1. Federated AI for
Communication & Trust
A Proposal in Response to AI and the News Challenge
Proposed by
Sabine Brunswicker & Alok Chaturvedi, Purdue University
2. The Federated-AI for Communication and
Trust (FACT) platform is envisioned to be
an open, federated AI that will
continuously evaluate professional and
civic news streams, assign trust rating to
content and sources, and progressively
adjust their ratings as new and also
untold stories are generated and shared.
What’s the FACT idea?
3. How does FACT federate?
FACT episodic
memory engine
FACT
intelligence
engine
• Generates
narratives based on
facts
• Produces
counterfactuals and
”untold” stories
• Rates trust rating of
new articles
• Stores historical facts
and sources of news
• Traces news content
and degree of
distortion over time
• Assigns trust rating
FACT integrates an assemble of AI models that augment the memory and the
“intelligence” of users. It’s a distributed AI platform that federates through
self-organization and novel human-AI interaction design
4. What’s the target audience of FACT?
Citizens (news
consumers) Professional journalists
Civic writers
The primary audience of FACT are ~ 30 % of the citizens living in the U.S, who are active on
the social media.
1 2 3 AI community and
researchers on AI
and the news
• Literacy about AI and
distortion of news
content
• Increased trust in news
• Confidence in writing
trustworthy stories
• Attention to untold
stories
• New models for federated AI
combining agent-based
modeling, GAN, and human-AI
interaction/design
• Behavioral studies/insights
5. The FACT AI framework is based on principles of self-
organization
World
Events
Other AI
Green Narrative
Red Narrative
Blue Narrative
• Economic
• Rule of Law
• Institution &
Infrastructure
Individual
Social Group
Recall augmented with
episodic memory (Hebbian
learning)
Trust
Model
Perceived trust
Desired trust
Trust rating
########
########
########
Actions augmented by
Intelligence (GAN)
Social
Media
Professional
Media
Context of Event
• Political Stability
• Social Stability
• Security..…
Word of Month
Media or
Direct Sense
Behavioral
implications
• Perceptions
• Attitudes &
Beliefs
• Actions
Augmented by FACT
Single Actions
• Propagate/retweet
• Influence
• Support narrative
• Generate Narrative
• …
6. FACT takes a unique focus on human-AI interaction
ANALYTICS/ALGORITHMS
Cognitive complexity of ratings
(e.g., descriptive, predictive).
Contextual Information
(e.g., transparency of algorithm).
VISUALIZATION
Visual Grammar Features
(e.g., color, shape, layout).
Interactions
(e.g., real-time, alert, automatic)
TEXT & language
(e.g., positive, negative sentiment)
FACT Interface
End-user Behavior
Perception
(e.g., attention to
facts)
1x Action Continued Action
(e.g. more fact-based retweets,
more untold stories)
7. What are the deliverables of FACT?
The first year of the FACT project will generate three innovative and tangible deliverables
Novel FACT platform
distributed AI with novel
human-AI interaction design
• Memory engine (Hebbian
learning & agent-based
simulation
• Intelligence engine
(generative adversarial
networks & agent-based
modeling)
• Human-AI interaction
design
Data and evaluation from
controlled experiments with
500+ citizens
• Lab & field experiments
• Testing effect of FACT for
citizens vis-à-vis
professional writers
• Impact measures: literacy,
perceived trust, confidence,
and behavioral change
• ~250 new FACT-based
stories
1 2 3
FACT channel
reporting to broader
audience
• News produced
with FACT by
citizen journalists
• Co-authored work
with advisory
board
8. The team behind FACT
Core Team Additional team members (excerpt)
The core team and its team members is diverse and interdisciplinary spanning AI, statistics, and
social science. We have have gathered experiences in developing and applying computational
methods, including AI and machine intelligence, in a social and communication context.
Dr.Sabine Brunswicker
• Associate ProfessorforDigitalInnovation,PurduePolytechnicandBrian
SchoolofCommunication,DirectoroftheResearchCenterforOpenDigital
Innovation(www.rcodi.org),AAAIaffiliate
• Computationalsocialscience(agent-basedsimulation,networkdynamics,
AI)andsocialcomputing,behavioralimplicationsofplatformsand
algorithmicfeedback,“digitalnudging”,largescale experiments
• Involvedthousandsofcitizensinthedesign&evaluationsofdigitalplatforms
Dr.AlokChaturvedi
• ProfessorinPurdueUniversity’sKrannertSchoolofManagementandthe
DepartmentofComputerSciences;DirectoroftheInstituteforSocial
EmpowermentthroughEntrepreneurshipandKnowledge(ISEEK)
(https://krannert.purdue.edu/centers/iseek/).
• AhsdevelopedandappliedAItechniquesforover30years,intensiveexperiencein
largescale modelinganddistributedAI,agent-basedmodeling
• DevelopedAIfortheUSgovernmentandNATO: e.g.theSyntheticEnvironmentfor
AnalysisandSimulation– VirtualInternationalSystem(SEAS-VIS),wasdeployedin
IraqandAfghanistantotestcommunicationstrategies(https://goo.gl/ETFuVa).
