DOCSLIB.ORG

Introduction

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Introduction 15-780: Graduate AI Introduction Instructors: Tuomas Sandholm and Manuela Veloso TAs: John Dickerson and Prateek Tandon http://www.cs.cmu.edu/~15780 Carnegie Mellon University Evaluation n 4 Homeworks 40% n Midterm Test 20% n Final Project 30% n Class participation 10% n Homeworks are done individually n 8 total “mercy” days (no penalty) for late homeworks - cannot use more than 3 mercy days in a single homework. No credit for late homeworks with no mercy days. n Final projects can be done in groups of up to 2 people 1 Resources n Lectures n Presentation and discussion in class n Lecture slides rd n Textbook: Russell and Norvig, 3 edition n Instructors – office hours by appointment n TAs – office hours will be announced More Information n TAs will hold recitations as needed n Doodle poll for best time n Audits n Register by fill out audit form n Must do final project n Waitlist n Let us know if you are on it n Mailing list n 15780students @ cs.cmu.edu n We will send a test message 2 What is Artificial Intelligence? What is “intelligence” ? Can we emulate intelligent behavior in machines ? How far can we take it ? Time Line (NYTimes 2011) 3 n Performed for 84 years n Defeated Napoleon and Franklin n Amazon Mechanical Turk: “artificial artificial intelligence” 7 Amazon Mechanical Turk 8 4 n Can a general machine think? n Can machines “pass” the imitation game? n Turing Test: Evaluator communicates via text channel with computer and human, must reliably identify the computer 9 Views of AI Think like humans Think rationally Cognitive Science Formalize inference into laws of thought Act like humans Act rationally Turing test Act according to laws 5 The Dartmouth Conference “We propose that a two-month, ten-man study of artificial intelligence be carried out during the summer of 1956 at Dartmouth College in Hanover, NH. The study is to proceed on the basis of the conjecture that every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it.” The Proponents n John McCarthy, assistant professor of mathematics at Dartmouth (2011) n Marvin Minsky, Harvard junior fellow in mathematics and neurology (MIT) n Nathaniel Rochester, manager of information research at IBM, NY (2001) n Claude Shannon, information theory, mathematician at Bell Labs (2001) 6 The Invited n Trenchard More, IBM n Arthur Samuel, IBM n Oliver Selfridge, Lincoln Labs, MIT n Ray Solomoff, MIT And “two vaguely known persons from RAND and Carnegie Tech… a significant afterthought.” (Pamela McCorduck, “Machines Who Think”, page 94) Herbert A. Simon and Allen Newell 7 Computer Chess at CMU n A long history n HiTech – Hans Berliner, Murray Campbell n ChipTest – Feng-Hsiung Hsu n Deep Thought – Hsu, Campbell, Anantharaman n Deep Blue – CMU team hired by IBM Research 8 Problem Solving n Allen Newell and Herb Simon – 1950s n Given: n an initial state n a set of actions n a goal statement n Search for a plan, a sequence of actions that transform the initial state into a state where the goal is satisfied 9 Intelligent Systems Three key steps (Craik, 1943): 1. the stimulus must be translated into an internal representation 2. the representation is manipulated by cognitive processes to derive new internal representations 3. internal representations are translated into action “agent” perception cognition action 10 Intelligent Agents n Sensing: vision, hearing, touch, smell n Cognition: think, reason, plan, learn n Action: motion, speech, manipulation 11 Perception – Sensors to State n Sensors – “signal” (data) collectors from the physical world: n Vision, sound, touch, sonar, laser, infrared, GPS, temperature,…. n Signal-to-symbol challenge: n Recognize the state of the environment n …wall at 2m… door on the left… green light… person in front… personX entering the room… ball at 1m and 30o East… Learning n Automatically generate strategies to classify or predict from training examples Classification: Is the Training data: Example object present in the images of object input image, yes/no? 12 Learning n Automatically generate strategies to classify or predict from training examples Mpg good/bad Predict mpg on new data Training data: good/bad mpg for example cars Web Search 13 Statistical Learning Recommender Systems 14 Question Answering – Watson! Multiagent Systems and Learning n How can an agent learn from experience in a world that contains other agents too ? n Other agents’ learning makes the world nonstationary for the former agent n Games n Learn to play Nash equilibrium n Learn to play optimally against static opponents 15 Games n Multiple agents maybe competing or cooperating to achieve a task n Capabilities for finding strategies, equilibrium between agents, auctioning, bargaining, negotiating. n Business n E-commerce n Multi-robot systems n Investment management n ….. Kidney Exchange n In US, ≥ 50,000/yr are diagnosed with lethal kidney disease n Cure = transplantation, but cadaver kidney have a long waiting list (2-5 yrs) n Potential donors may be incompatible with patient 32 16 Kidney Exchange