AUGUST 2021 The Spectrum of Artificial Intelligence Companion to the FPF AI Infographic Contents The Spectrum of Artificial Intelligence Companion to the FPF AI Infographic I. EXECUTIVE SUMMARY ______________________________________ 4 A. Symbolic AI _________________________________________ 4 AUTHORED BY B. Machine Learning _____________________________________ 5 C. Risks and Benefits of AI ________________________________ 5 Brenda Leong II. INTRODUCTION ___________________________________________ 6 Senior Counsel & Director of Artificial Intelligence and Ethics III. FOUNDATION DISCIPLINES __________________________________ 7 Philosophy _____________________________________________ 7 Dr. Sara R. Jordan Ethics _________________________________________________ 7 Senior Researcher, Artificial Intelligence and Ethics Logic __________________________________________________ 8 Mathematics ____________________________________________ 8 Physics ________________________________________________ 8 IV. MODERN COMPONENTS ____________________________________ 8 A. Data _______________________________________________ 8 B. Statistics ____________________________________________ 9 C. Design _____________________________________________ 9 Security _______________________________________________ 9 Hardware ______________________________________________ 9 COVER ART V. ARTIFICIAL INTELLIGENCE (NON-ML) – OVERVIEW ______________ 10 Artist: Dr. Lydia Kostopoulos A. Rules Based AI ______________________________________ 10 B. Symbolic AI __________________________________________11 Title: Luncheon of the Tech Enabled Boating Party i. Search ________________________________________11 Year: 2020 ii. Planning and Scheduling __________________________11 iii. Expert Systems ________________________________ 12 C. Computer Sensing ___________________________________ 12 D. Robotics ___________________________________________ 13 E. Knowledge Engineering _______________________________ 13 F. Natural Language Processing ___________________________ 14 G. Risks and Benefits – Rules-based AI ______________________ 14 VI. ARTIFICIAL INTELLIGENCE (MACHINE LEARNING) _______________ 15 A. Reinforcement Learning _______________________________ 15 B. Neural Networks _____________________________________ 16 C. Deep Learning ______________________________________ 17 D. Generative Adversarial Networks ________________________ 17 E. Risks and Benefits – Machine Learning ____________________ 17 VII. CONCLUSION ____________________________________________ 19 THE SPECTRUM OF ARTIFICIAL INTELLIGENCE | COMPANION TO THE FPF AI INFOGRAPHIC 1 THE SPECTRUM OF ARTIFICIAL INTELLIGENCE Produced by Artificial Intelligence (AI) is the computerized ability to perform tasks commonly associated with human intelligence, including reasoning, discovering patterns and meaning, FPF.ORG generalizing, applying knowledge across spheres of application, and learning from experience. The growth of AI-based systems in recent years has garnered much attention, particularly in the sphere of Machine Learning. A subset of AI, Machine Learning (ML) systems “learn” from the success or accuracy of their outputs, and can change their processing over time, with MACHINE LEARNING minimal human intervention. But there are non-ML types of AI that, alone or in combination, lie behind the real-world applications in common use. General AI — a human-level ML Algorithms improve through experience computational system — does not yet exist. But Narrow AI exists in many fields and applications where computerized systems greatly enhance human output or outperform humans at defined tasks. This chart explains the main types of AI, their relationships to each other, and provides specific examples of how they are currently appear in our day-to-day lives. It also demonstrates how AI exists within the timeline of human knowledge and development. GENERATIVE ADVERSARIAL NETWORKS GAN Two NNs learn by fighting SYMBOLIC AI SA Human-readable logic problems AI USE CASES AND CONTEXTS DEEP LEARNING DL Multiple layers of FINANCE neural networks TAX COMPLIANCE A software platform that distills tax laws into a EXPERT SYSTEMS ES Complex solutions program, creates a personalized decision system, through reasoning REINFORCEMENT LEARNING and enables individuals to quickly and accurately RL Learning to complete a task file their taxes. Value of AI: Tax compliance requires complete accuracy. This efficient, interactive system that provides precise and logically connected results SEARCH S Steps from initial allows taxpayers to understand, confirm, and have state to goal confidence in the outcome. KE provides transparent and clear explanations. NEURAL NETWORKS ROBOTICS DEEP LEARNINGNN Learning by making R Multi-sensing and/or DL Multiple layers of DESIGN