Inventing Things That Matter Through AI Innovation
•Télécharger en tant que PPTX, PDF•
0 j'aime•340 vues
This document discusses IBM's global research capabilities and focuses on inventing things that matter to the world. It provides an overview of IBM's research areas such as healthcare, government, financial services, industry cloud, IoT, blockchain, cognitive robotics, and more. It highlights IBM's leadership in patents and the deep skills of its scientists. It also discusses IBM's investments in quantum computing, AI, healthcare/life sciences, and more. The document emphasizes that foundational breakthroughs have led to recognition like Nobel Prizes and that IBM outpaces competitors in patents. It aims to convey that IBM researchers invent things that can make a difference globally.
Recommandé
Contenu connexe
Tendances
Tendances (20)
Similaire à Inventing Things That Matter Through AI Innovation
Similaire à Inventing Things That Matter Through AI Innovation (20)
Plus de ISSIP
Plus de ISSIP (20)
Dernier
Dernier (20)
Inventing Things That Matter Through AI Innovation
- 1. Inventing Things That Matter To The World Jim Spohrer Director, IBM Cognitive OpenTech July 24, 2020 Presentations on line at: http://slideshare.net/spohrer
- 2. IBM’s global research capability Healthcare Government Financial Services Healthcare Industry Cloud IoT Blockchain Cognitive Robotics Financial Services Accessibility Core AI Capabilities Cloud & IoT Industry Solutions Blockchain Cognitive Fashion Education & Skilling Cognitive Financial Services Cognitive Healthcare IoT & Mobile Security Security Analytics Nanotechnology Exascale Cognitive IoT AI for Healthcare Edge ComputingBig Data & Cognitive Cloud Healthcare / Life Sciences Quantum Computing POWER Mobile Aging Cognitive Oil & Gas Insurance Analytics Industry Cloud Big Data Nanomaterials Neurosynaptics 3,000+ researchers Australia Tokyo China Almaden Haifa Zurich Africa Ireland Brazil Watson Austin India © 2019 IBM Corporation 2
- 3. Foundational breakthroughs have made us famous 6 Nobel Laureates 10 National Medals of Technology 6 Turing Awards 5 National Medals of Science © 2019 IBM Corporation 3
- 4. 2018 patents: IBM vs. competition © 2019 IBM Corporation 4 0 9500 9100 IBM 2353 Microsoft Amazon 2070 Google 2035 GE 954 HP 703 Oracle 640 Facebook 200 Accenture 1597 148 Symantec NumberofPatents 2018 patent data sourced from IFI Claims Patent Services
- 5. Our scientists have deep skills in a range of core disciplines © 2019 IBM Corporation 5 Behavioral Science Biology Chemistry Computer Science Electrical Engineering Materials Science Mathematics Physics
- 6. IBM Quantum 7/25/2020 (c) IBM MAP COG .| 6
- 7. “Instrumenting trust into data sets and machine learning models will accelerate the adoption of AI and engender increased confidence in these general-purpose technologies.” Aleksandra Mojsilovic IBM Fellow Head of Foundations of Trusted AI Building trust into AI <https://www.ibm.org/responsibility/2018/trusted-ai#story> (© Copyright IBM Corporation 1994, 2019).
- 8. “If we fail to make ethical and inclusive artificial intelligence we risk losing gains made in civil rights and gender equity under the guise of machine neutrality.” Joy Buolamwini Gender Shades MIT Media Lab Joy Buolamwini – MIT Media Lab <https://www.media.mit.edu/people/joyab/overview/> (CC BY 4.0).