Pair 1 Pair 2 Patient 1 Patient 2 Donor 1 Donor 2 Kidneys usually travel inside donors Optimizing kidney exchange n Cycle cover = optimize matched vertices under cycles 1 2 3 4 of length ≤ L n Problem is NP-hard 5 6 34 17 RoboCup n Robot soccer competition and symposium n Challenge: By 2050, a team of fully autonomous humanoid robot soccer players shall win the soccer game, complying with the official rule of the FIFA, against the winner of the most recent World Cup n Pioneering multi-robot domain, since the early 90s, adversarial, with fully autonomous robots n www.robocup.org 35 CoBot – Collaborave Robot n Tasks n Go to a locaon n Deliver a message n Escort a visitor n Transport object between locaons n Semi-autonomous telepresence n Visitor companion n Deliver mail n Tour guide 18 Autonomy but Limitaons n Perceptual n No universal object recognizer, object finder n No universal speech understanding n Cognive n No complete certainty of posi+on n No certainty about all decisions, what to learns n No stac op+mal knowledge representaon n Actuaon n No ability to manipulate all objects n No constant mobility in all surfaces Symbio+c Autonomy Autonomous Robots Tasked Autonomous Robots Tasked ““To DO MORETo DO MORE”” than than what they actually what they actually ““CAN DOCAN DO””: : Intelligent ProacAve Intelligent ProacAve ““Ask for HelpAsk for Help”” 19 AI of the 21st century n The integration of MANY pieces n Use of unlimited resources n Own built-in and compiled knowledge n Search, infer, process information, remember, adapt n All external digital partners (sensors, robots, cell phones, toasters, vaccuum cleaners, cars, roads) n Digital talk, digital translate n The Web n Mine human-understandable data n People n Ask, learn from, be taught, observe, copy Main Technical Threads n Problem / knowledge representation n Search, optimization, constraint thinking n Reasoning under uncertainty n Modeling and planning n Multiple agents, complex decision making n Hidden information n Learning 20 Part I – No Uncertainty n Symbolic knowledge representation n Pure search n Deterministic planning Part II – Uncertainty n Probabilistic knowledge representation n Planning under uncertainty n Learning n Other agents 21 Schedule n (2) Jan 14 – Jan 16 – Introduction and agents n Knowledge representation – inference n (8) Jan 21 – Feb 13 – Search n Uninformed – SAT, CSP n Informed – Heuristics, MIP n (2) Feb 18 – Feb 20 – Planning n Classical planning and path planning n (3) Feb 25 – Mar 4 – Probabilistic reasoning n Representation n MDPs/POMDPs n (1) Wednesday, March 6 – Midterm Exam Schedule n (2) Mar 18 – Mar 20 – Learning n Supervised and RL n (2) Mar 25 – Mar 27 – Introduction to games n Representation, solving, complexity n (4) Apr 1 – Apr 10 – Multi-robot systems n Adversarial, playbook planning and learning n Learning from demonstration, reuse, correction n (5) Apr 15 – Apr 29 – Game solving n Normal-form games, sequential complete- and incomplete-information games n (1) May 1 – Discussion, open problems 22 .
Load more

Recommended publications
  • The Ethics of Machine Learning
    SINTEZA 2019 INTERNATIONAL SCIENTIFIC CONFERENCE ON INFORMATION TECHNOLOGY AND DATA RELATED RESEARCH DATA SCIENCE & DIGITAL BROADCASTING SYSTEMS THE ETHICS OF MACHINE LEARNING Danica Simić* Abstract: Data science and machine learning are advancing at a fast pace, which is why Nebojša Bačanin Džakula tech industry needs to keep up with their latest trends. This paper illustrates how artificial intelligence and automation in particular can be used to en- Singidunum University, hance our lives, by improving our productivity and assisting us at our work. Belgrade, Serbia However, wrong comprehension and use of the aforementioned techniques could have catastrophic consequences. The paper will introduce readers to the terms of artificial intelligence, data science and machine learning and review one of the most recent language models developed by non-profit organization OpenAI. The review highlights both pros and cons of the language model, predicting measures humanity can take to present artificial intelligence and automatization as models of the future. Keywords: Data Science, Machine Learning, GPT-2, Artificial Intelligence, OpenAI. 1. INTRODUCTION Th e rapid development of technology and artifi cial intelligence has al- lowed us to automate a lot of things on the internet to make our lives eas- ier. With machine learning algorithms, technology can vastly contribute to diff erent disciplines, such as marketing, medicine, sports, astronomy and physics and more. But, what if the advantages of machine learning and artifi cial intelligence in particular also carry some risks which could have severe consequences for humanity. 2. MACHINE LEARNING Defi nition Machine learning is a fi eld of data science which uses algorithms to improve the analytical model building [1] Machine learning uses artifi cial Correspondence: intelligence (AI) to learn from great data sets and identify patterns which Danica Simić could support decision making parties, with minimal human intervention.
