20 by Abbe, Karleen, responding to message 1	5	6	2
6	1509458542	message 5 of 20 by Abbe, Karleen, responding to message 4	5	6	5
7	1509458542	message 6 of 20 by Abbe, Karleen, responding to message 3	5	6	4
8	1509458542	message 7 of 20 by Abbe, Karleen, responding to message 1	5	6	2
9	1509458542	message 8 of 20 by Abbe, Karleen, responding to message 1	5	6	2
10	1509458542	message 9 of 20 by Aaren, Abbie, responding to message 2	3	6	3
11	1509458542	message 10 of 20 by Abbey, Abbi, responding to message 5	7	6	6
12	1509458542	message 11 of 20 by Abbe, Karleen, responding to message 10	5	6	11
13	1509458542	message 12 of 20 by Abbey, Abbi, responding to message 7	7	6	8
14	1509458542	message 13 of 20 by Aaren, Abbie	3	6	
15	1509458542	message 14 of 20 by Abbe, Karleen, responding to message 8	5	6	9
16	1509458542	message 15 of 20 by Abbe, Karleen, responding to message 11	5	6	12
17	1509458542	message 16 of 20 by Abbe, Karleen	5	6	
18	1509458542	message 17 of 20 by Abbe, Karleen, responding to message 4	5	6	5
19	1509458542	message 18 of 20 by Aaren, Abbie, responding to message 14	3	6	15
20	1509458542	message 19 of 20 by Aaren, Abbie, responding to message 2	3	6	3
    • cleaning up some cases where scenario is set to null. Fixed. It’s the first array index problem. Grrrrr. Ok, broke some things trying to make things better….
    • Then it’s time to make some REST interfaces
    • Meeting with Cindy. Much progress!
      • User-specified scenarios, seeded with some fun topics like conspiracy theories
      • Private deliberations.
      • Esperanto for verdict: verdikto
      • Lobbies for collecting users
      • Game starts when an DM-specified minimum is met, though there may be time to accumulate into a max as well
      • Game ‘dies’ if no contribution (by all players?) in a certain window
      • One user can kill a game by withdrawing. This can be attached to a user (troll), so the player can anonymously block in the future
      • Games can be respawned, optionally without a triggering troll from the last time
      • Games/Scenarios can be cloned
      • Highest-quality games that reach a verdict are featured on the site. Quality could be determined by tagging or NLP+heuristics.

 

Phil 10.30.17

7:00 – 4:30 ASRC MKT

  • The discussion and conclusion
  • Tweaked the “Future Work” section of the CHIIR DC proposal to reflect the herding work. More words means less bullet points!
  • Updated Java and XAMMP on my home machine
  • Pointed the IDE at the correct places
  • I don’t think I have PhpInspections (EA Extended) installed at work? It does nice things – Have it now
  • Working through creating a strawman game. Having some issues with a one-to-many relationship with RedBeanPHP. Ah – it’s because you have to sync the beans. I think rather than have a game point at all the players, I’ll have the players point at the scenario, and the chat messages point at the game and players.
  • Got that mostly working, but having a null player issues
  • Important PHP issue – arrays don’t need to start at zero! The bean arrays are indexed with respect to their db id!
  • Meeting with Wayne
  • The DC is good to submit
  • Start working on a JCMC article that connects the flocking model to qualitative theory.
  • Keep on working on the game. Possible project for a class/group in either 729 – design and evaluate class (Komlodi) or 728 – Online Communities & Social Media (Branham)

