agile and automation conclave 2018 - accenture · 2018-07-03 · agile and automation conclave 2018...
TRANSCRIPT
AGILE AND AUTOMATION CONCLAVE 2018
FUTURE OF ENTERPRISE AICHALLENGES AND OPPORTUNITIES
JANARDAN MISRA
Agile and Automation Conclave 2018
JANARDAN MISRATECHNOLOGY RESEARCH SR. PRINCIPAL, ACCENTURE LABS• 18+ years of R&D experience with contributions in areas of Unstructured
Data Analytics, Information Retrieval, Applied Machine Learning, and Complex
Adaptive Systems.
• 30+ peer reviewed research papers and a monograph.
• 40+ patents (issued and pending) across multiple geographies.
Agile and Automation Conclave 2018
AGENDA
• ENABLING TECHNOLOGIES
• WORKFORCE IMPLICATIONS
• POTENTIAL PITFALLS
• CURRENT STATE OF AI
• CHALLENGES WITH TODAY’S AI
• EMERGING TECHNIQUES
Agile and Automation Conclave 2018
CURRENT STATE OF AI
What can AI Do?
“If a typical person can do a mental task with < 1 second of thought, we can probably automate it using AI either now or in the near future”
-- Andrew Ng
• Often most helpful in complex environments
Probable Futures
Information
AI
Funnel
Agile and Automation Conclave 2018
AI – CURRENT STATE (CONT.)
• Major Breakthroughs
• Key Paradigm
• Deep Neural Networks based Supervised Learning
Speech and Image Recognition Language TranslationPersonalized Recommendations Credit Card Fraud DetectionSpam Filtering Search
Agile and Automation Conclave 2018
CHALLENGES WITH TODAY’S AI
Data Challenges• Effectiveness may come only with millions of data-points • Difficult to create ‘gold standard’ data set for training and validation
Engineering Challenges• Software Engineering for AI is still evolving! • Difficult to debug and incrementally improve in contrast to classical programming
Agile and Automation Conclave 2018
CHALLENGES WITH TODAY’S AI (CONT.)
Functional Challenges
• Causal Inferencing• Learning causation beyond correlations
Are these two definitions equivalent?• “A number that is divisible only by
itself and 1” • “a natural number greater than 1
that cannot be formed by multiplying two smaller natural numbers”
Temperature and Ice-cream sales are correlated!
• Do high temperatures cause high sales or vice versa?
• Reasoning• Commonsense and open-ended
inferences• Comprehension
• Learning abstractions through definitions
Agile and Automation Conclave 2018
EMERGING TECHNIQUES
COMPOSABLE AI SYSTEMS• Model vs Action
Composition
AI-SPECIFIC ARCHITECTURES• Domain Specific
ML models and hardware
NATURAL LANGUAGE PROCESSING• Conversational
and Q&A Agents
DEEP LEARNING + BIG DATA• Ability to learn
indefinitely as more data comes in
MISSION-CRITICAL AI• Acting in
Dynamic environments
Agile and Automation Conclave 2018
ENABLING TECHNOLOGIES
Never-ending Active Learning• Continuously learn as you predict with human-in-the-loop
Transfer Learning • Knowledge Reuse• Example: To be able to learn on open data to solve closed enterprise problems
Unsupervised Learning• Learning autonomously without explicit training
Agile and Automation Conclave 2018
WORKFORCE IMPLICATIONS
Data
Prediction
Judgement
Action
Output
Feedback
Employing Prediction Machines
• AI trainers vs designers
• Creative thinking vs routine execution
Experts-in-the-loop
• Complexity of judgements will be deciding
factor
• Skills to make right judgements will be critical
for the future workforces
Agile and Automation Conclave 2018
POTENTIAL PITFALLS
Technical Debt
• Potentially high maintenance costs after quick design wins
Lack of Explainability
• Most successful AI techniques are opaque
• “Right to Explanation” – GDPR
Security Concerns
• Data Poisoning attacks
• Lack of robustness against adversaries
Ethical Concerns
• How to ensure fairness?
Agile and Automation Conclave 2018
REFERENCESBooks• Human + Machine: Reimagining Work in the Age of AI. Paul Daugherty and H. James Wilson, Harvard
Business Review Press 2018
• The Master Algorithm: How the Quest for the Ultimate Learning Machine Will Remake Our World. Pedro
Domingos, Basic Books, 2018
• Humans Need Not Apply: A Guide to Wealth and Work in the Age of Artificial Intelligence. Jerry Kaplan,
Yale University Press, 2015
Articles• Artificial Intelligence and Life in 2030: One Hundred Year Study on Artificial Intelligence. Peter Stone et
al., Stanford University, 2016
• A Berkeley View of Systems Challenges for AI. Ion Stoica et al., arXiv.org, 2017
• What Artificial Intelligence Can and Can’t Do Right Now. Andrew Ng, HBR, 2016
• Future progress in Artificial Intelligence: A Survey of Expert Opinion. V. C. Müller and N. Bostrom, Springer
2016
• Deep Learning: A Critical Appraisal. Gary Marcus, arXiv.org, 2018
• What can Machine Learning Do? Workforce Implications. Eric B. and Tom Mitchell, Science, 2017
Agile and Automation Conclave 2018
Q&A
Agile and Automation Conclave 2018
FOLLOW USLinkedIn – SolutionsIQ India | Twitter – SIQIndia | Facebook – SolutionsIQ India
Thank you!
Janardan [email protected]