    [Show full text]
  • Supervised Machine Learning
    Tikaram Acharya Supervised Machine Learning Accuracy Comparison of Different Algorithms on Sample Data Metropolia University of Applied Sciences Bachelor of Engineering Information Technology Bachelor’s Thesis 11 January 2021 Author Tikaram Acharya Title Supervised Machine Learning: Accuracy Comparison of Differ- ent Algorithms on Sample Data Number of Pages 39 pages Date 11 January 2021 Degree Bachelor of Engineering Degree Programme Information Technology Professional Major Software Engineering Instructors Janne Salonen, Head of School The purpose of the final year project was to discuss the theoretical concepts of the analyt- ical features of supervised machine learning. The concepts were illustrated with sample data selected from the open-source platform. The primary objectives of the thesis include explaining the concept of machine learning, presenting supportive libraries, and demon- strating their application or utilization on real data. Finally, the aim was to use machine learning to determine whether a person would pay back their loan or not. The problem was taken as a classification problem. To accomplish the goals, data was collected from an open data source community called Kaggle. The data was downloaded and processed with Python. The complete practical part or machine learning workflow was done in Python and Python’s libraries such as NumPy, Pandas, Matplotlib, Seaborn, and SciPy. For the given problem on sample data, five predictive models were constructed and tested with machine learning with different techniques and combinations. The results were highly accurately matched with the classification problem after evaluating their f1-score, confu- sion matrix, and Jaccard similarity score. Indeed, it is clear that the project achieved the main defined objectives, and the necessary architecture has been set up to apply various analytical abilities into the project.
    [Show full text]
  • Seinfeld's Recent Comments on Zoom…. Energy
    Stanford Data Science Departmental 2/4/2021 Seminar Biomedical AI: Its Roots, Evolution, and Early Days at Stanford Edward H. Shortliffe, MD, PhD Chair Emeritus and Adjunct Professor Department of Biomedical Informatics Columbia University Department of Biomedical Data Science Seminar Stanford Medicine Stanford, California 2:30pm – 4:50pm – February 4, 2021 Seinfeld’s recent comments on Zoom…. Energy, attitude and personality cannot be “remoted” through even the best fiber optic lines Jerry Seinfeld: So You Think New York Is ‘Dead’ (It’s not.) Aug. 24, 2020 © E.H. Shortliffe 1 Stanford Data Science Departmental 2/4/2021 Seminar Disclosures • No conflicts of interest with content of presentation • Offering a historical perspective on medical AI, with an emphasis on US activities in the early days and specific work I know personally • My research career was intense until I became a medical school dean in 2007 • Current research involvement is largely as a textbook author and two decades as editor-in-chief of a peer- reviewed journal (Journal of Biomedical Informatics [Elsevier]) Goals for Today’s Presentation • Show that today’s state of the art in medical AI is part of a 50-year scientific trajectory that is still evolving • Provide some advice to today’s biomedical AI researchers, drawing on that experience • Summarize where we are today in the evolution, with suggestions for future emphasis More details on slides than I can include in talk itself – Full deck available after the talk © E.H. Shortliffe 2 Stanford Data Science Departmental
    [Show full text]
  • Artificial Intelligence Machine Learning and Deep Learning
    ARTIFICIAL INTELLIGENCE MACHINE LEARNING AND DEEP LEARNING LICENSE, DISCLAIMER OF LIABILITY, AND LIMITED WARRANTY By purchasing or using this book and its companion files (the “Work”), you agree that this license grants permission to use the contents contained herein, but does not give you the right of ownership to any of the textual content in the book or ownership to any of the information, files, or products contained in it. This license does not permit uploading of the Work onto the Internet or on a network (of any kind) without the written consent of the Publisher. Duplication or dissemination of any text, code, simu- lations, images, etc. contained herein is limited to and subject to licensing terms for the respective products, and permission must be obtained from the Publisher or the owner of the content, etc., in order to reproduce or network any portion of the textual material (in any media) that is contained in the Work. MERCURY LEARNING AND INFORMATION (“MLI” or “the Publisher”) and anyone involved in the creation, writing, production, accompanying algorithms, code, or computer programs (“the software”), and any accompanying Web site or software of the Work, cannot and do not warrant the performance or results that might be obtained by using the contents of the Work. The author, developers, and the Publisher have used their best efforts to insure the accuracy and functionality of the textual material and/or pro- grams contained in this package; we, however, make no warranty of any kind, express or implied, regarding the performance of these contents or programs. The Work is sold “as is” without warranty (except for defective materials used in manufacturing the book or due to faulty workmanship).