Phil 10.27.17

7:00 – 5:00 ASRC MKT

  • Nicely written paper on GANs:
    • Abstract: We describe a new training methodology for generative adversarial networks. The key idea is to grow both the generator and discriminator progressively: starting from a low resolution, we add new layers that model increasingly fine details as training progresses. This both speeds the training up and greatly stabilizes it, allowing us to produce images of unprecedented quality, e.g., CELEBA images at 10242. We also propose a simple way to increase the variation in generated images, and achieve a record inception score of 8.80 in unsupervised CIFAR10. Additionally, we describe several implementation details that are important for discouraging unhealthy competition between the generator and discriminator. Finally,we suggest a new metric for evaluating GAN results, both in terms of image quality and variation. As an additional contribution, we construct a higher-quality version of the CELEBA dataset.
    • With cool video
    • And code
  • Working on adding UI and batch interaction for the adversarial herding
    • Enable/disable switch – Done
    • Field for power – don’t know what the scale should be so no slider yet – Done
    • Set<String, Set<Flockingshape, weight>> If this doesn’t work, make shape comparable by name. Done! 
      HashMap<FlockingShape, Double> alignedShapeMap;
if(flock.size() > 0 && !alignedFlockMap.containsKey(flockName)){
    alignedShapeMap = new HashMap<>();
    alignedFlockMap.put(flockName, alignedShapeMap);
}else{
    alignedShapeMap = alignedFlockMap.get(flockName);
}
    • Do I want to delay the triggering of the herding on a separate timer? Waiting on this.
    • It’s done, and the results are kind of scary. If I set the weight of the herder to 15, I can change the change the flocking behavior of the default to echo chamber.
    • Normal: No Herding
    • Herding weight set to 15, other options the same: HerdingWeight15
  • Did some additional tweaking to see if having highly-weighted herders ignore each other (they would be coordinated through C&C) would have any effect. It doesn’t. There is enough interaction through the regular populations to keep the alignment space reduced.
  • It looks like there is a ‘sick echo chamber’ pattern. If the borders are reflective, and the herding weight + influence radius is great enough, then a wall-hugging pattern will emerge.
    • The influence weight is sort of a credibility score. An agent that has a lot of followers, or says a lot of the things that I agree with has a lot of influence weight The range weight is reach.
    • Since a troll farm or botnet can be regarded as a single organization,  interacting with any one of the agents is really interacting with the root entity.  So a herding agent has high influence and high reach. The high reach explains the border hugging behavior.
    • It’s like there’s someone at the back of the stampede yelling YOUR’E GOING THE RIGHT WAY! KEEP AT IT! And they never go off the cliff because they are a swarm Or, it never goes of the cliff, because it manifests as a swarm.
    • A loud, distributed voice pointing in a bad direction means wall hugging. Note that there is some kind of floating point error that lets wall huggers creep off the edge.Edgecrawling
    • With a respawn border, we get the situation where the overall heading of the flock doesn’t change even as it gets destroyed as it goes over the border. Again, since the herding algorithm is looking at the overall population, it never crosses the border but influences all the respawned agents to head towards the same edge: DirectionPreserving
  • Paper thoughts:
    • Armys have different patterns from emergent groups. They are imposed formations and reflect a commander’s will
    • From a distance, they look different, but close up, they may look the same. One of the reasons for the success of the Roman Legion was the use of formations against the less sophisticated structures of their adversaries [ref]

