Module 1 AI Disruptions & Introduction to AI AI and Society Prof Yu Chien Siang yucs@outlook.com This module exposes the student on background knowhow to understand how AI will impact Society, both positively and negatively. Its practical value is to teach leadership and confidence, motivating and enhancing the student’s value when he enters the job market; e.g. arm him with the skills to lead or contribute towards AI automation projects, which are fundamentally different from standard IT projects, as AI systems tend to be both highly diverse and complex. Skills for Future Jobs Most AI Deep Learning and data science projects fail, despite promising test results and clear problem statements at the start. This is because many users may start off with unrealistic expectations due to an inadequate understanding of AI; what it can and cannot do, and business leaders may fail to understand the privacy, ethics, cyber security and governance issues related to this new disruptive science. Additionally, there may not be enough high quality data available, and users do not understand the importance of garbage in, garbage out. Therefore, a key aim of this module is to provide a business deployment and societal perspective of foundational Deep Learning AI capabilities, ranging from Natural Language Processing to Machine Vision; in a manner that requires little or no coding. Neural networks and how Deep Learning works, the data input requirements, project lifecycle and the development framework would be introduced. What problems that each specific AI algorithm can solve would be illustrated via virtual lab workshop sessions and the enterprise benefits would be highlighted via case studies; e.g. enhanced productivity for corporate decision making, process automation or state-of-art unattended (autonomous) systems. The focus is on rapid problem understanding and to explore AI project viability for real world problem solving. Importantly, the student will also gain insights to discuss AI regulatory compliance, privacy and cyber security requirements at the design phase. Skills for Future Jobs Finally, modern AI governance challenges such as ensuring man-in-the-loop AI oversight, cyber security hardening against adversarial AI attacks and privacy obligations would be covered as new requirements for the digital economy. The future of AI continues to shine brightly, and we aim to highlight major high value digitalisation and enterprise automation possibilities as AI and Big Data systems converge pervasively, and 5G and IoT AI at the edge become commodity. With these insights, the student will master multidisciplinary innovation and thinking skills that will be the main drivers to deliver AI strategic benefits to the corporation. Applied Learning Objectives Very up-to-date, inspirational, valuable, broad and deep coverage of AI Confidence boosting, get used to learning complicated things, be street smart, non-stop • Big Picture Section (minus coding and the mathematical details) • Architecture and History • Appreciate the value and power of the knowledge gained • Usage examples • Innovation and opportunities • Learn also the negatives really well • Engineering Section (modular, key components, problem solving) • Learn how AI works, how to get problems solved and projects done via coding ideas. • Rapid Prototyping, Agile Mindset • Toolset empowerment – learn an idea, correlate it to the Big Picture and be able to exploit what has been learnt to solve problems immediately by exploiting codeless/code lite production and code simulation using Colab/Juypter Notebook “AI has the potential to change my life for the better, agree 71%.” - Gemalto Applied Learning Pedagogy Multi-modal Learning, self driven learning - Extensive use of animation - Video sequences and multimedia - Instant gratification via virtual lab exercises, facilitated by experimental tuning Team-based Learning - Modified for remote learning (to cope with Covid) Reinforcement Techniques - 7 repeats (Learning the same thing 7 times but with different points of view) - Try and then see what happens (open lab experimentation) Life Long Learning and Skills for Thinking • The cognitive development objective is to guide the students to develop their own "compass" for learning and achievements, be critical about existing systems and norms as these may be disrupted quickly. • Knowing important AI issues in depth helps to nurture our students to be deep thinkers. • Think innovation all the time! Lessons and Test Plan Lectures 13 of about 3 hrs each • Week 1 – Introduction to AI and Smart Nation Society • Week 2 – Usage and Models (Lab 1) • Week 3 – Natural Language Processing (NLP) Use Cases (Lab 2) • Week 4 – Natural Language Processing (NLP) Smart Factory Quiz1 (for 1 and 2) 30 MCQ questions • Week 5 – AI Vision Applications (Lab 3) • Week 6 – AI Vision Embedded Systems Mid-course Test, 40 MCQ Questions • Week 7 – IoT, Robots and Drones, RPA (Lab 4) • Week 8 – Economics, AI Project Management and Leadership • Week 9 – AI Governance and Compliance • Week 10 – AI Ethics, Legislation and Politics Quiz2 (for 3 and 4) 30 MCQ Questions • Week 11 – AI and Cyber Security (Lab 5) • Week 12 –The Future of AI • Week 13 – AI and Education End of Course Final Test, 60 MCQ Questions eLab Sessions Five lab sessions