Introduction to Artificial Intelligence

Transcript

1. Ali Akbar Septiandri Introduction to Artificial Intelligence
2. TABLE OF CONTENTS WHAT IS AI? Why do people talk about it? IN DEPTH What is deep learning? Why is it different from machine learning? 01 03 02 04 05 06 AI in INDONESIA Where are we now? ETHICAL AI Is AI innately unbiased? FUTURE of AI What’s next for AI? AI for EVERYONE How to prepare individuals and companies for AI era?
3. WHAT IS AI? 01 Why do people talk about it?
4. Alan Turing, Computing Machinery and Intelligence I propose to consider the question, “Can machines think?” This should begin with definitions of the meaning of the terms “machine” and “think.”
5. • Can machines be more intelligent than humans? • If so, what will happen to humanity? • How can we measure human intelligence? SOME PHILOSOPHICAL QUESTIONS
6. Source: Digital Wellbeing
8. Full Self-Driving Tesla
9. AI Adoption by McKinsey Global Institute (2017)
10. AI DESIGNS • Rational actions • Mathematical and empirical evaluation • Psychology, neuroscience • Optimization theory, statistics, game theory, etc.
11. Source: NVIDIA
12. Jake VanderPlas “Fundamentally, machine learning involves building mathematical models to help understand data.”
14. WHY IS IT HARD?
15. WHY IS IT HARD?
16. WHAT IS IN THIS PICTURE?
17. WHAT IS IN THIS PICTURE?
18. ENTER: DEEP LEARNING This is the one responsible for the AI hype nowadays
19. A DEEP NEURAL NETWORK
21. IN DEPTH 02 What is deep learning? Why is it different from machine learning?
22. LEVELS OF INTELLIGENCE Simplest form of thought process Search problems, Markov decision processes, adversarial games Constraint satisfaction problems, Bayesian networks First-order logic, knowledge base
23. MACHINE LEARNING TASKS UNSUPERVISED LEARNING SUPERVISED LEARNING REINFORCEMENT LEARNING
24. SUPERVISED LEARNING Even better, can we tell that there is a cat and a dog in the image? How to identify cats or dogs in an image?
25. BEFORE (CLASSICAL ML) IMAGE PROCESSING Edge detection, texture analyser, color histogram FEATURE EXTRACTION Eye position, eye colour, nose colour, fur type, leg counts MODEL Logit model, SVM, k-NN
26. y = σ(β 0 + β 1 x 1 + β 2 x 2 + β 3 x 3 ) Trying to find the optimum weights for each feature
27. AFTER (DEEP LEARNING) INPUT You only need raw images! FEATURE EXTRACTION + MODEL TRAINING Your model will do the feature extraction step for you OUTPUT Just define how many classes
28. ...but we need millions of images!
29. images from 20k categories AlexNet achieved a top-5 error of 15.3% 14 mio
30. Convolutional Neural Networks
31. Generative Adversarial Networks (GANs)
32. Recurrent Neural Networks
33. UNSUPERVISED LEARNING Customer segmentation (clustering) Topic modelling Anomaly or outlier detection
34. UNSUPERVISED LEARNING Word Representations
35. REINFORCEMENT LEARNING Monte Carlo Tree Search for AlphaGo
36. Source: NVIDIA
37. AI in INDONESIA 03 Where are we now?
43. ETHICAL AI 04 Is AI innately unbiased?
50. BIAS and FAIRNESS in AI
51. “...a project to look for skin cancer in photographs. It turns out that dermatologists often put rulers in photos of skin cancer, for scale, but that the example photos of healthy skin do not contain rulers. To the system, the rulers (or rather, the pixels that we see as a ruler) were just differences between the example sets, and sometimes more prominent than the small blotches on the skin. So, the system that was built to detect skin cancer was, sometimes, detecting rulers instead.” (Evans, 2019) BIAS in AI
52. UNDERSTANDING CAUSALITY Confounders Graphical presentation of confounding in directed acyclic graphs (Suttorp et al., 2014) Age Chronic kidney disease Mortality
53. Dermatoscopic images See (Finlayson et al., 2019) ADVERSARIAL ATTACKS
54. FUTURE of AI 05 What’s next for AI?
55. Is it even possible? ARTIFICIAL GENERAL INTELLIGENCE
56. AlphaStar by DeepMind
57. Generating Images from Brain Signals
59. AI-Aided Prediction Deep Neural Networks Improve Radiologists’ Performance in Breast Cancer Screening (Wu et al., 2019) Human+AI AUC, Malignant Prediction
60. AI for EVERYONE 06 How to prepare individuals and companies for AI era?
62. AI Transformation Execute pilot projects to gain momentum Customer segmentation, review categorization, photo classification 1st Build an in-house AI team Find people with specific skills related to ML 2nd Provide broad AI training Everyone should learn what AI is! Be it managers, team leaders, etc. 3rd Develop an AI strategy What are some vital things to enhance? 4th Develop internal & external communications Talk with board of directors, investors, your precious stakeholders 5th
63. WHAT ML CAN DO Anything you can do with 1 second of thought, we can probably now or soon automate
64. WHAT ML CANNOT DO e.g. Market research and run an extended market report, give empathetic responses
65. WORKING with an AI TEAM 1 Define acceptance criteria Sometimes, you don’t really need 100% accuracy 3 Ensure clean data Big data without correct labels can mislead you! 2 Find sufficient data The bigger, the better
66. AI PITFALLS to AVOID Don’t think you need superstar AI engineers Don’t expect traditional planning w/o changes Don’t expect it to work the first time Don’t just hire 2/3 super ML engineers Don’t expect AI to solve everything
67. by Ainun Najib AI/ML for KIDS
69. CREDITS: This presentation template was created by Slidesgo, including icons by Flaticon, and infographics & images by Freepik. Please keep this slide for attribution. Does anyone have any questions? pm@aliakbars.id @aliakbars uai.aliakbars.id THANKS