Sarit Kraus Bar-Ilan University

Multi-issue Negotiation Employer and job candidate. Objective: reach an agreement over hiring terms after successful interview. Chat based negotiations 3

Training People in Negotiations 4

5/19 The Training Methods Classical role playing with another human counterpart:  simple  scheduling, requires other people Agent role playing  accessible and available 24/7  modeling different counterparts no significant differences were found between the different training methods ( 148 human subjects)

Virtual Suspect to Train Investigators

7

Culture Sensitive Agents for Culture related Studies The development of standardized agents to be used in the collection of data for studies on culture and negotiation Bargaining

Automated Mediators for Resolving Conflicts

Automated Speech Therapist

11 First place of the TedMed 2014 innovation competition for startups in medicine

Language Tele-Rehabilitation: Monitoring and Intervention 12 Patient: Currently in use by patients. Cyndi:גם זה פרי Patient:תפוח טפוך האותיות השניה והשלישית נכונות

Past deliberations accumulative data Agent Current deliberation Update Offer arguments = Obtains information Agent Supports Deliberation 13

15 Supporting Robots-Human Teams

Sustainability: Reducing Fuel Consumption Persuasion

Why not use only game theory? Game theory is a study of strategic decision making: "the study of mathematical models of conflict and cooperation between intelligent rational decision- makers“ Results from the social sciences suggest people do not follow game theory strategies.

People Often Follow Suboptimal Decision Strategies Irrationalities attributed to sensitivity to context lack of knowledge of own preferences the effects of complexity the interplay between emotion and cognition the problem of self control

Why not Only Behavioral Science Models? There are several models that describe human decision making Most models specify general criteria that are context sensitive but usually do not provide specific parameters or mathematical definitions

Why not Only Machine Learning? Machine learning builds models based on data It is difficult to collect human data Collecting data on specific user is very time consuming. Human data is noisy “Curse” of dimensionality

Methodology Human Prediction Model Take action machine learning Game Theory Optimization methods Human behavior models Data (from specific culture) Human specific data

22

An Experimental Test-Bed Interesting for people to play: analogous to task settings; vivid representation of strategy space (not just a list of outcomes). Possible for computers to play. Can vary in complexity repeated vs. one-shot setting; availability of information; communication protocol; Negotiations and teamwork 25

CT Game 100 point bonus for getting to goal 10 point bonus for each chip left at end of game Agreement are not enforceable 26 Collaborators: Gal, Haim, Gelfand

An Influence Diagram- Two rounds interaction Probability of acceptance Probability of transfer

Methodology Human Prediction Model Take action machine learning Game Theory Optimization methods Human behavior models Data (from specific culture) Human specific data

Average Task dep. Task indep. Co-dep People (Lebanon) People (US) Prediction of Acceptance and Reliability Data Collection: Human playing against adaptive agent (PURB) 29 Reliability Measure

Personality, Adaptive Learning (PAL) Agent 30 Human Prediction Model Take action Machine Learning Optimization methods Optimization methods Data from specific culture In this data set the Lebanon people almost always kept the agreements  PAL never kept agreements “Nasty Agent”: Less reliable when fulfilling its agreement People adapt their behavior to their counterparts.

Pal vs Humans 31 The model’s used by PAL had a very low accuracy prediction of people actually played against PAL

Multi-issue Negotiation: The Negotiation Scenario Employer and job candidate Objective: reach an agreement over hiring terms after successful interview 32

33 Chat-Based Negotiation

34 NegoChat’s Offers Find a “good” offer to use as anchoring Predict which offers will be accepted using on past negotiation sessions and create clusters of possible offers per time period Aspiration Adaptation Theory (AAT) proposes issues sequentially based on aspiration scale learned from data retreat from previous values 34 Collaborators: Avi Rosenfeld, Zukerman, Dagan, Gelfand

NegoChat vs KBagent in Israel 35

NegoChat in Egypt Collect new data of human vs human (negotiations much longer) Build classifier Extract aspiration list Got many complaints on the agent. Challenges in running experiments. Results: slower than in Israel; lower score; almost the same happiness and fairness; females scored much lower. 36

One operator – Multiple robots Search And Rescue (SAR) Warehouse operation Automatic air-craft towing Fire-Fighting Military applications Etc..

Semi-Autonomous Robots Controls the robots Noisy signals

Agent Prioritize tasks Provide warnings Controls the robots Noisy signals Filtered signals

Agent design Provide Advice Machine Learning Optimization Data on robots performance Data on human behavior Robot model Human model 150 hours of simulations (no human operator). 30 human Operators in simulation

Evaluation: Three Environments 16 subjects 12 subjects

Objects found per condition Simulated officePhysical office Simulated warehouse yard

Relatively low impact Possible Reason: lack involvement of the teachers Better results with a human teachers in the loop. Computer Games for Learning

Capabilities for Agents in Learning Planning Monitoring Intervention Encouragement 45

Human-agent Working Together Planning: human forms the basic learning plan; agent adapts it to the student’s progress. Monitoring & Intervention: Agent performs M&I and asks the human for help when it is not sure. Encouragement: point up to the human that a student needs encouragements. 46

Team of Agents and Teacher Many automated agents provide service to many students. One human teacher supervises many students. Special agent supports the teacher. 47

Reducing Energy Consumption by Advising People on how to set the Climate Control System GM Chevrolet Volt 2011

Presenting Advice to User

~80% of drivers explicitly accepted.

Methodology Collecting data on 38 subjects for an hour session in the car) Predicting drivers reactions to offers Modeling the car and environment Integrating into an MDP Solving the MDP 51

Evaluation 45 drivers - 15 per condition, 3 rounds. The lower the better.

Why Did MDP Outperform the SAP? SAP was aggressive. Some subjects stopped clicking on the advice. AgentAvg. go eco %Avg. save %Avg. consumption MACS SAP agent

This does not look as a real person!! Remove the Avatar!!

AniMed evaluation

56 Agents interacting proficiently with people is important Human Prediction Model Take action machine learning Game Theory Optimization methods Human behavior models Data (from specific culture) Human specific data Challenges: Experimenting with people is very difficult !!! Working with people from other disciplines is challenging. Challenges: How to integrate machine learning and behavioral models? How to use in agent’s strategy?