31
Design & Intelligence Lab - Georgia Institute of Technology Using Watson to Enhance Human-Computer Co-Creativity Presented by: Bryan Wiltgen Based on work by: Ashok Goel, Brian Creeden, Mithun Kumble, Shanu Salunke, Abhinaya Shetty, & Bryan Wiltgen

Using Watson for Enhancing Human-Computer Co-Creativity

Embed Size (px)

Citation preview

Design & Intelligence Lab - Georgia Institute of Technology

Using Watson to Enhance Human-Computer Co-Creativity

Presented by: Bryan Wiltgen

Based on work by: Ashok Goel, Brian Creeden, Mithun Kumble,

Shanu Salunke, Abhinaya Shetty, & Bryan Wiltgen

Design & Intelligence Lab - Georgia Institute of Technology

Acknowledgments

● IBM

○ Watson Engagement Advisor

○ Two IBM Faculty Awards in Cognitive Systems.

● Students in the Georgia Tech’s Spring 2015

CS 4803/8803: Computational Creativity class

● Dianne Fodell for inviting me to speak!

Design & Intelligence Lab - Georgia Institute of Technology

IBM’s Watson Engagement Advisor

CS 4803/8803: Computational Creativity

Human-Computer Co-Creativity

Watson Best Practices

Biologically Inspired Design Apps

Watson in Education

Design & Intelligence Lab - Georgia Institute of Technology

Outline

1. Background

2. Class Projects

3. Conclusions

Design & Intelligence Lab - Georgia Institute of Technology

Background

Design & Intelligence Lab - Georgia Institute of Technology

The Design & Intelligence Lab (http://dilab.gatech.edu)

FacultyGoel Joyner McGreggor Rugaber

Ph.D. StudentsBanerjee Delgado Ehsan Fitzgerald Wiltgen

AwasthyAzad

FrazerKulkarni

KoushikSarathy

ShettyHartman Tuchez

Spiliopoulou

M.S. Students UG Students

Design & Intelligence Lab - Georgia Institute of Technology

ComputationalCreativity

Design Thinking

SystemsThinking

AnalogicalThinking

VisualThinking

MetaThinking

Abductive Thinking

The Design & Intelligence Lab (http://dilab.gatech.edu)

Design & Intelligence Lab - Georgia Institute of Technology

CS 4803/8803: Computational Creativity ClassSpring 2015

24 Students - 21 graduate, 3 undergraduate

Learning Goals:

1. To become familiar with the literature on computational creativity (concepts,

methods, tasks)

2. To become familiar with the state of art in computational creativity (systems,

techniques, tools)

3. To learn about the processes of designing, developing and deploying

interactive/autonomous creative systems from ideation to realization

4. To acquire experience in designing an interactive creative tools

5. To become an independent thinker in computational creativity.

Design & Intelligence Lab - Georgia Institute of Technology

Class Projects in Computational Creativity Class

Six, self-organized teams of four students each

Goal:

Deeply address a problem in computational creativity,

specifically in the domain of biologically inspired design.

Design & Intelligence Lab - Georgia Institute of Technology

Class Projects in Computational Creativity Class

Phase 1: Initial Learning Phase

1. Select case-study of Biologically Inspired Design

2. Seed Watson with articles

3. Generate and answer questions relative to selected case

4. Train Watson on Q&A pairs

5. Evaluate Watson for answering design questions relevant to selected case

Phase 2: Open-ended Research Phase

● Teams grew the number of documents in our Watson KB

● Teams developed custom-made software

● Teams evaluated aspects of their projects

● Teams wrote reflective design reports and made short videos

Design & Intelligence Lab - Georgia Institute of Technology

Biologically Inspired Design

Core Idea:

Take inspiration from nature to develop creative

solutions to design challenges.

Design & Intelligence Lab - Georgia Institute of Technology

Example of Biologically Inspired Design

Shinkansen 500: Japanese bullet train

Design & Intelligence Lab - Georgia Institute of Technology

Example of Biologically Inspired Design

Shinkansen 500: Japanese bullet train

Design & Intelligence Lab - Georgia Institute of Technology

IBM’s Watson(Specifically: Watson Engagement Advisor)

Automated question-answering.

Q&A Training

Styling

Our Students Inputted the Following:● About 500 Documents● About 1200 Questions

Design & Intelligence Lab - Georgia Institute of Technology

Class Projects

Design & Intelligence Lab - Georgia Institute of Technology

Ask Jill

Search for and manage relevant papers

in a literature survey writing context.