    [Show full text]
  • The 1970 United States Computer Chess Championship: the Start of the Longest-Running Experiment in Computer Science History
    The 1970 United States Computer Chess Championship: The Start of the Longest-Running Experiment in Computer Science History Jonathan Schaeffer University of Alberta Abstract On August 31, 1970, an experiment began that continues to this day. The first chess tournament for computers was held as part of the Association for Computing Machinery’s (ACM’s) National Conference. The interest generated was tremendous, leading to ACM sponsoring an annual event until 1994. Chess competitions continue to this day, allowing for 50 years of data on the growth of artificial intelligence capabilities in this domain. During this period, program ratings have soared from roughly 1400 in 1970 to over 3500 today. The 1970 event was the first continuous competition in computer science history, and it represents the longest ongoing experiment in computer science history.1 Introduction Creating a program capable of competing with the human world chess champion was one of the first “grand challenge” problems of the fledgling research area of artificial intelligence. The importance of computer chess at the dawn of the computer age is reflected in the many computer science luminaries that contributed to the early advances in the field. This includes Claude Shannon (father of information theory), Alan Turning (creator of the Turing Machine and the Turing Test), Herbert Simon (Nobel Laureate and winner of the prestigious Turing Award), Alan Newell (Turing Award winner), and John McCarthy (Turing Award winner). In the 1950s and the 1960s, progress in developing chess programs was slow. However a breakthrough occurred with the development of MACHACK VI by Massachusetts Institute of Technology (MIT) student Richard Greenblatt.
    [Show full text]
  • Building University-Industry Relations in the Computer Science Department at Stanford University 1963-1972
    Title: Academic Revolution and Regional Innovation: Building University-Industry Relations in the Computer Science Department at Stanford University 1963-1972 Author: Danny Crichton is beginning a Fulbright fellowship as a visiting researcher at KAIST in South Korea. Before beginning this position, he was an undergraduate at Stanford University, where he majored in Mathematical and Computational Sciences with honors in Science, Technology, and Society. His senior thesis was sponsored by a grant from the university, and he won Stanford’s Firestone Award for Undergraduate Research. He can be contacted at [email protected] Subtheme: Silicon Valley – Exploring the ‘works of the engine’, History and Conditions for success Keywords: Stanford, Computer Science, History, Academic Entrepreneurship, University- Industry Relations Copyright of the paper belongs to the author(s). Submission of a paper grants permission to the Triple Helix 9 Scientific Committee to include it in the conference material and to place it on relevant websites. The Scientific Committee may invite accepted papers accepted to be considered for publication in Special Issues of selected journals after the conference. Introduction There remains little consensus in regional studies on the origins of Silicon Valley or other innovation hubs. Different approaches, reflecting the interdisciplinary nature of the field, have examined the issue from institutional, cultural, and network analy- sis perspectives. At the same time, historians of science are beginning to construct a more detailed narrative of the development of computer science in the United States, particularly in the divide between academic theory and industrial practice. Akera has developed a theoretically rigorous synthesis of the field’s rise as part of his study on the pluralism of computation in the Cold War era.