Phil 10.26.17

7:00 – 3:30 ASRC MKT

  • Listening to BBC Business Daily this morning on Facebook vs Democracy:
    • Presenter Ed Butler hears a range of voices raising concern about the existential threat that social media could pose to democracy, including Ukrainian government official Dmytro Shymkiv, journalist Berit Anderson, tech investor Roger McNamee and internet pioneer Larry Smarr.
  • …and had some thoughts on adversarial herding in information space
    • Herders can teleport, since they are not emotionally invested in their belief space position and orientation
    • Herders appear like multiple individuals that may seem close and trustworthy, but they are actually a distant monolithic entity that is aware of a much larger belief space.
    • Herders amplify the most extreme positions and may also amplify opposition. The insight is that they are not herding in a direction, but to increase polarization
    • To add this to the model, I need to do the following:
      • Make the size of the agent a function of the weight
      • When in ‘herding mode’ the overall heading of the group is calculated, and the agent that is furthest in that direction is selected
      • The weight is increased to X, and the radius is increased to Y.
        • X represents amplification of the concept, by trolls, bots, etc.
        • A large Y means that the bots can swamp other, normally closer signals. This models the effect of a monolithic entity controlling thousands of bots across the belief space
    • Got it! Note that the influence radius is 1/3 of the range normally needed to polarize. Also note that the amplified agent is not leading. This reflects Arendt’s insight that totalitarian rulers follow and amplify the mobTrollfarm I need to make it so that there is UI support (on/off, amount), and make it so that each flock can have its own bots.
    • I expect this to produce extreme polarization (low time to border) more quickly than emergent, ‘organic’, echo chambers
    • This may describe some applicable group hunting behavior: Predator-prey interactions in two schooling fishes, Caranx ignobilis and Stolephorus purpureus
      • Interactions between the jack, Caranx ignobilis, a facultative schooling species, and the Hawaiian anchovy, Stolephorus purpureus, an obligate schooler were studied within an enclosure in the field in Hawaii. Single predators were the most successful at capturing isolated (individual) prey, and relatively unsuccessful at capturing individuals in schools. Grouped (schooled) predators were the most successful at capturing schooled prey. The leading, or first, predator was the most successful member of a group or school at capturing isolated or schooled prey. Following predators tended to make it possible to catch more prey earlier in the experiments. Larger predator groups were able to break up schools of prey quickly, resulting in increased numbers of prey becoming isolated. These prey were captured before they could reform or join other schools. As the initial size of the prey school increased, the per cent of individuals captured declined. Schooling in prey reduces the time a visually orienting predator has to align himself with an individual prey. Schooling in predators may have co-evolved as an adaptation, making it possible for predators to break up and isolate schooled prey. Larger prey schools may have co-evolved to satiate or swamp the feeding capacity of a finite number of schooled predators and decrease the probability that a specific given individual would be captured.
  • Finishing up Suppressing the Search Engine Manipulation Effect

Phil 10.25.17

 