USER Types of AI: connections PLANNING & SCHEDULING mobile AI neural networks P&S Multi-dimensional strategies and action sequences EXPERIENCE KE NN NLP USER INTERFACE RULES BASED HEALTHCARE RB Deductions based on curated rules KNOWLEDGE ENGINEERING KE Rules based on human expertise AMBIENT CHARTING PHIL The use of background voice-to-text processing OSOPHY FOU during a patient/medical provider exchange to record those interactions into the patient’s chart, ND along with extracting tasks, symptoms, and ATION recommendations for further action as required. MA A Value of AI: Medical providers spend significant L TECHNO THEMA COMPUTER SENSING time documenting, with uneven outputs, as well as CS Human sense-based inputs difficulty in correlating between providers. Ambient ETHICS TICS systems encode conversations, target key phrases, LO NATURAL LANGUAGE PROCESSING GY NLP Understand, interpret, manipulate language and present a summary for provider edit/acceptance. STATISTICS PHYSICS DATA BUSINESS Types of AI: LOGIC ANAL ANALYTICS YSIS SECURITY SA DL NLP ENCRYPTION MODELING TRACKING MOBILITY AND TRANSPORTATION SOCIAL MEDIA FORECASTING WORKPLACE MONITORING TURN-BY-TURN NAVIGATION SPEECH OR CONTENT MODERATION SUPPLY CHAIN MANAGEMENT Embedded systems can monitor physical and digital Location-based software that provides detailed Systems can facilitate human teams in identifying, Systems to improve traditional inventory and traffic, data usage, device management, and some instructions for travelers to reach a selected flagging, and deleting posts with defined, prohibited forecasting beyond historical/internal trend data, employee behaviors for efficiency and security designation, customizable mode of transportation, terms (such as “hate speech” or profanity). to weight and include external factors such as management of time, assets, and resources. multiple stops, services en route, and real-time Categorizing and selectively reacting based on weather, consumer sentiment, demographic trends, Value of AI: Monitoring enables necessary adjustments based on traffic, tolls, and weather. platform policies, usually embedded in analysis of portal traffic, stock fluctuations, and enforcement of data security policies and protocols. Value of AI: This is a “shortest path” problem human/computer systems for review and decision. service levels HARDWARE Also, systems can monitor and manage time solver, able to consider and weight variables such Value of AI: More efficient at scale than human-alone Value of AI: Systems can increase accuracy and reporting and project management tools, as well as as speed, cost, and personal preferences, and reviews. Additionally, well-designed systems can efficiency, as well as provide improved transparency ensuring appropriate supervision, training, and allow personalization based on repeated journeys, potentially adapt to variations in context, intent, and reliable, predictive analytics; enable aggregate support, including for remote workers as well as link to calendar and scheduling data, cultural norms, and user expectations more forecasting from individual impact up through and interactive prompts. consistently across platforms. regional levels. Types of AI: Types of AI: Types of AI: Types of AI: RB CS NN S SA DL GAN KE NLP RL P&S ES KE ML I. Executive Summary This paper is a companion piece to the FPF Spectrum several algorithmic designs that are generally consid- image) based upon the initial human programming. The of Artificial Intelligence (AI) Infographic (fpf.org/blog/ ered to be examples of Symbolic AI, including Search, B. Machine Learning other network, the “discriminator,” has been programmed the-spectrum-of-artificial-intelligence-an-infographic-tool/), Planning and Scheduling, and Expert Systems. to what the correct output should be (e.g., what the image to expand the information included in that educational The types of AI described above tell computers how to should look like). The discriminator evaluates the output, When computers are programmed to find a specific pat- resource, and describe how the graphic can be used sift through information according to rules and process- and critiques it. Initial outputs are likely to be extremely tern in a set of symbols and then to perform a designated as an aide in developing legislation or other regulatory es crafted by humans, such as language or mathematics. inaccurate. The discriminator’s feedback is then incorpo- action, we can say that the AI is engaged in a “search.” guidance impacting AI-based systems. We identify spe- Machine learning is different. Machine learning works rated, the generator continues to churn out results, and Planning and scheduling AI are what enable a comput- cific use cases for various AI technologies and show how because machines