Augmented Political Intelligence
Rating Persuasion of Political/Public Communication
Persuasion Engine
Persuasion workflow:
OCEAN psychometric assessment to break audience into the 5 groups.
Deep customer segmentation.
addressing each group with customized persuasion.
Behavioral Microtargeting. Political Campaign Support.
Data-driven behavior change.
The Psychology of Persuasion
Robert Cialdini - Harnessing The Science Of Persuasion.pdf
http://content.yudu.com/Library/A17ln5/RobertCialdiniHarnes/resources/1.htm
Identify possibly six different patterns in persuasive TED talks.
http://changingminds.org/techniques/general/cialdini/cialdini.htm
if Social Proof - action words, positive sentiment
pre-processing by parsing and extracting action verbs extract how strongly audience can relate to the story see also https://www.amazon.com/Pre-Suasion-Revolutionary-Way-Influence-Persuade/dp/1501109790/ref=pd_sbs_14_t_0 Pre-Suasion: A Revolutionary Way to Influence and Persuade September 6, 2016 by Robert Cialdini Ph.D. Identify five parts of the persuasive speech calling for an action (charisma on command, Charlie Houpert) 5 Steps to Influence https://www.youtube.com/watch?v=H3aWbte8DxU&t=600s 1. Introduce the problem you have. What's getting done is not enough. Associate pain with the status quo. 2. Preemptively handling objections. Handling objections before people bring them up. 6-7 objections. 3/4 of the speech. If objections are not handled preemptively, people dig in into their positions expressing objections, 3. Convince of Consistency: The change you want to people to make is consistent with their beliefs. Make people to feel that it is a part of their identity. Give them implicit personal ownership of the change. Take pre-existing action and show how it's the beginning of the change he wants people to do. 4. Picture how things will be when the change is implemented. Give details (Obama skipped). 5. Ask for help. Because you can't make things yourself. Virtual Audience (NLP) strategy feedback methodology:
to use NLP feedback break down your strategy into specific purpose parts
Strategics steps politician takes to win a competition ( Trump, see Charlie Houpert bet on Trump youtube video)1. input one part of your five part strategy 2. let NLP to score the entered part against the training set 1. Get attention 2. Win one at a time 3. Remove threats one at a time 4. Define the terms of the engagement as though you are already a winner 5. Be flexible to attract only the people who follow you
Mistakes politician makes to help his opponent (Hillary, see Charlie Houpert bet on Trump youtube video)
1. Be clever rather than impactful
2. Let others define the terms of engagement
3. Don't change your strategy
4. Just allow Donald to dismantle your weapons
******************************************** body language, facial expressions, voice modulation https://www.youtube.com/watch?v=IqsAhmTn7n4 Paul Ekman's facial expressions research - over 50 points, 7 basic emotions, over 3000 expressions, notice micro expressions. Create Dialog: questions and suggestions in response to a speech. Task - optimizing persuasion. http://arxiv.org/abs/1606.01269 End-to-end LSTM-based dialog control optimized with supervised and reinforcement learning Jason D. Williams, Geoffrey Zweig (Submitted on 3 Jun 2016) This paper presents a model for end-to-end learning of task-oriented dialog systems. The main component of the model is a recurrent neural network (an LSTM), which maps from raw dialog history directly to a distribution over system actions. The LSTM automatically infers a representation of dialog history, which relieves the system developer of much of the manual feature engineering of dialog state. In addition, the developer can provide software that expresses business rules and provides access to programmatic APIs, enabling the LSTM to take actions in the real world on behalf of the user. The LSTM can be optimized using supervised learning (SL), where a domain expert provides example dialogs which the LSTM should imitate; or using reinforcement learning (RL), where the system improves by interacting directly with end users. Experiments show that SL and RL are complementary: SL alone can derive a reasonable initial policy from a small number of training dialogs; and starting RL optimization with a policy trained with SL substantially accelerates the learning rate of RL. |
Argument: Politics and marketing - any public speech - is a technology. We can estimate the effect of the speech and improve it.
ref: Politics and Religion are technologies - TED talk by Noah Feldman
1. Providing Feedback based on the testing of a given text for desired effects
- Addressing different types of audience
- Pro-Government vs Libertarian
see Snowden
- Young Open Minded Technical Savvy
- Evangelicals what makes political speech to appeal to a specific population
- Persuasion, Openness, Intellectual Brilliance, and Leadership, etc.