Jay Cheon,
ComputerScience,Student, AI&Computational Science
GoogleDeep Mind Scholars
SatyamMukherjee
Physics
ProfessorDataScience,India
Rachael Tan
ComputerEngineering, Student
AmateurJournalist
Advisory Board Members & Partners (in progress)
• Opennews.org,ErikaOwens
• Washington Post,tbd
• Kate Crawford,AIand SocietyInstitute, NYU
9. Some examples of our Publication in AI, computational sciences and feedback design (more upon request)
1) Wan-Lin Hu, Janette Meyer, Zhaosen Wang, Tahira Reid, Alok Chaturvedi, Douglas Adams and Sunil Prabhakar, “Dynamic
Data Driven Approach for Modeling Human Error,” International Conference on Computational Science, Reykjavik, Iceland, June
2015
2) Chaturvedi, R., B. Armstrong, A. Chaturvedi, D. Dolk, and P. Drnevich, “Got a Problem? Agent-based modeling becomes
mainstream,” Global Economics and Management Review, Vol 18:2, pp 33-39, 2013.
3) Chaturvedi, A. R., Dolk, D. R; Drnevich, P. L., “Design Principles for Virtual Worlds,” MIS Quarterly, Vol. 35, No. 3, pp 673-684,
2011.
4) Shawn C. McKay, Alok Chaturvedi, Douglas E. Adams, “A Process for Anticipating and Shaping Adversarial Behavior,” Naval
Research Logistics, Vol. 58, No. 3, pp 255-280, 2011.
5) Moskowitz, H., Drnevich, P., Ersoy, O., Altinkemer, K., & Chaturvedi, A. Using Real-Time Decision Tools to Improve Distributed
Decision Making Capabilities in High-Magnitude Crisis Situations. Decision Sciences Vol. 42, No. 2, 477-493, 2011.
6) Chaturvedi, A., “Society of Simulation Approach to Dynamic Integration of Simulations,” Winter Simulation Conference,
Monterey California, December 3-6, 2006.
7) Chaturvedi, A. R., Chaturvedi, R., and Dolk, D. R., “Agent Based Modeling of International System,” Agent 2004, University of
Chicago, 2004.
8) Chaturvedi, A.R., "FMS Scheduling and Control: An AI Approach to Achieve Multiple Decision Goals," Expert Systems with
Applications: International Journal, vol. 6, pp. 267-286, 1993.
9) Chaturvedi, A.R., "Acquiring Implicit Knowledge in a Complex Domain," Expert Systems with Applications: International
Journal, vol. 6. pp. 23-35, 1993.
10) Chaturvedi, A.R., Hutchinson, G. K. and Nazareth, D. "Supporting Complex Real-Time Decision Making through Machine
Learning," Decision Support Systems, vol 9, pp. 1-21, 1993.
Our experience in AI, system design, and its behavioral
implications
10. Some examples of our Publication in AI, computational sciences and feedback design (more upon request)
1) Brunswicker, S., Almirall, E., & Majchrzak, A. (forthcoming). The interplay between platform architecture and producers’ design
strategies for platform performance. An agent-based simulation. Management Information Systems Quarterly
2) Promann, M., & Brunswicker, S. (2018, August). The effect of proximal data structure on perceived group unity. IEEE Visual
Analytics Science and Technology (VAST). Berlin, Germany (acceptance rate of 30.4 %, Google Scholar h5-index 15, Ranking
2017: rank #12 in Computer Graphics).
3) Swan, M., and Brunswicker, S. (2018, August). Blockchain Economic Networks and Algorithmic Trust. Proceedings of the
Americas Conference on Information Systems (AMCIS), New Orleans: Association of Information Systems.
4) Brunswicker, S., Jensen†, B., Song, C., & Majchrzak, A. (2018). Transparency as design choice of open data contests. Journal
of the Association of Information Science and Technology.
5) Brunswicker, S., Matei, S. A., Zentner, M. G., Zentner, L. K., & Klimeck, G. (2017). Creating impact in the digital space: digital
practice dependency in communities of digital scientific innovations. Scientometrics, 110(1), 417-426.
6) Brunswicker, S., & Prietula, M. (2017, August). Transparency and reuse in digital innovation contests: A simulation study.
Proceedings of the 76th Annual Meeting of Academy of Management. Atlanta, Georgia: Academy of Management.
7) Promann, M., & Brunswicker, S. (2017, August). Four affordances of feedback design. American Conference on Information
Systems (AMCIS). Boston, MA: Association of Information Systems.
8) Brunswicker, S., Bertino, E., & Matei, S. (2015). Big data for open digital innovation: A research roadmap. Big Data Research,
2(2), 53-58.
9) Koch, G., Füller, J., & Brunswicker, S. (2011). Online crowdsourcing in the public sector: How to design open government
platforms. Online Communities and Social Computing, 6778, 203-212. doi:10.1007/978-3-642-21796-8_22 (34 citations on
Google Scholar in September 2018)
Our experience in AI, system design, and computational social
science
11. Find out more
A teaser video about FACT: https://youtu.be/b2LQSe1QtGg
A website explaining FACT: www.rcodi.org/FACT/
Our centers
• RCODI: www.rcodi.org
• ISEEK: https://krannert.purdue.edu/centers/iseek/
Our google scholar profiles:
• Sabine Brunswicker https://goo.gl/fSR5Ns
• Alok Chaturvedi: https://goo.gl/zk18sj