AI summary by TwinMind
Juan Corvalan | Incubating Predictive Systems: Application Challenges | Talks at Google
Talks at Google
1h 2m 22s
Introduction and Background
- Virtual Google Detox edition hosted by Belén Carrera from Google Cloud technical residents program
- Guest: Juan Corvalán, director/co-founder of ialam AI laboratory at University of Buenos Aires, Argentina
- Juan's credentials: lawyer, MIT machine learning specialist, Oxford AI specialist
- Co-creator of predictive systems Promethea and Pretor
- Co-created Uva and Alabe metaverse (first proto metaverse by Latin American public university)
- Author of 8 books, 4 on Artificial Intelligence
- Academic consultant at Latin American Development Bank
Historical Context and Technology Convergence
- 1906 San Francisco video example showing technological accumulation
- Multiple centuries of innovation (4-6 centuries) overlapping in single street
- Trams, carriages, bicycles, early automobiles coexisting
- State regulation trying to manage technological convergence
- Parallel to current AI disruption and technological convergence
- Need for technologies to "decant" and find coexistence space
Commodity Predictions Concept
- AI becoming epicenter of 4th Industrial Revolution
- Feeds back into other emerging technologies: biotechnology, nanotechnology, 3D printing, immersive computing, quantum computing
- Commodity predictions = reusable, scalable predictions
- Google Translate as prime example of commodity prediction
- Spanish language processing advanced significantly 2010-2020
- Benefits: increased task efficiency, reduced development costs, foundation for other predictions
Latin American AI Challenges
- 2019 Haiti summit on strategic AI plans for Latin American countries
- Regional challenges vs. advanced countries (US, Estonia, Europe, Japan)
- Less evolved data ecosystem and interoperability
- Need for shared strategies to anchor in certain commodities
- Digital transformation by design = Artificial Intelligence
- Otherwise transforming in obsolete technological/conceptual terms
Alliance Between Sectors
- Difficult alliance between companies, academia, civil society, state
- Innovating in Latin America = facing historical gaps not yet closed
- Asymmetric situation: discussing AI agenda while citizens demand closing other gaps
- >80% population lacks space to download applications
- Issues: data consumption resources, IT culture, literacy, awareness
- Fear of replacement by machines and intelligent systems
Technological Adaptation Principles
- Empathy and adaptation to local culture critical
- Principle of technological adaptation
- People should embrace technology in their own way
- Cobotization, augmented intelligence, complementarity paradigm
- Working with machines/predictive machines
- Two keys learned: commodity predictions + adaptation to context/culture
- Starting from scratch very difficult - many commodities already exist
Automation Paradigms
- Three concepts: automation, adaptive automation, adaptive and inclusive automation
- Automation curve growing vertiginously (exponential/doubly exponential - studies ongoing)
- Two views on automation challenges/benefits
- Regional perception: automation will help/empower but unclear agenda
- Double message: understanding automation benefits but unclear human role in equation
Adaptive Automation Requirements
- Must have very specific implementations, not just concepts/presentations
- People must understand they're part of automation, not displaced by it
- Even if 50-85% of tasks automated, people remain part of definitions
- Must be designed from predictive system cycles
- Tailor-made predictions vs. highly standardized commodities
- Equal importance: tasks to automate vs. tasks strategically not automated
- Design roadmap for people's adaptability in organizations
Intelligent by Design Approach
- Organizations visited: clubs, civil registries, gaming companies, transportation companies, public administrations
- Pattern noticed: those thinking holistically don't do process reengineering with AI as design component
- Advantage: analyzing processes in light of possible predictions that could optimize them
- AI DNA logic from process design
- Machine learning as epicenter of intelligent processes/tasks
- Human role determined after AI-centric design
- Always person-centered at conceptual level
Low Code/No Code Paradigm
- Monumental gap between organizational needs and programming capabilities
- Automation paradigm always mediated by programmer/coding personnel
- Complement with paradigm allowing easier user automation without programming
- Predictive systems surrounded by automation combos/capsules/micro services
- Users can change automations themselves
Small Data/Tiny Data Approach
- Big Data traditional approach (Google Translate example)
- Recent years: predictions with small data
- Active learning, semi-labeling techniques, machine labeling
- Transfer learning techniques becoming more sophisticated
- Artificial vision reusing object/image learning for natural language processing
- Learning with 20-30 examples achieving >90% accuracy rates
- Challenge: bias evaluation with small data
- Opportunity for Latin America to scale AI despite limited data history
Holistic Organizational Approach
- Avatar movie metaphor: everything interconnected, quantum vision
- Tasks articulated in network with mutual impact
- Understanding task impact logic crucial
- Example: transportation company CEO wanting gasoline/naphtha predictions
- Changed strategy after understanding broader task impacts
- 269 total tasks in two large processes example
- Tasks covered by predictions vs. tasks complemented by predictions
- Semi-automatable tasks vs. non-automatable tasks with predictive input
Inter-American Development Bank Case Study
- Saving crucial analysis time for individual consultant hiring
- 32 subtasks with predictive assistant
- Critical thinking/problem solving tasks not displaced
- Fine tasks, subtasks, microtasks distinction important
10-Step Implementation Method
- Method for overcoming economic, social, cultural barriers, data availability issues
- University approach: research to knowledge transfer
- Holistic, empathy, adaptability, inclusive approach
- Creating sustainable ties for long-term sustainability
- Trust challenge: people must trust AI can help them
- Machines complement and enhance human work
- Historical distrust: "heard this promise before" with computers, internet
- Trust generation requires empathy, context adaptation, network creation
Gender Violence Predictive System Project
- Strategic technological plan to address gender violence cases
- December 2nd plan presentation with 5 major verticals
- 45 major technological solutions in certain verticals within process
- Context for predictive algorithms crucial
- Peru work with brief concept, now expanding with Latin American Development Bank
- Artificial vision for PDF complaints
- Open source algorithm for detecting symbolic violence in complaint text
- Anonymized labeled data available
- Learning path documentation for Uruguay, Argentina based on Peru experience
- Sustainable prediction input for Latin America
Latin American Opportunities
- Talent as first/most important asset
- People accustomed to innovating with available resources in adverse scenarios
- Cultural resilience: difficult to give up, no chance of giving up
- Empathetic culture: likes to hug people, warm culture
- Talent + warm culture = major automation era assets
- Need to accelerate talent training ecosystem
- Universities' role in developing predictive systems leaders/developers
Recommendations for Youth
- Three key recommendations for young people finishing education:
- Learn to start startups - not just conceptual, requires robust learning process
- Fail fast approach - don't think linearly, try multiple things simultaneously
- Overcome Instagram culture - work as real teams, not just social media posts
- Alpha generation challenges: liquid, fresh, empathetic but need robust learning
- Network collaboration culture vs. individualistic technology paradigm
- Transdisciplinary approach like Avatar movie concept
- Empathy training for deeper team collaboration
- Recognize achievements/potential of team members
- Iterate permanently, learn from mistakes