    [Show full text]
  • AI Watch Historical Evolution of Artificial Intelligence
    JRC TECHNICAL REPORTS AI Watch Historical Evolution of Artificial Intelligence Analysis of the three main paradigm shifts in AI EUR 30221 EN This publication is a Technical report by the Joint Research Centre (JRC), the European Commission’s science and knowledge service. It aims to provide evidence-based scientific support to the European policymaking process. The scientific output expressed does not imply a policy position of the European Commission. Neither the European Commission nor any person acting on behalf of the Commission is responsible for the use that might be made of this publication. Contact information Blagoj DELIPETREV European Commission, Joint Research Centre, TP262, Via Fermi, 21027 Ispra (VA), ITALY [email protected] Tel.: + 39 0332 786 352 AI-Watch functional mailbox [email protected] EU Science Hub https://ec.europa.eu/jrc JRC120469 EUR 30221 EN PDF ISBN 978-92-76-18940-4 ISSN 1831-9424 doi:10.2760/801580 Luxembourg: Publications Office of the European Union, 2020 © European Union, 2020 The reuse policy of the European Commission is implemented by the Commission Decision 2011/833/EU of 12 December 2011 on the reuse of Commission documents (OJ L 330, 14.12.2011, p. 39). Except otherwise noted, the reuse of this document is authorised under the Creative Commons Attribution 4.0 International (CC BY 4.0) licence (https://creativecommons.org/licenses/by/4.0/). This means that reuse is allowed provided appropriate credit is given and any changes are indicated. For any use or reproduction of photos or other material that is not owned by the EU, permission must be sought directly from the copyright holders.
    [Show full text]
  • Bounded Rationality and Artificial Intelligence
    No. 2019 - 04 The Game of Go: Bounded Rationality and Artificial Intelligence Cassey Lee ISEAS – Yusof Ishak Institute [email protected] May 2019 Abstract The goal of this essay is to examine the nature and relationship between bounded rationality and artificial intelligence (AI) in the context of recent developments in the application of AI to two-player zero sum games with perfect information such as Go. This is undertaken by examining the evolution of AI programs for playing Go. Given that bounded rationality is inextricably linked to the nature of the problem to be solved, the complexity of Go is examined using cellular automata (CA). Keywords: Bounded Rationality, Artificial Intelligence, Go JEL Classification: B41, C63, C70 The Game of Go: Bounded Rationality and Artificial Intelligence Cassey Lee 1 “The rules of Go are so elegant, organic, and rigorously logical that if intelligent life forms exist elsewhere in the universe, they almost certainly play Go.” - Edward Lasker “We have never actually seen a real alien, of course. But what might they look like? My research suggests that the most intelligent aliens will actually be some form of artificial intelligence (or AI).” - Susan Schneider 1. Introduction The concepts of bounded rationality and artificial intelligence (AI) do not often appear simultaneously in research articles. Economists discussing bounded rationality rarely mention AI. This is reciprocated by computer scientists writing on AI. This is surprising because Herbert Simon was a pioneer in both bounded rationality and AI. For his contributions to these areas, he is the only person who has received the highest accolades in both economics (the Nobel Prize in 1978) and computer science (the Turing Award in 1975).2 Simon undertook research on both bounded rationality and AI simultaneously in the 1950s and these interests persisted throughout his research career.
    [Show full text]
  • TEX: a Branch in Desktop Publishing Evolution, Part 1
    IEEE ANNALS OF THE HISTORY OF COMPUTING THEME ARTICLE: HISTORY OF DESKTOP PUBLISHING, PART 1 TEX: A Branch in Desktop Publishing Evolution, Part 1 Donald Knuth began the development of TEX in 1977 and had an initial version running in 1978, with the Barbara Beeton American Mathematical Society aim of typesetting mathematical documents with the Karl Berry highest quality, and with archival reproducibility far TEX Users Group into the future. Its usage spread, and today the TEX David Walden system remains in use by a vibrant user community. However, the world of TEX has always been somewhat out of sync with the commercial desktop publishing systems that began to come out in the mid-1980s and are now ubiquitous in the worlds of publishing and printing. TEX is a typesetting system that was developed by Donald Knuth, starting in 1977, to “do beau- tiful typesetting,” particularly of mathematics. At the time, PostScript and PDF did not yet exist, people were in the early stages of mathematically defining digital fonts, and WYSIWYG user interfaces were just being conceived. What we today think of as desktop publishing systems (for example, WYSIWYG, PostScript output) were half a dozen years away, and the computers of the time were slow and small compared with today. Thus, TEX was implemented using markup of plain text files, as was typical with prior systems such as RUNOFF, to instruct TEX about docu- ment formatting. TEX also had its own way of outputting page descriptions for print as well as its own companion system, Metafont, for designing and formatting type.
