Computers that can Negotiate ERCIM Cor Baayen Award Tim Baarslag - PowerPoint PPT Presentation

Computers that can Negotiate ERCIM Cor Baayen Award Tim Baarslag Researcher in Centrum Wiskunde & Informatica (CWI), Research institute for Mathematics and Computer Science in the Netherlands

Negotiation  Negotiation is everywhere around us.  Many human deficiencies: – Leaving money on the table, bounded rationality – Biases & emotions, time & costs

Research line I NEGOTIATION SUPPORT 3

Negotiation support requirements 2. Generate bids 1. Learn about the user and opponent 3. Advise you when to accept.

Automated negotiation challenges Good for opponent Contract 𝑪 Fte : 1.0 Car : no Salary : $3500 𝑪 Contract 𝑩 Fte : 0.8 Car : yes Salary : $3000 dominanto utcomes 𝑩 5 Good for me

Automated negotiation challenges Good for opponent learning 6 Good for me accepting bidding

The Automated Negotiating Agent Competition Baarslag et al., Evaluating Practical Negotiating Agents: Results and Analysis of the 2011 International Competition , Artificial Intelligence, 2013.

Research line II PRIVACY NEGOTIATIONS 8

Privacy in the digital economy  Our data is the currency of many digital services  Problems – Take it or leave it approach – One size fits all – Opaque business models  What if we could negotiate our privacy decisions?  Agent representation with incomplete preferences

State of the art: State-of-the-art no uncertainty no uncertainty Opponent utility dominanto utcomes 10 User utility

State of the art: State-of-the-art opponent uncertainty no uncertainty opponent uncertainty Opponent utility 11 User utility

Key future challenge: State-of-the-art full uncertainty no uncertainty opponent uncertainty Opponent utility Unexplored full uncertainty  New concepts required: – Elicitation on-the-fly: Reduce uncertainty which queries to ask? with costly queries – What is (costly) user information worth? 12 User utility

First results: personalized privacy negotiations  Tested with mobile app and real, personal, publically published data  Results show that negotiation gives users control, and more meaningful consent Baarslag et al., Negotiation As an Interaction Mechanism for Deciding App Permissions , CHI Late Breaking Work, 2016.

Current applications  Internet of Things privacy management  Social media preferences  Smart energy cooperatives

Further pointers Lewis et al. Deal or No Deal? End-to-End Learning for Negotiation Dialogues. Facebook AI, 2017. Baarslag et al. How would a machine conduct our salary negotiations? Wired, 2017. Baarslag et al. How artificial intelligence could negotiate better deals for humans. Science, 2017.