will be conducted on related weeks. The objective of these exercises will be to further elaborate on the practical aspects via hands-on practice using the concepts introduced during the lectures. • Lab 1: Understanding Neural networks CNN • Lab 2: NLP Applications • Lab 3: Machine Vision Applications • Lab 4: Robotic Process Automation (RPA) • Lab 5: AI Governance and Cyber Security Overseas Example: A powerful collaboration: MIT Sloan and MIT CSAIL Artificial Intelligence: Implications for Business Strategy (self-paced online) Certificate Track: Management and Leadership Location: Online, 6 weeks Tuition: US$3,200 Program Days (for ACE Credit) 2 Assessments All the MCQ tests are open book • Team Paper with a video • • Team paper Video 30% 20% 10% • Individual Powerpoint 10% • Individual Quiz/Assessment 30% • Lab Quiz1 MCQ 30 Questions – End of Week 4 (first 3 lectures plus Lab 1 and 2) – 5% • Test2 Midterm MCQ 40 Questions – End of Week 6 (1-6 lectures) – 20% • Lab Quiz2 MCQ 30 Questions – End of Week 10 (7-9 lectures plus Lab 3 and 4) – 5% • Final Test/Exam 30% • MCQ 60 Questions (coverage of the entire course plus Lab5) Team Project Assessments • Team Paper with a video • Topic of paper to be about AI implications or opportunities for the future society. Students are required to work on a group project with a written report (20 A4 pages) plus a video summary (8 minutes). • Font size for Text to be 11 (e.g. Times Roman) and Headings to be font size 12. • Margins - minimum 2.5 cm, maximum 3 cm all around. • Single line spacing. • Group project outcomes should depict thoughtful discussion and debate on how best to operationalize new AI opportunities cum its impact to the new digital economy and to also study potential challenges. The video presentation should show original thinking, creativity and communication skills. • Individual Powerpoint • Each team member to produce his own individual powerpoint (20 slides) to show active participation. • Rubric for Team Project Paper Marking • • • • • Holistic Content (well organised, comprehensive, logically structured) Innovation (novel, interesting, flexible design) Critical Thinking (compared pros and cons, handle negatives and possible issues) Operacy Thinking (how practical, how relevant to real world problem solving and economics) Paper, video and powerpoints to be submitted by end of week 13 - 40% - 20% - 20% - 20% Possible Team Projects (some examples) • Non-Technical Papers • • • • • The impact of AI on the economy Jobs Changes and AI, or The Future of Jobs as impacted by AI How would AI Ethics impact enterprise governance Is AI security a really serious problem? Why must we care about AI implications to Smart Nation? • More-technical papers • • • • • Future of AI and AGI How would AI impact office IT and boost workers’ productivity? Chatbots and AI AI Security, or AI and its Operational Safety What is XAI and why is this important? Overview of Google Colab as Lab Platform • Free cloud service hosted by Google to encourage Machine Learning and Artificial Intelligence. • Options to choose to run CPU,GPU or TPU • Utilises the Jupyter notebook environment to execute your code • Allows simultaneous access by different team members for editing • Supports many popular machine learning libraries which can be loaded easily in the notebook • Most importantly, you will need to get a Google account. • Give it a try. Start off with https://www.tensorflow.org/tutorials/quickstart/beginner • Click on… Google Colaboratory notebook file. Colab Usage Guide: Resources • https://www.geeksforgeeks.org/how-to-use-google-colab • https://analyticsindiamag.com/a-beginners-guide-to-using-googlecolab/ • https://www.tutorialspoint.com/google colab/google colab quick g uide.htm • https://www.analyticsvidhya.com/blog/2020/03/google-colabmachine-learning-deep-learning/ • https://towardsdatascience.com/10-tips-for-a-better-google-colabexperience-33f8fe721b82 Important things to note 1. Mounting your google drive to Colab 2. Activate GPU or TPU 3. Run bash commands 4. Stop Colab from disconnecting during training 5. Installing libraries 6. Upload files 7. Other accounts needed 1. Mounting your google drive to Colab • This function is useful for getting access to the files in google drive and saving the data and models once the training is down • The code for it is: • from google.colab import drive drive.mount('/content/gdrive’) • After running the code, a link will be given in the output and redirected to the permission page for the authentication code that is to be entered in the output Redirected permission page 1. Mounting your google drive to Colab • Once you have allowed permission, a verification code will be generated and you have to copy the code into the textbox given in the output 1. Mounting your google drive to Colab • Once you have entered the verification code, you should receive an output: • Mounted at /content/drive 2. Activate GPU or TPU • The default hardware used in Google Colab is CPU • You must enable either GPU or TPU for more computationally demanding tasks such as machine learning • Click on: “Runtime” → “Change runtime type” → “Hardware accelerator”. Then select the desired hardware. 