    [Show full text]
  • Collective Test
    COLLECTIVE TEST: CAN GENERATIVE TOOLS WORKING AUTONOMOUSLY CREATE NEW CULTURES IN THE LONG RUN? Architectural history thesis spring 2021 spring 2021 AR2A011 ARCHITECTURAL HISTORY THESIS Students: Haruka Maeda Leticija Aleksandra Petrova Tutor: Dan C. Baciu COLLECTIVE TEST: CAN GENERATIVE TOOLS WORKING AUTONOMOUSLY CREATE NEW CULTURES IN THE LONG RUN? Architectural history thesis spring 2021 Index Abstract 00 Introduction 01 -Brief history of AI and generative tools -Current testing of generative tools& limitation -Neccessity for an additional test, breaking from limitations -Methodology -The necessary steps -The choice of the 3 types of case studies -The specific projects & the tools used Generative tools: 02 created Illusion and human trace -Introduction for case studies -Approach of a human designer -Case study 1: AI & Text Illusion and the turing test The process and human trace -Case study 2: AI & Art Illusion and the turing test The process and human trace -Case study 3: AI & Architecture Illusion and the turing test The process and human trace Summary of the created illusions and human inputs Can generative tools 03 autonomously create human cultures on a larger scale? Conclusion: Collective test 04 Bibliography 05 Appendix 06 Abstract The term AI has entered the creative industry in the last decades. Computer based design is now ominiprescent in everyday life. Will AI take over the creative industry? Can they autonomously be creative? These are some of the headlines propagated by the mainstream media. This paper will unveil the human labor behind these computer generated designs by analyzing how they were publicized and how they were actually developed. Through the related case studies, it is possible to identify how much of the AI work is still human originated, and how a total automation is still not attained.
    [Show full text]
  • John Von Neumann
    John von Neumann ! Hungarian mathematician, computer scientist, cyberneticist, all-around genius ! Worked on atomic bomb project in WW II ! Invented game theory and developed theory of self-replicating automata ! Originated key concept of stored-program computer in 1945 ! Program instructions = data ! Easily reprogrammable ! Von Neumann architecture is still the universal standard EDVAC ! Electronic Discrete Variable Automatic Computer ! Designed by Mauchly, Eckert, and Von Neumann ! Stored-program design ! Used binary instead of decimal to represent information ! Version called UNIVAC I was the first commercially available computer system ! Sold to the U.S. Census Bureau in 1951 First Generation Computers ! Mid 1940s to late 1950s ! Stored-program design with ~ 1000 words of RAM ! Used vacuum tubes, but required less space than ENIAC ! Punched cards for input and output ! Vacuum tube or magnetic core memory for data storage ! Programmed directly in binary machine language ! Included EDVAC and UNIVAC First Generation Computers Transistors ! Invented at Bell Labs in 1947 by William Shockley, John Bardeen, and Walter Brattain ! Generated far less heat than vacuum tubes ! Required far less power ! Much faster, smaller, cheaper, and more reliable Transistors ! Incorporated into Second Generation computers in the late 1950s and early 1960s Origins of AI ! 1956 Summer Dartmouth Conference – Organized by John McCarthy, who coined the term “artificial intelligence” – Participants: ! John McCarthy ! Marvin Minsky ! Claude Shannon Minsky Shannon
    [Show full text]
  • In Memoriam-Arthur Samuel: Pioneer in Machine Learning
    Memoriam AI Magazine Volume 11 Number 3 (1990) (© AAAI) In Memoriam Arthur Samuel: Pioneer in Samuel was a modest man, and the strate the Machine Learning importance of his work was widely power of recognized only after his retirement electronic Arthur Samuel (1901–1990) was a from IBM in 1966, in part because he comput- pioneer of artificial intelligence didn’t relish the politics that were ers. He research. From 1949 through the late required to have his research more didn’t fin- 1960s, he did the best work in mak- vigorously followed up on. He was ish the ing computers learn from their expe- also realistic about the large difference program while he was at the universi- rience. His vehicle for this work was between what had been accomplished ty of Illinois, perhaps because the the game of checkers. in understanding intellectual mecha- computer wasn’t finished in time. Programs for playing games often nisms and what would be required to In 1949, Samuel joined IBM’s fill the role in artificial intelligence reach human-level intelligence. Poughkeepsie Laboratory, where he research that the fruit fly Drosophila Samuel’s papers on machine learn- worked on IBM’s first stored program plays in genetics. Drosophilae are ing are still worth studying. Possess- computer, the 701. This computer’s convenient for genetics because they ing great creativity and essentially memory was based on Williams tubes, breed fast and are cheap to keep, and working alone, doing his own pro- which stored bits as charged spots on games are convenient for artificial gramming, he invented several semi- the screen of a cathode ray tube.
    [Show full text]