FROM AI TO IA AI: Artificial Intelligence IA: Intelligence Amplification Mieke De Ketelaere, AI/CI @ SAS NEMEA
About myself G.M. De Ketelaere University of Stuttgart, DE G.M. De Ketelaere and H.W. Guesgen PhD, University of Auckland, NZ A neural network system to predict the future energy consumption. A neural network system to aid decision making in induction of labour. Master in Engineering Stuttgart, Germany, 1994 IAAAI Spring Symposium on AI in Medicine, Stanford, US, 1996
Inspiring AI Cases Health Smart City Manufacturing Marketing Defense Fraud
About myself using Google AI ENGINEER DATA SCIENTIST DIRECTOR
How it works: Training the System INPUT Historical Data Train Test Learn OUTPUT Decide AI System
How it works: Using the System INPUT Example: [A, B, C, D,...] New Data Data OUTPUT Example: 0.97 Decide AI System
Types of Machine Learning Different types of Learning by Experience Supervised Learning Unsupervised Learning Reinforcement Learning
How it works: AI & Decisioning New Data Data Listen INPUT Example: Decide Act OUTPUT AI & Decisioning System Example: Mieke De Ketelaere
AI in today s world Moving from the backend to the frontend Goodmorning, Mieke! Blah? Blahblah... Blah!!
How it works: AI & Decisioning New Data Data Listen INPUT Example: Decide Act OUTPUT AI & Decisioning System Example: Mieke De Ketelaere
OUTPUT INPUT Input <-> Output New Data Data Listen Learn Decide Act AI & Decisioning System
Second Wave AI Autogenerated (Perfected) Images AUTOGENERATED PERFECTED IMAGES AUTOGENERATED PERFECTED IMAGES
Second Wave AI Autogenerated Writing and Visual Speech AUTOGENERATED WRITTEN LETTERS AUTOGENERATED SPEECH
Second Wave AI Autogenerated Text AUTOGENERATED CONTENT AUTOGENERATED SUMMARIES
Second Wave AI Autogenerated Art & Music AUTOGENERATED ART AUTOGENERATED MUSIC
Relevance: When things go wrong... Image Recognition
Relevance: When things go wrong... Voice Recognition
Relevance: When things go wrong... Facial Recognition
Relevance: When things go wrong... Natural Language Processing
Relevance: When things go wrong... Facial Recognition
AI has become a Multi-Dimensional Discipline Computer Science Computer Science Business Public Policy Behavioural Science 1960 Now Time
How to gain trust? The FAT Principle: Fair, Accountable and Transparent
Ethical Considerations Art of Decisioning Act Individual Smart Energy Group Robotics Human Subjects Human Involvement Non-Human Subjects
When do we trust decisions made by systems? Relevance + Convenience = Acceptance
Human versus Artificial Intelligence Elementary Differences HUMAN INTELLIGENCE ARTIFICIAL INTELLIGENCE Speed of processing Copy Accuracy on Big Data Modus Size of storage Energy
Human versus Artificial Intelligence Elementary Differences HUMAN INTELLIGENCE ARTIFICIAL INTELLIGENCE Learn Experience Adopt Change Parallel Processing Type of Input Data Creativity Empathy
Human versus Artificial Intelligence Cognitive Bias introduces Algorithmic Bias HUMAN INTELLIGENCE ARTIFICIAL INTELLIGENCE Bias Availability Bias Enchoring Bias Confirmation Bias Overconfidence Bias EXAMPLES
Known Issues Known Issues Historical Data Train Test New Data Data Listen INPUT Intentional Bad Quality Incomplete Inappropriate Legal Ownership Unintentional Cognitive Data Bias Decide Act OUTPUT AI & Decisioning System
Known Issues Historical Data Train Test New Data Data Listen INPUT Known Issues Quality of the model Explainability Model Bias Decide Act OUTPUT AI & Decisioning System
Known Issues Historical Data Train Test New Data Data Listen INPUT Decide Act AI & Decisioning System OUTPUT Known Issues Intentional Timing wrong Full context missing Unneeded automation Unintentional Cognitive Data Bias
The Road Less Travelled Diagnosis Treatment Prevention Listen Computer Science Public Policy Decide Act Business Behavioural Science
Until Algorithmic Accountability is Clear Industrial Community Long-Term Internal High Social Individual Short-Term External Low Context Impact Results Objectives Human Involvement
Summary Getting ready for the future Let s embrace the use of AI, but let s in the mean time also: Request ethics around it: Fair Accountable Transparant Cooperate with the insights generated by AI in order to come to the most optimal decision. Be responsible of training the systems around us, like we would train our children.
Thank you! Mieke De Ketelaere, Director SAS NEMEA Date: October 2018