Methodology: Comparison of two Contextual Mappings - the wordings in great speeches and tested texts.
-Using factorial analysis in Hyperbase. Graphical representation of the speech - spatial distribution - clouds - of words with smaller distances corresponding to higher frequencies of co-occurrences of the words.
-marketing - golden circle-Apple example in Simon Sinek 'How great leaders inspire action' TED talk
https://www.ted.com/talks/simon_sinek_how_great_leaders_inspire_action?language=en
recher vocabulary -> higher engagement
2. Stretch Goal - Recommending speech rewording by translating a given fragment to the one from a recognized high ranking speaker.
Technical details:
RNN with LSTM - translation, see
Ilya Sutskever's home page
www.cs.toronto.edu/~ilya
Publications - RNN text generation demo - Award
1. Collecting a corpora - training set of the phrases/paragraphs rated by experts in a binary fashion as
- persuasive/non-persuasive (convincing/unconvincing) (similar to positive/negative sentiment)
- rhetorical quality (compare to great rhetoric in famous speeches)
Analysis -
sub-properties of "convincing":
self promotion - verbal, explicit, implicit, tone, body language - posture, pose, gestures, facial expression
audience can relate to the key concepts
establishing trust
perceived openness
perceived leadership
perceived credibility
key position statements
key promises that are important to an audience
etc.
experts can rate the phrases or entire documents by subproperties. the resulting score can be used to validate a general conclusion on "convincing" property.
2. Building a model for scoring a phrase convincibility (persuasion).
3. Using ensemble model that includes RNN model.
4. Testing prepared speeches, answers to anticipated questions, political platform descriptions against strong persuasion model
5. Improving a corpora by updating it based on the real life results.
Proving the effectiveness of the RNN scoring:
scoring social media - Twitter - for positive and negative sentiment in responses to high score- and low-score speeches and debates
RESPONSE SIMULATION FOR ACHIEVING MAXIMUM RESONANCE
Creating a sandbox for adjusting the input message to receive desirable response
1. Collecting large amounts of feedback messages in response to the various political input speeches.
2. Visualizing the clusters of responses to each input speech
3. Simulating the response to the input speech based on the prior experiences.
This work lies in the intersection between the areas of computational creativity and information retrieval. In our approach, we assume that users have a certain concept in their mind, formulated as a sequence of lines, and their information need is to find the missing lines, composing a convincing sequence. Such an information need does not have a factual answer; nevertheless, users will be able to assess the relevance of the response provided by the system. The relevance of the response depends on factors that include expressive force, vocabulary, unexpectedness, semantic coherence, and humor. We use an open-source speech synthesizer, eSpeak, to produce the transcription. eSpeak has also text to-phonemes functionality.
eSpeak text to speech
http://espeak.sourceforge.net/
VISUALIZATION AUGMENTING HUMAN INTELLIGENCE
LAYERING ANALYTICAL VISUALIZATIONS
EXAMPLE:
Quid’s article-analyzing app can tell you many things — like why you lost the Senate race in Iowa
http://venturebeat.com/2014/11/27/quids-article-analyzing-software-is-amazing-it-can-even-tell-you-what-went-wrong-in-your-political-campaign/
Main SW components: web portal, secure gateway, API manager, backend app deployer, DB, operational analytics.
Business Marketing Model: Native ads that look like a content (see LinkedIn as an example)
Everybody wants to maximize his interest. Find the person and a group whose interest maximization will maximize your interest. This is a multi-entity one-shot game.