- 9. 97/25/2020
- 10. 107/25/2020
- 11. Is it fair? Is it easy to understand? Is it accountable? So what does it take to trust a decision made by a machine? (Other than that it is 99% accurate)? Did anyone tamper with it? #21, #32, #93 #21, #32, #93
- 12. Is it fair? Is it easy to understand? Is it accountable? Did anyone tamper with it? FAIRNESS EXPLAINABILITYROBUSTNESS LINEAGE Our vision for Trusted AI Pillars of trust, woven into the lifecycle of an AI application
- 13. 137/25/2020 LightGBM GPy Supported ML/DL frameworks: 30+ SOTA attacks (evasion, poisoning, extraction, inference) 25+ baseline defenses Modules for detection, metrics and certification
- 14. Adversarial Threats 14 › Adversarial threats against machine learning models and applications have a wide variety of attack vectors. › Evasion: Modifying input to influence model › Poisoning: Modify training data to add backdoor › Extraction: Steal a proprietary model › Inference: Learn information on private data
- 15. Real Adversarial Threats › Evasion. › Imperceptible modifications to medical images to influence classification. › Poisoning. › Imperceptible patterns in training data create backdoors that control models. › Extraction. › Theft of proprietary models through model queries. › Inference. › Derive properties of the model’s training data up to identifying single data entries. 15
- 16. Adversarial Threat Combinations › Combinations of adversarial threats become more effective than their sum. › Extraction attacks enable stronger white- box evasion attacks › Extraction attacks steal models that could leak more private information in inference attacks Evasion Poisoning Inference Extraction 16
- 17. Adversarial Robustness Toolbox (ART) › ART is a Python library for machine learning security. › github.com/IBM/adversarial-robustness- toolbox › 1500+ Stars (~500 in last 6months) › providing tools to developers and researcher › Evaluating, Defending, Certifying and Verifying of machine learning models and applications › All Tasks: Classification, Object Detection, Generation, Encoding, Certification, etc. › All Frameworks: TensorFlow, Keras, PyTorch, MXNet, scikit-learn, XGBoost, LightGBM, CatBoost, GPy › All Data: images, tables, audio, video, etc. › Contributions and feedback are very welcome! 17 LightGBM GPy
- 18. The Tools of ART › ART 1.3 - art.metrics - Methods to quantify robustness - art.estimators - Abstractions for models Evasion Poisoning Extraction Inference art.attacks examples • 21 (+8) • White-box (e.g. FGSM, PGD, Carlini&Wagner, …) • Black-box (HopSkipJump, Boundary, ZOO, …) • 3 (+1) • Backdoor, Feature Collision, SVM, … • 3 • FunctionallyEquivalent, KnockOffNets, CopyCat, … • 4 (+4) • Model Inversion (MIFace, …) • Attribute Inference art.defences examples • 15 (+4) • Adversarial Training (Madry, Fast is Better than Free, …) • Preprocessing • Transformer • Detection • 4 (+1) • Detection (Activation, Provenance, RONI, Spectral Signature, …) • 6 • Postprocessing (Reverse Sigmoid, …) • DiffPrivLib 18
- 19. New Attacks and Defenses –Dpatch (Liu et al., 2019) • Adversarial patches for object detectors –Shadow Attack (Ghiasi et al., 2020) • Breaking/spoofing robustness certificates –Feature Adversaries (Sabour et al., 2016) • Imitates feature representation of benign samples – Frame Saliency Attack (Inkawhich et al., 2018) • Attack on action recognition systems –Wasserstein Attack (Wong et al., 2019) • Large but naturally looking perturbations –Auto Attack (Croce and Hein, 2020) • Multiple white- and black-box attacks optimized for achieving state-of-the-art robustness evaluation performance of leading experts completely automated – Auto-PGD (Croce and Hein, 2020) • multiple attack losses and automated learning rate adjustment – Square Attack (Croce and Hein, 2019) • very efficient black-box attack based on random search –DefenseGAN (Samangouei et al., 2018), InvGAN (Lin et al., 2019) • Defense based on Generative Adversarial Networks (GAN) –MP3 compression, resampling (Carlini et al., 2018) –MPEG compression, frame-wise JPEG and spatial smoothing –Fast is Better than Free (Wong et al., 2019) • Fastest adversarial training protocol 19
- 20. 20 –Application of ART to Speech classification –Dataset: Audio-MNIST, spoken digits [0-9] with multiple speakers –Baseline for evaluating defenses against evasion on audio data –Starting point for ART towards speech recognition and sequence-to- sequence models –https://github.com/IBM/adversarial-robustness- toolbox/blob/master/notebooks//adversarial_audio_examples.ipynb ART Audio Example – Speech Classification
- 21. AI Fairness 360 21 aif360.mybluemix.net Most comprehensive open source toolkit for detecting & mitigating bias in ML models: • 70+ fairness metrics • 10 bias mitigators • Interactive demo illustrating 5 bias metrics and 4 bias mitigators • extensive industry tutorials and notebooks
- 22. AI Fairness 360 aif360.mybluemix.net Designed to translate new research from the lab to industry practitioners: tutorials, education, glossary, resources.