The Class ProjectsSustArch

Search tool and marketplace for

sustainable architecture.

Design & Intelligence Lab - Georgia Institute of Technology

The Class ProjectsWatson BioMaterial

Search for relevant materials.

Watsabi

Automated and community-driven

Q&A for agriculture.

Design & Intelligence Lab - Georgia Institute of Technology

The Class ProjectsTwenty Questions

Find relevant document through an iterative, game-inspired process.

Design & Intelligence Lab - Georgia Institute of Technology

The Class ProjectsTwenty Questions

Find relevant document through an iterative, game-inspired process.

Design & Intelligence Lab - Georgia Institute of Technology

The Class ProjectsTwenty Questions

Find relevant document through an iterative, game-inspired process.

Design & Intelligence Lab - Georgia Institute of Technology

The Class ProjectsTwenty Questions

Find relevant document through an iterative, game-inspired process.

Design & Intelligence Lab - Georgia Institute of Technology

The Class ProjectsErasmus

Explore a concept-space.

Design & Intelligence Lab - Georgia Institute of Technology

The Class ProjectsErasmus

Explore a concept-space.

Design & Intelligence Lab - Georgia Institute of Technology

The Class ProjectsErasmus

Explore a concept-space.

Design & Intelligence Lab - Georgia Institute of Technology

The Class ProjectsErasmus

Explore a concept-space.

Design & Intelligence Lab - Georgia Institute of Technology

Conclusions

Design & Intelligence Lab - Georgia Institute of Technology

IBM’s Watson Engagement Advisor

CS 4803/8803: Computational Creativity

Human-Computer Co-Creativity

Watson Best Practices

Biologically Inspired Design Apps

Watson in Education

Design & Intelligence Lab - Georgia Institute of Technology

Watson Best Practices● Well structured/annotated data is needed and is time-

consuming to produce.

○ Automation

■ Including automated chunking of documents

○ Crowd-Sourcing

○ Wikipedia is a good source.

● Training Watson

○ Bridge questions to link questions together.

○ Crowd-Sourcing

○ Developer-Responsible Feedback Loop

○ 3-Step Approach to Training

■ Train on all possible questions that the corpus papers can answer

■ Train Watson on alternative versions of those questions

■ If first two steps prove insufficient, add additional papers and train on

those.

Design & Intelligence Lab - Georgia Institute of Technology

Human-Computer Co-Creativity

When creativity emerges from interactions

between humans and computers.

Hypothesis: Human-Watson interaction, when mediated through

technologies like those produced by the students, results in

human-computer co-creativity.

Why?

● Users typically will engage in longer-term interactions.

● Techs added semantics and context to Watson interactions.

Design & Intelligence Lab - Georgia Institute of Technology

Watson in Education

Variety in the Projects

● Topics

○ 5 Bio-inspired Design

■ 3 Domain Independent

■ 2 Targeted specific domains (Resilient

Materials and Built Architecture)

○ 1 Agriculture

● Interaction

○ All supported human-computer interaction.

○ 2 supported human-human interaction.

● All were Internet-enabled.

● 2 were Android apps.

● 1 inspired by a game.

● 1 integrated another IBM technology

(AlchemyAPI).

Exploring Watson for Online

Classroom FAQ

● Goel teaches the Udacity class CS 7637:

Knowledge-based AI: Cognitive Systems as

part of Georgia Tech’s OMSCS program.

● The class uses Piazza for online discussion.

● In Fall 2014:

○ ~6950 Messages

○ ~170 Students

● In Spring 2015:

○ ~11,000 Messages

○ ~240 Students

● We are developing a Watson-powered

technology to automatically answer FAQ’s in

this future offerings of this class.

Design & Intelligence Lab - Georgia Institute of Technology

Thank You For Listening

Bryan Wiltgen

[email protected]

Design & Intelligence Lab

http://dilab.gatech.edu

Class Project Video Playlist on YouTube

https://www.youtube.com/playlist?list=PL44rHkM-p0hu5H7oS3OXYgK9qDkVajyqY

Published Version of This Work

Ashok Goel, Brian Creeden, Mithun Kumble, Shanu Salunke, Abhinaya Shetty, & Bryan Wiltgen (2015). Using

Watson for Enhancing Human-Computer Co-Creativity. To appear in procs. of AAAI 2015 Fall Symposium.