Application of Image Recognition
power and confidence posture recognition for politicians
Flow Chart of an Adviser - Augmented Intelligence - can be created with
https://www.draw.io/
notes:
check if anything new in http://arxiv.org/pdf/1601.06733v3.pdf
Long Short-Term Memory-Networks for Machine Reading
Jianpeng Cheng, Li Dong, Mirella Lapata
(Submitted on 25 Jan 2016 (v1), last revised 1 Feb 2016 (this version, v3))
generative model example:
DopeLearning: A Computational Approach to Rap Lyrics Generation
18 May 2015
http://arxiv.org/pdf/1505.04771.pdf
IBM's Watson can sense sadness in your writing
The AI just got a big emotion-detecting upgrade.
http://www.engadget.com/2016/02/22/ibm-emotion-detection-upgrade/
Scott Adams' analysis of the Trump's strategy with the voters: similar to hypnosis - lead and pace.
Lead: express similar feelings as your audience, win their sympathy and trust
Pace: diverge in the direction you want to go and see your audience following you because you already won their sympathy.
Enhancing charisma, breakdowns from Charlie Houpert Trump Prediction. Facts beat feelings:
1. Don't memorise your speech word by word. Only 3-9 logical parts. And exact opening and closing of your speech.
2. Speak with ease and conviction
3. Sound like authority being specific with numbers and facts. Expertise enhances charisma. Sound credible.
4. Feel comfortable answering difficult questions
5. Conviction (rather than softer tone) and congruency (genuine smile and laughter rather than a forced one)
6. Create stronger emotional connection with the audience:
6.1 Images and stories you invoke connect you to others
6.2 Speaking in the third person "they" and calling people by their group names (African Americans, Latinos) distances you from those people
6.3 Speak in second person, meaning you - we need your talents, your skills...
6.4 Look into the camera dead on makes an impression that you speak directly with the audience
AI vs Human Intelligence -
The Rise of AI Makes Emotional Intelligence More Important
sub-properties of "convincing":
self promotion - verbal, explicit, implicit, tone, body language - posture, pose, gestures, facial expression
audience can relate to the key concepts
establishing trust
perceived openness
perceived leadership
perceived credibility
key position statements
key promises that are important to an audience
etc.
experts can rate the phrases or entire documents by subproperties. the resulting score can be used to validate a general conclusion on "convincing" property.
2. Building a model for scoring a phrase convincibility (persuasion).
3. Using ensemble model that includes RNN model.
4. Testing prepared speeches, answers to anticipated questions, political platform descriptions against strong persuasion model
5. Improving a corpora by updating it based on the real life results.
Proving the effectiveness of the RNN scoring:
scoring social media - Twitter - for positive and negative sentiment in responses to high score- and low-score speeches and debates
RESPONSE SIMULATION FOR ACHIEVING MAXIMUM RESONANCE
Creating a sandbox for adjusting the input message to receive desirable response
1. Collecting large amounts of feedback messages in response to the various political input speeches.
2. Visualizing the clusters of responses to each input speech
3. Simulating the response to the input speech based on the prior experiences.
This work lies in the intersection between the areas of computational creativity and information retrieval. In our approach, we assume that users have a certain concept in their mind, formulated as a sequence of lines, and their information need is to find the missing lines, composing a convincing sequence. Such an information need does not have a factual answer; nevertheless, users will be able to assess the relevance of the response provided by the system. The relevance of the response depends on factors that include expressive force, vocabulary, unexpectedness, semantic coherence, and humor. We use an open-source speech synthesizer, eSpeak, to produce the transcription. eSpeak has also text to-phonemes functionality.
eSpeak text to speech
http://espeak.sourceforge.net/
VISUALIZATION AUGMENTING HUMAN INTELLIGENCE
LAYERING ANALYTICAL VISUALIZATIONS
EXAMPLE:
Quid’s article-analyzing app can tell you many things — like why you lost the Senate race in Iowa
http://venturebeat.com/2014/11/27/quids-article-analyzing-software-is-amazing-it-can-even-tell-you-what-went-wrong-in-your-political-campaign/
Finding patterns of lie in a competitor's speech
elevated jargon use in fundamentally deceptive papers
http://news.stanford.edu/news/2015/november/fraud-science-papers-111615.html
*********************************************elevated jargon use in fundamentally deceptive papers
http://news.stanford.edu/news/2015/november/fraud-science-papers-111615.html
Main SW components: web portal, secure gateway, API manager, backend app deployer, DB, operational analytics.