- 23. Three categories of bias mitigation algorithms Pre-processing algorithm – a bias mitigation algorithm that is applied to training data In-processing algorithm – a bias mitigation algorithm that is applied to a model during its training Post-processing algorithm – a bias mitigation algorithm that is applied to predicted labels The choice among algorithm categories can partially be made based on the user persona’s ability to intervene at different parts of a machine learning pipeline. If the user is allowed to modify the training data, then pre-processing can be used. If the user is allowed to change the learning algorithm, then in-processing can be used. If the user can only treat the learned model as a black box without any ability to modify the training data or learning algorithm, then only post-processing can be used.
- 24. AI Explainability 360 The most comprehensive open source toolkit for explaining ML models and data: • 10 innovated algorithms to explain data and AI models + 2 metrics • An interactive demo that provides a gentle introduction through a credit scoring application • 13 tutorial notebooks covering use cases in finance, healthcare, lifestyle, retention, etc. • documentation that guides the practitioner on choosing an appropriate explanation method. One Explanation Does Not Fit All: A Toolkit and Taxonomy of AI Explainability Techniques by Arya et al. https://arxiv.org/abs/1909.03012 http://aix360.mybluemix.net/
- 25. Designed to translate new research from the lab to industry practitioners: tutorials, education, glossary, resources. AI Explainability 360 aix360.mybluemix.net
- 26. One Explanation Does Not Fit All: A Toolkit and Taxonomy of AI Explainability Techniques by Arya et al. https://arxiv.org/abs/1909.03012
- 27. Meaningful Explanations Depend on the Explanation Consumer Must match the complexity capability of the consumer Must match the domain knowledge of the consumer Regulatory Bodies • Who: EU (GDPR), NYC Council, US Gov’t, etc • Why: ensure fairness for constituents Affected Users • Who: Patients, accused, loan applicants, teachers • Why: understanding of factors AI System builders, stakeholders • Who: data scientists, developers, prod mgrs • Why: ensure/improve performance End Users • Who: Physicians, judges, loan officers, teacher evaluators • Why: trust/confidence, insights(?) “We couldn’t explain the model to them because they didn’t have the training in machine learning.” Nautilus, Sept 2016
- 28. Trusted AI Committee Updates 2 Technical Working Group: Topics presented and discussed AI Fairness 360: MLOps Section created Kubeflow Pipelines Integration https://github.com/IBM/AIF360/tree/master/mlops/kubeflow Apache Nifi Integration: https://github.com/IBM/AIF360/tree/master/mlops/nifi SKLearn API support for AIF360https://github.com/IBM/AIF360/tree/master/aif360/skle arn The AI Fairness 360 R package https://github.com/IBM/AIF360/tree/master/aif360/aif360-r AI Factsheets https://www.ibm.com/blogs/research/2018/08/factsheets-ai/ KFServing Integration http://bit.ly/kubeflow-trusted-ai KPMG: Trusted AI in field https://lists.lfai.foundation/g/trustedai- committee/files/KPMG%20AI%20in%20Control%20May14202 0.pdf Adversarial Robustness 360: MLOps Section created Kubeflow Integration https://github.com/IBM/adversarial-robustness- toolbox/tree/master/mlops Principles Working Group: Materials submitted to Trusted AI Committee from --- Orange (document draft 4.1 dated 12 August 2019) --- AT&T (Working Draft Artificial Intelligence Operating Principles Under Development version dated Nov 7, 2019 ) --- TenCent (Jeff Cao - Tencent Research Institute - slides) --- IBM ( https://wiki.lfai.foundation/display/DL/Trusted+AI+Committee#TrustedAICom mittee-Assets) --- Institute of Ethical AI https://github.com/EthicalML/awesome- artificial-intelligence-guidelines https://ethical.institute/ Initial PWG Trusted AI documents produced. Tencent, Orange and IBM have signed off. Seeking an AT&T signoff Orange Responsible AI Presentation https://lists.lfai.foundation/g/trustedai- committee/files/2020_Responsable_AI_Orange_LFAI.pdf Upcoming Work:: Finalize the LFAI Principle document outlining, among other things- -- scope - who creates AI - humans or machines/AIs -- bias in definitions -- consider how to organize principles - perhaps in a hierarchy - particular contribution of document: linking principles to implementation; incorporating global principles and thinking, linking to business throug use cases -- more on correlation to business e.g., how explainability links to business (edited)
- 29. How will COVID-19 effect the need for and use of robots in a service world with less physical contact? Will robots improve or harm livelihoods/jobs? Robots Rule Retail? Taking away jobs Telepresence Robot World? Adding more jobs Robots at Home? Reducing need to have a job You will be assigned to a small team to discuss. Please have a team member to take notes of the most important insights and/or questions that emerge from your discussion. Your notes will be crucial for us to create a conference report, send to contact@creatingvalueconf.com What is most probable to happen? What is desirable? Spohrer
- 30. IBM-MIT $240M over 10 year AI mission 7/25/2020 (c) IBM 2017, Cognitive Opentech Group 30
- 31. Narrow AI Emerging Broad AI Disruptive and Pervasive General AI Revolutionary ▼ We are here 2050 and beyond 31IBM Research AI © 2018 IBM Corporation The evolution of AI Borrowed from David Cox, IBM-MIT Lead
- 32. 32September 2018 / © 2018 IBM Corporation Petaflops = 1,000,000,000,000,000 or a million billion = 10 ** 15 Megaflops = 1,000,000 = million = 10 ** 6 Gigaflops = 1,000,000,000 = billion = 10 ** 9 Larges Super Computer in the World, = 13 MegaWatts of Power (HOT!)
