2. Introspective Systems
Overview
1. What is AI?
2. Why now?
3. AI for Business
4. AI in Marketing
5. Machine Learning
6. Reinforcement Learning
7. Temporal-difference Learning
8. Example: N-Armed Bandit
9. The Next Frontier: Collaborative AI
6. Introspective Systems
AI for Business
● Business Intelligence - trends and irregularities
● Failure Prognostication - equipment
● Marketing
● Product Inventory
● Sales Forecasting
● Chatbots - 24/7 support
● HR - automate screening, paperwork, and data entry
● Security - cyber or otherwise
● Supply Chain and Logistics
7. Introspective Systems
AI for Marketing
Twenty percent of all business advertising content is already AI generated/optimized - Gartner
● Chatbots for great user experience
● Personalization of marketing at scale
● Micro-moments or right information at the
right time, in a real-time environment
● Increasing volume of dark social sharing
● The advent of modern AI and machine
learning tools
● Predictive intelligence based content
creation
Technologies
● Effective risk prediction and interventions
● Efficient predictive data modeling
● Machine learning based pay per click
campaigns
● ML insight based content campaigns
● Highly targeted email campaigns
● Real-time content help through chatbots
and other tools
Enabled Outcomes
8. Introspective Systems
Machine Learning Samuel 1959 - Can learn without being explicitly programmed
Supervised Learning
● Instructive feedback
● Independent of the action taken
● Requires Labeled Training Set
Examples:
● Image Recognition
● Natural Language Parsing
● Pattern Classification
Unsupervised Learning
● Evaluative feedback
● Dependent of the action taken
● Can learn in real-time
Examples:
● Alpha Go
● Adaptive Controllers
● Function optimization
9. Introspective Systems
Machine Learning Samuel 1959 - Can learn without being explicitly programmed
Goal
● Environmental Representations
● Value Functions
Tools
● Neural Networks (deep learning)
● Decision tree learning
● Inductive logic programming
● Clustering
● Reinforcement learning
● Bayesian networks
11. Introspective Systems
Temporal-difference Learning
A combination of:
● Monte Carlo ideas (learn directly from raw experience; no model) and
● dynamic programming (iteratively update estimates).
Policy
Environment
actionreward
state
Value
Function
Actor
Critic
13. Introspective Systems
The Next Frontier: Collaborative AI
Any Questions or Comments?
Introspective Systems Research:
● Department of Energy
● AFRL & USGS
● Starcraft