Systems Engineering: Optimizing Creation of Virtual Conversational - PowerPoint PPT Presentation

Systems Engineering: Optimizing Creation of Virtual Conversational Human Agents Daniel P. Burns Dan M. Davis & CAPT, USN-Ret Julianne M. Nordhagen Home Ports Solutions, LLC Univ. of Southern California Los Angeles, California Los Angeles, California Norfolk, Virginia  April 24-26, 2018

Major Theses • Personnel & funding constraints are limiting mentoring • Computer-generated conversational agents address this • Complex implementation issues are problematic • System Engineering (SE) tools are available and useful • These can often be applied by researchers and users • Optimizing Virtual Humans with SE has shown success • A quick introduction to those tools will be helpful Norfolk, Virginia  April 24-26, 2018 Norfolk, Virginia  April 24-26, 2018 2

Existing Problem • Declining budgets and demanding Ops Schedules • Burgeoning need for more mentoring and technicians • Virtual Conversational Humans can fill the need • Virtual mentors available 24/7 anywhere net reaches • Research and implementation complex & expensive • SE tools not in widespread use in academia and DoD • Constraints not likely to improve in foreseeable future Norfolk, Virginia  April 24-26, 2018 Norfolk, Virginia  April 24-26, 2018 3

Research in Virtual Humans (VH) • Three USC VH programs • New Dimensions in Testimony – Holocaust memories preserved • MentorPal – Advising STEM students on career paths • SimCoach – Counseling PTSD Veterans via internet • All complex and difficult to implement efficiently Norfolk, Virginia  April 24-26, 2018 Norfolk, Virginia  April 24-26, 2018 4

SE Optimization • Staffing, processing, and producing deliverable items • Especially useful to help organize several systems • SE Goals – Design, develop, implement and operate systems – Forecast behaviors in various environments – Deliver intended functions in sustainable fashion • Can be utilized by journeyman researcher/user • Helpful even at elementary level of implementation Norfolk, Virginia  April 24-26, 2018 Norfolk, Virginia  April 24-26, 2018 5

Commonly Used SE Tools • V-Model – reminds user that for every stage of development, there is a corresponding evaluation phase • Stakeholder Diagram – a formalization of recognizing all of the entities who have and interest in the initiative Norfolk, Virginia  April 24-26, 2018 Norfolk, Virginia  April 24-26, 2018 6

SE Tools Continued • Context Diagram – examines and documents environment in which the system will operate • Interfaces, boundaries, … • Pugh Matrix – plots design satisfaction vs. implementation options • Numerical analysis of various paths to goal Norfolk, Virginia  April 24-26, 2018 Norfolk, Virginia  April 24-26, 2018 7

More SE Tools • Analytic Hierarchy Process – Quantifies the decision processes • Quantitative analysis by assigning numerical values to components • Design Structure Matrix – ID’s subsystems and interfaces • X indicates connectivity and origin of data flow Norfolk, Virginia  April 24-26, 2018 Norfolk, Virginia  April 24-26, 2018 8

Two SE Tools • Functional Modeling – identifies and documents functionality and critical interconnection • Breaks model down into small sub-systems for better comprehension and analysis • Implementation Means Analysis - a second way to look at alternative paths to goal • Not quantified, but shows choices Norfolk, Virginia  April 24-26, 2018 Norfolk, Virginia  April 24-26, 2018 9

Last SE Tools • Failure Mode & Effects Analysis – specifies and logs potential causes and effects of failure • System Requirements Model – puts requirements into a framework: Operational and Functional • N 2 Analysis – examines impact of growth of system • Requirement Relationship Analysis – relates sponsor needs and leadership vision to requirements Norfolk, Virginia  April 24-26, 2018 Norfolk, Virginia  April 24-26, 2018 10

MentorPal Project • DoD needs technically trained personnel • Several factors inhibit students’ pursuing tech careers – Lack of effective mentors – Misunderstanding instilled by media – Poor understanding of career characteristics • Low SES and geographically remote get little help • Virtual Human capable of conversation can augment • ICT has developed a series of successful VH systems • VH can respond to questions from students Norfolk, Virginia  April 24-26, 2018 Norfolk, Virginia  April 24-26, 2018 11

SE Tools Applicable to MentorPal • V-Model • Context Diagram • Stakeholder Diagram • Articulating Requirements • System Requirement Model • Implementation Mean Analyses • Failure Mode and Effects Analysis Norfolk, Virginia  April 24-26, 2018 Norfolk, Virginia  April 24-26, 2018 12

Record/Estimates of Time Norfolk, Virginia  April 24-26, 2018 Norfolk, Virginia  April 24-26, 2018 13

Impact to Date • Drove formalization and documentation of issues • Function analysis found areas in the heterogeneous code base that were causing crashes • Implementation Means study identified huge time investment in transcription, which was cut back • Stakeholders’ Diagram helped suggest follow -on work which will drive new proposals • Context diagram illuminated issues for implementation Norfolk, Virginia  April 24-26, 2018 Norfolk, Virginia  April 24-26, 2018 14

Future Research • Would be useful to seek funding to – Assess needs for this approach – Quantify benefit • Social, professional and personal resistance to change when SE is introduced into processes – Need to proselytize – Careful selection of SE’s to work well with researchers • Seek funding during implementation phase of MentorPal to prove efficacy of SE Norfolk, Virginia  April 24-26, 2018 15

Getting Student Evaluations Norfolk, Virginia  April 24-26, 2018 Norfolk, Virginia  April 24-26, 2018 16

Conclusions • SE’s are useful and beneficial • Few researcher take the time to apply SE • Can give good return on investment • Save costs • Improve morale • Help meet schedule • Taking a few moments to formalize and record issues is a good investment Norfolk, Virginia  April 24-26, 2018 17

Acknowledgements & Caveats Much of the work described above was conducted in response to a Office of Naval Research contract named MentorPal: Growing STEM Pipelines with Personalized Dialogs with Virtual STEM Professionals, N00014-16-R-FO03. The opinions expressed herein are the authors’ own and do not necessarily reflect those of the Department of the Navy, the Department of the Army or the U.S. Government. Norfolk, Virginia  April 24-26, 2018 18