- 33. 33September 2018 / © 2018 IBM Corporation Exascale = 1,000,000,000,000,000,000 or a billion billion = 10 ** 18 Megaflops = 1,000,000 = million = 10 ** 6 Gigaflops = 1,000,000,000 = billion = 10 ** 9 Human Brain = 20 Watts (COOL!)
- 34. Questions • What is the timeline for solving AI and IA? • TBD: When can a CEO buy AI capability <X> for price <Y>? • Who are the leaders driving AI progress? • What will the biggest benefits from AI be? • What are the biggest risks associated with AI, and are they real? • What other technologies may have a bigger impact than AI? • What are the implications for stakeholders? • How should we prepare to get the benefits and avoid the risks? 7/25/2020 (c) IBM 2017, Cognitive Opentech Group 34
- 35. Timeline: Short History 7/25/2020 © IBM Cognitive Opentech Group (COG) 35 Dota 2 “Deep Learning” for “AI Pattern Recognition” depends on massive amounts of “labeled data” and computing power available since ~2012; Labeled data is simply input and output pairs, such as a sound and word, or image and word, or English sentence and French sentence, or road scene and car control settings – labeled data means having both input and output data in massive quantities. For example, 100K images of skin, half with skin cancer and half without to learn to recognize presence of skin cancer.
- 36. Timeline: Every 20 years, compute costs are down by 1000x • Cost of Digital Workers • Moore’s Law can be thought of as lowering costs by a factor of a… • Thousand times lower in 20 years • Million times lower in 40 years • Billion times lower in 60 years • Smarter Tools (Terascale) • Terascale (2017) = $3K • Terascale (2020) = ~$1K • Narrow Worker (Petascale) • Recognition (Fast) • Petascale (2040) = ~$1K • Broad Worker (Exascale) • Reasoning (Slow) • Exascale (2060) = ~$1K 367/25/2020 (c) IBM 2017, Cognitive Opentech Group 2080204020001960 $1K $1M $1B $1T 206020201980 +/- 10 years $1 Person Average Annual Salary (Living Income) Super Computer Cost Mainframe Cost Smartphone Cost T P E T P E AI Progress on Open Leaderboards Benchmark Roadmap to solve AI/IA
- 37. Timeline: GDP/Employee 7/25/2020 (c) IBM 2017, Cognitive Opentech Group 37 (Source) Lower compute costs translate into increasing productivity and GDP/employees for nations Increasing productivity and GDP/employees should translate into wealthier citizens AI Progress on Open Leaderboards Benchmark Roadmap to solve AI/IA
- 38. Timeline: Leaderboards FrameworkAI Progress on Open Leaderboards - Benchmark Roadmap Perceive World Develop Cognition Build Relationships Fill Roles Pattern recognition Video understanding Memory Reasoning Social interactions Fluent conversation Assistant & Collaborator Coach & Mediator Speech Actions Declarative Deduction Scripts Speech Acts Tasks Institutions Chime Thumos SQuAD SAT ROC Story ConvAI Images Context Episodic Induction Plans Intentions Summarization Values ImageNet VQA DSTC RALI General-AI Translation Narration Dynamic Abductive Goals Cultures Debate Negotiation WMT DeepVideo Alexa Prize ICCMA AT Learning from Labeled Training Data and Searching (Optimization) Learning by Watching and Reading (Education) Learning by Doing and being Responsible (Exploration) 2018 2021 2024 2027 2030 2033 2036 2039 7/25/2020 (c) IBM 2017, Cognitive Opentech Group 38 Which experts would be really surprised if it takes less time… and which experts really surprised if it takes longer? Approx. Year Human Level -> +3 See: https://paperswithcode.com/sota