7:00 – 1:00 ASRC MKT

  • Collective Agency and Cooperation in Natural and Artificial Systems (some chapters:)
    • The Participatory Turn: A Multidimensional Gradual Agency Concept for Human and Non-human Actors
    • Collective Agency and Cooperation in Natural and Artificial Systems
    • Planning for Collective Agency
    • Simulation as Research Method: Modeling Social Interactions in Management Science
    • How Models Fail
    • Requested the library copy. Request #426721
  • Continuing with Suppressing the Search Engine Manipulation Effect
    • One quick thought on Likert scales: If the selection is a slider, then we can see how the user interacts with the slider. This could let us see how decisive the users are. This could also be dome by looking at the area surrounding the radio buttons and tracking mouse motion and number of clicks in the area
    • Introduction
      • Recent research has shown that society’s growing dependence on ranking algorithms leaves our psychological heuristics and vulnerabilities susceptible to their influence on an unprecedented scale and in unexpected ways
      • Experiments conducted on Facebook’s Newsfeed have demonstrated that subtle ranking manipulations can influence the emotional language people use
      • Similarly, experiments on web search have shown that manipulating election-related search engine rankings can shift the voting preferences of undecided voters by 20% or more after a single search
      • While “bias” can be ambiguous, our focus is on the ranking bias recently quantified by Kulshrestha et al. with Twitter rankings
      • Our results provide support for the robustness of SEME and create a foundation for future efforts to mitigate ranking bias. More broadly, our work adds to the growing literature that provides an empirical basis to calls for algorithm accountability and transparency [24, 25, 90, 91] and contributes a quantitative approach that complements the qualitative literature on designing interventions for ranking algorithms
      • Our results also suggest that proactive strategies that prevent ranking bias (e.g., alternating rankings) are more effective than reactive strategies that suppress the effect through design interventions like bias alerts. Given the accumulating evidence, we speculate that SEME may be impacting a wide range of decision-making, not just voting
    • Related Work
      • Order effects are among the strongest and most reliable effects ever discovered in the psychological sciences [29, 88]. These effects favorably affect the recall and evaluation of items at the beginning of a list (primacy) and at the end of a list (recency).
        • There does not seem to be an equivalent primacy effect in maps that I can find
      • online systems can: (1) provide a platform for constant, large-scale, rapid experimentation, (2) tailor their persuasive strategies by mining detailed demographic and behavioral profiles of users [1, 6, 9, 18, 121], and (3) provide users with a sense of control over the system that enhances their susceptibility to influence
        • Is this flocking from the flock’s perspective? Sort of an Ur-flock?
        • This is that Trust/Awareness equation again
      • A recent report involving 33,000 people found that search engines were the most trusted source of news, with 64% of people reporting that they trust search engines, compared to 57% for traditional media, 51% for online media, and 41% for social media [10]. Similarly, a 2012 survey by Pew found that 73% of search engine users report that “all or most of the information they find is accurate and trustworthy,” and 66% report that “search engines are a fair and unbiased source of information” [105].
      • Suggestions for fostering resistance can be broken down into two primary strategies: (1) providing forewarnings [43, 49] and (2) training and motivating people to resist [79, 120].
        • Interesting that alternate, non-ordered design approaches aren’t even mentioned
      • Part of the reason that forewarnings work is explained by psychological reactance theory [12], which posits that when people believe their intellectual freedom is threatened – by exposing an attempt to persuade, for example – they react in the direction opposite that of the intended one
      • In the context of online media bias, researchers have primarily explored methods for curbing the effects of algorithmic filtering and selective exposure [87, 96] rather than ranking bias [71]. In this vein, researchers have developed services that encourage users to explore multiple perspectives [97, 98] and browser extensions that gamify and encourage balanced political news consumption [19, 20, 86]. However, these solutions are somewhat impractical because they require users to adopt new services or exert additional effort.
    • Methods – Experiment Design
      • To construct biased search rankings we asked four independent raters to provide bias ratings of the webpages we collected on an 11-point Likert scale ranging from -5 “favors Cameron” to +5 “favors Miliband”. We then selected the 15 webpages that most strongly favored Cameron and the 15 that most strongly favored Miliband to create three bias groups
      • The query in the search engine was fixed as “UK Politics ‘David Cameron’ OR ‘Ed Miliband’”, and subjects could not reformulate it.
      • On top of assignment to a bias group, subjects were randomly assigned to one of three alert experiments.We drew from the literature on decision-making and design intervention to implement so-called debiasing strategies for improving decision-making in the presence of biased information [39, 78, 82]. Specifically, we constructed and placed alerts in the search results produced by our mock search engine that provided forewarnings with salient graphics, autonomony-supportive language, and details on the persuasive threat
    • Methods – Procedure
      • After providing informed consent and answering basic demographic questions
        • Do this and use this phrase!
      • Subjects then rated the two candidates on 10-point Likert scales with respect to their overall impression of each candidate, how much they trusted each candidate, and how much they liked each candidate. Subjects also indicated their likelihood of voting for one candidate or the other on an 11-point Likert scale where the candidates’ names appeared at opposite ends of the scale and 0 indicated no preference, as well as on a binary choice question where subjects indicated who they would vote for if the election were held today.