3. Run bash commands • Google Colab allows users to run terminal codes • Bash commands can be run by prefixing the command with “!” • Examples: • Download dataset from the web with !wget <ENTER URL> • Install libraries with !pip install <LIBRARY> or !apt-get install <LIBRARY> • Run an existing .py script with !python script.py • Make sure that you are in the correct directory before executing • %cd <file path> • Clone a git repository with !git clone <REPOSITORY URL> 4. Stop Colab from disconnecting during training • This is important especially if you are doing model training • Google Colab will automatically disconnect after running a task for 12 hours or idle for more than 30 minutes • Open your Chrome DevTools by pressing F12 or ctrl+shift+i on Linux and enter the following JavaScript snippet in your console: • function KeepClicking(){ console.log("Clicking"); document.querySelector("colab-connect-button").click() } setInterval(KeepClicking,60000) • This function makes a click on the connection-button every 60 seconds to ensure that the notebook is not idle and model training will continue until the end Usage of the JavaScript code 5. Installing libraries • Although most of the commonly used python libraries are preinstalled, there will be some missing libraries required • Code: • !pip install [package name] OR • !apt-get install [package name] • Place the code in a cell and execute it 6. Upload files • https://colab.research.google.com/notebook s/io.ipynb • On the left side of the GUI, you will a file folder that you can click on it. It will show an explorer view of the file directories at your colab site. Do a mouse click at 3 dots icon, for options to create a new directory, delete files or folder or “upload” a file. • If you have python code, you could cut the entire text out and paste it into a Juypter cell. 7. Other accounts that could be useful • Get a Kaggle account, to get access to data and tutorials. • Kaggle.com • https://www.youtube.com/watch?v=GJBOMWpLpTQ Join competition • https://www.kaggle.com/learn/overview?utm medium=youtube&utm source=c hannel&utm campaign=yt-learn free courses • Got a Github account to get access to codes/Juypter notebooks. • Go to https://github.com/join • Type a user name, your email address, and a password. • Choose Sign up for GitHub, and then follow the instructions. Module 1: Introduction to AI (Deep Learning) and AI-driven disruptions? • Why is Deep Learning/AI revolutionary? • Explaining how Deep Neural Networks (DNN) works. • Compare Deep Learning (DL) with Machine Learning (ML). • Digitalisation and the 4th Industrial revolution. • How are jobs affected? Deep Learning (deep neural network) • https://en.wikipedia.org/wiki/Deep learning • Family of machine learning methods based on learning representations of data. • An observation (e.g., an image) can be represented in many ways such as a vector of intensity values per pixel, or in a more abstract way as a set of edges, regions of particular shape, etc.. • Some representations make it easier to learn tasks (e.g., face recognition or facial expression recognition) from examples. • Replacing handcrafted features with efficient algorithms for unsupervised or semisupervised feature learning and hierarchical feature extraction. • http://www.datarobot.com/blog/a-primer-on-deep-learning/ • Youtube: https://www.youtube.com/watch?v=vT1JzLTH4G4&list=PL3FW7Lu3i5JvHM8ljYjzLfQRF3EO8sYv Stanford 2017 Lecture 1 | Introduction to Convolutional Neural Networks for Visual Recognition Who’s who in Deep Learning • Mar 2013, Geoffrey Hinton and two of his graduate students, Alex Krizhevsky and Ilya Sutskever, were hired by Google. (bought DNNResearch) • Dec 2013, Facebook hired Yann LeCun to head its AI Lab. • 2014 Google also acquired DeepMind Technologies, a British start-up that developed a system capable of learning how to play Atari video games using only raw pixels as data input. • 2014, Baidu hired Andrew Ng (ex-Stanford) and Adams Coates to head their new Silicon Valley based research lab. He started Coursera. Left Baidu. • Deep Speech Engine (noisy environment) for voice commands, GPU Difference between ML (statistical) vs DL • https://blogs.nvidia.com/blog/2016/07/29/whats-difference-artificialintelligence-machine-learning-deep-learning-ai/ • Machine Learning at its most basic is the practice of using algorithms to parse data, learn from it, and then make a determination or prediction about something in the world. So rather than hand-coding software routines with a specific set of instructions to accomplish a particular task, the machine is “trained” using large amounts of data and algorithms that give it the ability to learn how to perform the task. • Machine learning came directly from minds of the early AI crowd, and the algorithmic approaches over the years included decision tree learning, inductive logic programming. clustering, reinforcement learning, and Bayesian networks among others. As we know, none achieved the ultimate goal of General AI, and even Narrow AI was mostly out of reach with early machine learning approaches. Difference between ML (statistical) vs DL • https://blogs.nvidia.com/blog/2016/07/29/whats-difference-artificialintelligence-machine-learning-deep-learning-ai/ • As it turned out, one of the very best application areas for machine learning for many years was computer vision, though it still required a great deal of