- 39. Who is winning 7/25/2020 (c) IBM 2017, Cognitive Opentech Group 39 https://www.technologyreview.com/s/608112/who-is-winning-the-ai-race/
- 40. Robots by Country • Industrial robots per 10,000 people by country 7/25/2020 IBM #OpenTechAI 40 34
- 41. Economic Growth Rates 2035: AI Projected Impact 7/25/2020 (c) IBM MAP COG .| 41
- 42. AI Benefits • Access to expertise • “Insanely great” labor productivity for trusted service providers • Digital workers for healthcare, education, finance, etc. • Better choices • ”Insanely great” collaborations with others on what matters most • AI for IA = Augmented Intelligence and higher value co-creation interactions 7/25/2020 (c) IBM 2017, Cognitive Opentech Group 42
- 43. AI Risks • Job Loss • Shorter term bigger risk = de-skilling • Super-intelligence • Shorter term bigger risk = bad actors 7/25/2020 (c) IBM 2017, Cognitive Opentech Group 43
- 44. Other Technologies: Bigger impact? Yes. • Augmented Reality (AR)/ Virtual Reality (VR) • Game worlds grow-up • Blockchain/ Security Systems • Trust and security immutable • Advanced Materials/ Energy Systems • Manufacturing as cheap, local recycling service (utility fog, artificial leaf, etc.) 7/25/2020 (c) IBM 2017, Cognitive Opentech Group 44
- 45. “The best way to predict the future is to inspire the next generation of students to build it better” Digital Natives Transportation Water Manufacturing Energy Construction ICT Retail Finance Healthcare Education Government
- 46. Artificial Leaf • Daniel Nocera, a professor of energy science at Harvard who pioneered the use of artificial photosynthesis, says that he and his colleague Pamela Silver have devised a system that completes the process of making liquid fuel from sunlight, carbon dioxide, and water. And they’ve done it at an efficiency of 10 percent, using pure carbon dioxide—in other words, one-tenth of the energy in sunlight is captured and turned into fuel. That is much higher than natural photosynthesis, which converts about 1 percent of solar energy into the carbohydrates used by plants, and it could be a milestone in the shift away from fossil fuels. The new system is described in a new paper in Science. 7/25/2020 IBM Code #OpenTechAI 46
- 47. Food from Air • Although the technology is in its infancy, researchers hope the "protein reactor" could become a household item. • Juha-Pekka Pitkänen, a scientist at VTT, said: "In practice, all the raw materials are available from the air. In the future, the technology can be transported to, for instance, deserts and other areas facing famine. • "One possible alternative is a home reactor, a type of domestic appliance that the consumer can use to produce the needed protein." • According to the researchers, the process of creating food from electricity can be nearly 10 times as energy efficient as photosynthesis, the process used by plants. 7/25/2020 IBM Code #OpenTechAI 47
- 48. Exoskeletons for Elderly • A walker is a “very cost-effective” solution for people with limited mobility, but “it completely disempowers, removes dignity, removes freedom, and causes a whole host of other psychological problems,” SRI Ventures president Manish Kothari says. “Superflex’s goal is to remove all of those areas that cause psychological-type encumbrances and, ultimately, redignify the individual." 7/25/2020 IBM Code #OpenTechAI 48
- 49. 7/25/2020 49 1955 1975 1995 2015 2035 2055 Better Building Blocks
- 50. 7/25/2020 © IBM 2015, IBM Upward University Programs Worldwide accelerating regional development 50 I have… Have you noticed how the building blocks just keep getting better?