        • This is a good way to set up the game. People read the dilemma, formulate an initial solution and their level of commitment to it. They can choose to make it “public” as their first statement or to keep it private and display a “no opinion” initial statement
      • We asked: “While you were doing your online research on the candidates, did you notice anything about the search results that bothered you in any way?” and prompted subjects to explain what had bothered them in a free response format: “If you answered “yes,” please tell us what bothered you.” We did not directly ask subjects whether they had “noticed bias” to avoid the inflation of false positive rates that leading questions can cause
    • Methods – Participants
      • We recruited 3,883 subjects between April 28, 2015 and May 6, 2015 on Amazon’s Mechanical Turk (AMT; https://mturk.com), a subject pool frequently used by behavioral, economic, and social science researchers [8, 13, 102]. We excluded from our analysis subjects who reported an English fluency level of 5 or less (on a scale of 1 to 10) (n=26)
        • MTurk would be a good source of participants as well
    • Analysys – Search metrics
      • Utilizing Kolmogorov-Smirnov (K-S) tests of differences in distributions, we found significant differences in the patterns of time spent on the 30 webpages between subjects in the no alert experiment (correlation with ranking: Spearman’s ρ = -0.836, P <0.001) and the high alert experiment (ρ = -0.654, P <0.001) (K-S D = 0.467, P <0.01), and between subjects in the low alert experiment (ρ = -0.719, P <0.001) and the high alert experiment (K-S D = 0.400, P <0.01)
        • A way of looking for explore/exploit populations? And how fast can it be determined? Google uses a mechanism to stop an experiment once a confidence level is reached. Also, bootstrap would be good here
      • Similarly, we also found significant differences in the patterns of clicks that subjects made on the 30 webpages between subjects in the no alert experiment (ρ = -0.865, P <0.001) and the high alert experiment (ρ = -0.795, P <0.001) (K-S D = 0.500, P <0.001), and between subjects in the low alert experiment (ρ = -0.876, P <0.001) and the high alert experiment (K-S D = 0.367, P <0.05)
      • Among all conditions,we found that differences in the patterns of time and clicks on the individual rankings primarily emerged on the first SERP, but less so on the second, fourth, and fifth SERPs
    • Analysys – Attitude Shifts
      • we found that the mean shifts in candidate ratings for the bias groups significantly converged on the mean shift found in the neutral group as the level of detail in the alerts increased, with high alerts creating higher convergence than low alerts
        • As more diverse information is injected, populations compromise
    • Analysys – Vote Shifts
      • Vote Manipulation Power (VMP)is the percent change in the number of subjects, in the two bias groups combined, who indicated that they would vote for the candidate who was favored by their search rankings. That is, if x and x ′ subjects in the bias groups said they would vote for the favored candidate before and after conducting the search, respectively, then VMP = (x ′ − x)/x.
        • This could also be applied to the game to watch how votes for an outcome change over time. In the case of the game, new candidates can come into existence, so we need to watch for that.
    • Analysys – Bias Awareness
      • We found 8.1% of subjects that showed awareness of the bias in the no alert experiment, a figure identical to the 8.1% awareness rate found by Eslami et al. in their audit of Booking.com [37], and similar to the 8.6% of subjects who showed awareness in the original study [30]. The percentage of subjects showing bias awareness increased to 21.5% in the low alert experiment, and 23.4% in the high alert experiment.
    • Discussion
      • However, despite the additional suppression of the high alerts, the lowest VMP was found among the neutral group subjects: rankings alternating between favoring the two candidates prevented SEME.
        • This configuration forces users to “explore” more, within the context of a list affordance.
      • As with previous research on SEME [30], and with research on attitude change and influence more generally [3, 72, 120], we found that subjects vary in their susceptibility to SEME, as well as in their responsiveness to the alerts, based on their personal characteristics (Figure 6 and Figure 7 in the Appendix).
        • Explorer and exploiter populations?
      • As more people turn to the internet for political news [85, 115], designing systems that can monitor and suppress the effects of algorithm biases, like ranking bias, will play an increasingly important role in protecting the public’s psychological vulnerabilities.
        • And one of the big issues is finding bias at scale with domain independence
      • Real-time automated bias detection could potentially be achieved by utilizing a Natural Language Processing (NLP) approach. One could utilize opinions [75], sentiment [99], linguistic patterns [109], word associations [14], or recursive neural networks [59] with human-coded data to classify biased language.
        • Scale and domain problems.
      • Discussion – Awareness of bias
        • Awareness of ranking bias appears to suppress SEME only when it occurs in conjunction with a bias alert, perhaps because an alert is a kind of warning–inherently negative in nature.
          • According to Moscovici, an inherently negative construct should reduce polarization movement.
        • Awareness of ranking bias in the absence of bias alerts might increase VMP because people perceive the bias as a kind of social proof [111, 112], made all the more powerful because of the disproportionate trust people have in search rankings [10, 95, 105]. The user’s interpretation might be, “This candidate MUST be good, because even the search results say so.”
  • Polarization Game
    • Upgraded to PHP7. I went with the threadsafe version, which meant that I had to upgrade xdebug as well. and for some reason, I had to put the php_xdebug.dll file in the ext directory.
    • And now I have typing!
  • Proposal Work (1:00 – 5:00)
    • Skimmed the RFI, thought of Rick Satava
    • Building out template
    • Wrote the first pass with Aaron