hand-coding to get the job done. People would go in and write hand-coded classifiers like edge detection filters so the program could identify where an object started and stopped; shape detection to determine if it had eight sides; a classifier to recognize the letters “S-T-O-P.” From all those handcoded classifiers they would develop algorithms to make sense of the image and “learn” to determine whether it was a stop sign. • Good, but not mind-bendingly great. Especially on a foggy day when the sign isn’t perfectly visible, or a tree obscures part of it. There’s a reason computer vision and image detection didn’t come close to rivaling humans until very recently, it was too brittle and too prone to error. Possible H(v) And after Digital Human Example of DeepFakes Exciting Demos • GPT3 • https://research.aimultiple.com/gpt/ (several movie demos here) • More: • https://www.entrepreneur.com/article/368843 • https://www.theguardian.com/commentisfree/2020/sep/08/robot-wrotethis-article-gpt-3 An introduction to Reinforcement Learning • https://www.youtube.com/watch?v=JgvyzIkgxF0 • Video – 16 min 26 sec Business Models are being disrupted • Business Models are changing. • • • • • Banks going to the cloud CloudATM Taxi electronics get obsoleted by mobile app eSIM coming Paperless 59 Disruptions rocking the world of IT • Kodak was the leader in producing film negatives. In 2013, it went bankrupt. The company invented digital cameras. The pace of change was faster than expected. Interestingly, Fujitsu survived. • Disruptions, challenging IT today: • Cloud First (SIA) • Fintech /Blockchain • National Digital Identity (Smart Nation), mobile phone is the central enabler! • IoT and 4th Industrial Revolution • Connected Cars, Robots and Drones • Big Data, AI (embedded AI) • Quantum Computing • Facebook saga, GDPR, regulations on fake news 60 Late 2017, McKinsey Global Institute Report • https://insights.dice.com/2018/10/03/4-tech-jobs-risk-automation-artificialintelligence/ • Automation will leave many technology jobs relatively untouched. “Workers of the future will spend more time on activities that machines are less capable of, such as managing people, applying expertise, and communicating with others,” that report read. “The skills and capabilities required will also shift, requiring more social and emotional skills and more advanced cognitive capabilities, such as logical reasoning and creativity.” • Other studies have positioned automation as helping tech pros perform their jobs more effectively. For example, one study from Puppet demonstrated that automated tools had taken over roughly a third of testing and configuration-management duties from DevOps specialists, freeing them to pursue more proactive tasks. Late 2017, McKinsey Global Institute Report • But automation won’t prove a good thing for every tech pro. Many companies will look at these labour-saving tools and decide they can get by with smaller teams of technologists. • While high performers will benefit from automation, it stands to reason that not everyone will survive when new, more sophisticated platforms come online. With that in mind, here are four tech jobs most at risk from automation and A.I. The following will be negatively affected: • • • • Datacenter Administrators Help Desk Staff Programmers Data Analysts Andrew Ng’s words of wisdom • Much has been written about AI’s potential to reflect both the best and the worst of humanity. For example, we have seen AI providing conversation and comfort to the lonely; we have also seen AI engaging in racial discrimination. • Yet the biggest harm that AI is likely to do to individuals in the short term is job displacement, as the amount of work we can automate with AI is vastly bigger than before. • As leaders, it is incumbent on all of us to make sure we are building a world in which every individual has an opportunity to thrive. Understanding what AI can do and how it fits into your strategy is the beginning, not the end, of that process. Conclusion • Massive job loss is not immediate. • When will it start… already started, and will accelerate quickly. • Covid crisis will pressure all companies to go digital much faster, e.g. cosmetics manufacturing, schools, banking services. • Is it old wine in new bottle; like global warming, globalisation etc, that we have experienced before? • How will AI make its greatest impact? • Changes to world order, nature of our economy, how we will live, our hopes and aspiration for the future etc. • Who will benefit most? • The various Internet giants • You! As you have seen the future. • What else is important? • Cyber and data security Reading (Reinforcement Learning): • View this Rajiv production for reinforcement learning (2017), Deep Q Learning for Video Games - The Math of Intelligence #9, 12 Aug 2017, 9 min https://www.youtube.com/watch?v=79pmNdyxEGo • We're going to replicate DeepMind's Deep Q Learning algorithm for Super Mario Bros! This bot will be able to play a bunch of different video games by using reinforcement learning. This is the first video in this series that uses libraries (Keras & Gym) because if it didn't, the code would be way too long for a short video. I'll make a longer, in-depth version without libraries soon. Code for this video: https://github.com/llSourcell/deep_q_...