GOVERNMENT ANALYTICS LEADERSHIP FORUM SAS Canada & The Institute of Public Administration of Canada April 26 The Shaw Centre Ottawa
Artificial Intelligence Algorithms Automation Deep Learning Machine Learning Analytics Cognitive Computing Natural Language Processing Computer Vision Robotic Process Automation
Analysts Perspective AI is a Mega Trend Termed AI Everywhere What s interesting: Many components are at the peak Deep Learning has been added Cognitive experts moved up
Is AI something new? Gartner In January 2016, the term "artificial intelligence" was not in the top 100 search terms By May 2017, the term ranked at No. 7. Hype and "AI Washing" Is Creating Confusion, Obscuring the Real Benefits of the Technology AI Vs Analytics 120 Analytics WW AI WW 100 80 60 40 20 0 2010 2010 2011 2012 2013 2014 2015 2015 2016 2017 Siri Amazon Echo Waymo AlphaGo Source: Gartner https://www.gartner.com/newsroom/id/3763265 July 2017 Google Trends
Why now? Data Computing Power Algorithms
Artificial Intelligence is the science of training systems to emulate human tasks through learning and automation.
AI is a Spectrum Rules-based systems Simplest form of automation, the execution of rules Predictive Analytics Predict, advise, influence, recommend Machine Learning Advanced analytic algorithms create insight with more automation Text Processing The addition of unstructured text AI Components Deep Learning Self learning algorithms that deliver even more insight and automation Robotics Automate repetitive functions and processes Computer Vision The addition of images and video Natural Language Understanding Both natural language ingestion and generation.
Role of Data Machine Data Transactional Data M2M Sensors Unstructured Multi-media Real-time Mobile IoT Social Processes Assets Leverage events and sensors to generate data Allows the move to real time systems Integration of more data provides more meaningful interactions Enterprise Data Voluminous Data Geospatial We can store and consume all data Includes both structured and unstructured Data provides a more holistic picture
Role of Algorithms Machine Learning? Regression Bayesian Statistics Decision trees Gradient boosting Random forests SVM Gaussian processes SUPERVISED LEARNING Teaching by example. We have a target we are predicting. Map inputs to desired output. SEMI-SUPERVISED LEARNING A bit of both Similar to supervised learning. Helpful when volume or variety of data is too high to allow labelling. Regression Decision trees Gradient boosting Random forests Text Processing Image Processing A priori rules Clustering k-means clustering Factorization PCA Network Analysis Affinity Analysis Markov Models UNSUPERVISED LEARNING No answer key is provided. No target is defined. Draws inferences and conclusions based solely on analyzing input data. REINFORCEMENT AI becomes reality. Like teaching someone a game. The machine takes actions. Observes results (trial and error). Maximizes rewards or result. Deep Forward Neural Networks (DNNs) Convolutional Neural Networks (CNNs) Recurrent Neural Networks(RNNs)
What can these algorithms Do? Categorize Predict Identify Detect Automate, Optimize and Learn
What falls under AI? Chatbots Roboadvisor Smart Q&A and Search System Conversational system Facial recognition APPLICATIONS Natural Language Processing Deep Learning Algorithms Self Learning Models Reinforcement Algorithms Machine Learning CAPABILITIES
Government Use Cases Enhance Service Delivery Categorize Predict Identify Detect Ingest unstructured data to expedite. document handling. Implement Chatbots for routine inquires. Reduce processing times and backlog. Find tax evasion. Perform recruit assessment (DND). Predict health care outcomes. Forecast citizen demand. Optimization program spend. Implement sensors to for transportation use cases. Identify program eligibility. Use to analyze Opioid abuse. Look for fraud and waste in government spending.
Focus on the problem Consider the impact ARE YOU READY? Innovate and Experiment Have a Data Strategy Assess Your Risk Tolerance Build Expertise Have Realistic Expectations
Focus on the problem Bring the problem to AI not the other way around. High-volume, routine decision points. Need rules are easily intuited but not easily codified. Look for: Discrete outputs or actions. Determined by diverse inputs.
Consider the impact Augment don t replace. Consider the ethical issues. Not all problems are good candidates. Consider the impact of bias.
Innovate and Experiment
Have a Data Strategy AI thrives on data, make sure you data house is in order. Acknowledge data privacy issues. Deep Learning isn t always the answer you may not have the data for it. Understand the data context.
Assess Your Risk Tolerance
Build Expertise Technical Data engineering skills Technical architecture skills Math and algorithm knowledge Data Science expertise Business Subject Matter Experts Business process Problem solvers Creativity and communication
Have Realistic Expectations You re likely doing some form of it now. AI isn t a silver bullet - progress takes time. Make sure to keep human intervention Start small and grow.
Role of SAS
The Process
AI Capabilities ALGORITHMS Deliver robust analytic algorithms. More accurate analytics. AUTOMATION Automate the analytics lifecycle. Automatically generate models Automate data preparation 23 TEXT AND IMAGE PROCESSING Image processing, classification and recognition. Text processing, Analytics, & Sentiment Analysis. Natural language generation. AI APPLICATIONS Embed AI in SAS Applications Natural language interfaces to SAS.
Impact of Better Analytics Lost Opportunity This is a good outcome Apply a different approach Capture opportunity Traditional analytic techniques work, new techniques are better. Machine learning algorithms increase analytics accuracy. Gradient Boosting, Factorization Machines, Support Vector Machines, Principal Component Analysis, Deep Neural Networks AI can automate Model Selection.
The Lifecycle Bridge the agility gap DATA 1 DISCOVERY 2 DEPLOYMENT 3 1 - Inefficient data pipeline. - Data preparation is cumbersome. - Unable to answer new / unexpected questions efficiently. 2 - Multiple tools for discovery. - Variety of personas to enable - Lack of actionable insights. - Disjointed, inefficient workflows. 25 3 - Multiple hand offs & technologies. - Unable to automate effectively. - Difficult to communicate results & value.
Unstructured Text Valuable set of data often untapped. AI helps review, categorize and make sense of un structured information. Uses include: Program eligibility Waste or fraud detection
AI Applications AI Applications Person/Object Identification Image Processing 1 2 3 4 Personalized Recommendation IoT Analytics Citizen Services
Image classification Platform enables computer vison and image classification. Deliver a full set image classification functions for AI modelling. Paired with Deep Learning to built applications. Over 3000 IMAGES analyzed in < 10 minutes Model Accuracy 78%
We provide capabilities for organizations to build and embed AI into enterprise systems.
SAS AI Approach Delivers capabilities to build AI applications SAS is the analytic standard in the government Provide governed AI processes Support cloud and on premise deployment Simplified end-to-end experience Deliver real time analytics Supported and scalable
Transforming a world of data into a world of intelligence