Business Marketing Model: Native ads that look like a content (see LinkedIn as an example)
Visual comparison graph augments cognitive perception of similarity and differentiation of texts.
Feature Set for Prediction Of Political Success
1. IQ
http://www.businessinsider.com/the-15-smartest-us-presidents-of-all-time-2015-3
2. Financial support
3. Economic interests of the majority of voters
4. Strength of legal support
5. Charisma
6. Debate skills
7. Leadership skills
8. Relatable image
9. Ability to convince audience. Tools: use concepts to which audience relates.
10. Discerning untruthfulness, intentional fact distortion in a competitor speeches, statements.
1. IQ
http://www.businessinsider.com/the-15-smartest-us-presidents-of-all-time-2015-3
2. Financial support
3. Economic interests of the majority of voters
4. Strength of legal support
5. Charisma
6. Debate skills
7. Leadership skills
8. Relatable image
9. Ability to convince audience. Tools: use concepts to which audience relates.
10. Discerning untruthfulness, intentional fact distortion in a competitor speeches, statements.
Everybody wants to maximize his interest. Find the person and a group whose interest maximization will maximize your interest. This is a multi-entity one-shot game.
Application of Image Recognition
power and confidence posture recognition for politicians
Flow Chart of an Adviser - Augmented Intelligence - can be created with
https://www.draw.io/
notes:
check if anything new in http://arxiv.org/pdf/1601.06733v3.pdf
Jianpeng Cheng, Li Dong, Mirella Lapata
(Submitted on 25 Jan 2016 (v1), last revised 1 Feb 2016 (this version, v3))
generative model example:
DopeLearning: A Computational Approach to Rap Lyrics Generation
18 May 2015
http://arxiv.org/pdf/1505.04771.pdf
IBM's Watson can sense sadness in your writing
The AI just got a big emotion-detecting upgrade.
http://www.engadget.com/2016/02/22/ibm-emotion-detection-upgrade/
Scott Adams' analysis of the Trump's strategy with the voters: similar to hypnosis - lead and pace.
Lead: express similar feelings as your audience, win their sympathy and trust
Pace: diverge in the direction you want to go and see your audience following you because you already won their sympathy.
Enhancing charisma, breakdowns from Charlie Houpert Trump Prediction. Facts beat feelings:
1. Don't memorise your speech word by word. Only 3-9 logical parts. And exact opening and closing of your speech.
2. Speak with ease and conviction
3. Sound like authority being specific with numbers and facts. Expertise enhances charisma. Sound credible.
4. Feel comfortable answering difficult questions
5. Conviction (rather than softer tone) and congruency (genuine smile and laughter rather than a forced one)
6. Create stronger emotional connection with the audience:
6.1 Images and stories you invoke connect you to others
6.2 Speaking in the third person "they" and calling people by their group names (African Americans, Latinos) distances you from those people
6.3 Speak in second person, meaning you - we need your talents, your skills...
6.4 Look into the camera dead on makes an impression that you speak directly with the audience
AI vs Human Intelligence -
The Rise of AI Makes Emotional Intelligence More Important
by Megan Beck,
Barry Libert
FEBRUARY 15, 2017
The booming growth of machine learning and artificial intelligence (AI), like most transformational technologies, is both exciting and scary. It’s exciting to consider all the ways our lives may improve, from managing our calendars to making medical diagnoses, but it’s scary to consider the social and personal implications — and particularly the implications for our careers. As machine learning continues to grow, we all need to develop new skills in order to differentiate ourselves. But which ones?
It’s these human capabilities that will become more and more prized over the next decade. Skills like persuasion, social understanding, and empathy are going to become differentiators as artificial intelligence and machine learning take over our other tasks. Unfortunately, these human-oriented skills have generally been viewed as second priority in terms of training and education. We’ve all experienced the doctor, financial planner, or consultant who is more focused on his or her reports and data than on our unique situations and desires.
YOW! 2016 Eva Nahari - NLP: Something Old, Something New.
https://www.youtube.com/watch?v=InayOc_Dpyg
YOW! Conferences
Published on Mar 7, 2017
No comments:
Post a Comment