- 51. Learning to program: My first program 7/25/2020 © IBM 2015, IBM Upward University Programs Worldwide accelerating regional development 51 Early Computer Science Class: Watson Center at Columbia 1945 Jim Spohrer’s First Program 1972
- 52. 7/25/2020 © IBM UPWard 2016 52 Fast Forward 2016: Consider this…
- 53. Microsoft CaptionBot June 19, 2016 7/25/2020 © IBM UPWard 2016 53
- 54. Microsoft CaptionBot June 20, 2016 7/25/2020 © IBM UPWard 2016 54
- 55. IBM Image Tagging 7/25/2020 © IBM UPWard 2016 55
- 56. Today: November 10, 2017 7/25/2020 © IBM DBG COG 2017 56 IBM
- 57. 10 million minutes of experience 7/25/2020 Understanding Cognitive Systems 57
- 58. 2 million minutes of experience 7/25/2020 Understanding Cognitive Systems 58
- 59. 7/25/2020 (c) IBM MAP COG .| 59
- 60. 7/25/2020 (c) IBM MAP COG .| 60
- 61. Karpathy and Li, 2015
- 62. “Teddy Bear” Meret Oppenheim, Le Déjeuner en fourrure
- 63. 7/25/2020 © IBM 2015, IBM Upward University Programs Worldwide accelerating regional development 63
- 64. Icons of AI Progress • 1956: Dartmouth Conference organized by: • John McCarthy (Dartmouth, later Stanford) • Marvin Minsky (MIT) • and two senior scientists: • Claude Shannon (Bell Labs) • Nathan Rochester (IBM) • 1997: Deep Blue (IBM) - Chess • 2011: Watson Jeopardy! (IBM) • 2016: AlphaGo (Google DeepMinds) 7/25/2020 (c) IBM 2017, Cognitive Opentech Group 64
- 65. AI at IBM: Past (Nathan Rochester) 7/25/2020 (c) IBM MAP COG .| 65
- 66. Smartphones pass entrance exams? When? 7/25/2020 (c) IBM 2017, Cognitive Opentech Group 66 … when will your smartphone be able to take and pass any online course? And then be your coach, so you can pass too?
- 67. 7/25/2020 © IBM 2015, IBM Upward University Programs Worldwide accelerating regional development 67 Cognitive Mediators for all people in all roles
- 68. Occupations = Many Tasks 7/25/2020 © IBM 2015, IBM Upward University Programs Worldwide accelerating regional development 68
- 69. Watson Discovery Advisor 7/25/2020 © IBM 2015, IBM Upward University Programs Worldwide accelerating regional development 69 Simonite, T. 2014. Software Mines Science Papers to Make New Discoveries. MIT. November 25, 2014. URL: http://m.technologyreview.com/news/520461/software-mines-science-papers-to-make-new-discoveries/
- 70. Step Comment GitHub Get an account and read the guide MAX CODAIT’s Model Asset Exchange Learn 3 R's - Read, Redo, Report Read (Medium/arXiv), Redo (GitHub), Report (Jupyter Notebook) PapersWithCode Stay on top of recent advances; Do 3 R’s. Kaggle Compete in a Kaggle competition Leaderboards Compete to advance AI progress Linux Foundation AI Help end-to-end open source industry AI & Data infrastructure Mozilla Common Voice Donate your speech; Label and verify data; Recruit others. Figure Eight Generate a set of labeled data (also Mechanical Turk) Design New Challenges Build for Call for Code/Code and Response; Build your AI Helper; Build test-taker, that can switch to tutor-mode; Etc. Open Source Guide Establish open source culture in your organization 7/25/2020 IBM Code #OpenTechAI 70
- 71. 7/25/2020 (c) IBM MAP COG .| 71 Microsoft acquiring GitHub $7.5B 2018 John Marks on Open Source Models will run the world Why SW is eating the world
- 72. Sweden 7/25/2020 (c) IBM MAP COG .| 72
- 73. IBM AI Explain a transaction Deployment: Claim Approval Model name: Claim Model AI Fairness 360 toolkit Trust and transparency integral to AI on the IBM Cloud Explainability, fairness, lineage are critical principles of trusted AI Open source toolkit to check for unwanted bias in datasets and machine learning models © 2019 IBM Corporation 73 DENIED APPROVEDCONFIDENCE 90% 10% POLICY HOLDER AGE: 18 RESPONSIBLE PARTY: Self CAR BRAND: Oldsmobile Cutlass POLICE REPORT: Yes CAR VALUE: $20,000 POLICY AGE: 5 Years 65% 17% 23% 13% 13% 5% Factors contributing to a DENIED confidence level Factors contributing to an APPROVED confidence level
Notes de l'éditeur
- Susan’s references Internal https://w3.ibm.com/w3publisher/ibm-developer-marketing/code-and-response https://w3.ibm.com/w3publisher/cfc-finalists External https://developer.ibm.com/callforcode/ https://developer.ibm.com/code-and-response/
- The activities of our global labs reflect both local market needs and the deep technical expertise found in these key regions of the world. For example, our Dublin, Ireland lab collaborates with academic and industrial partners on research programs established by the European Union as well as collaborative projects developed side by side with University College Dublin scientists. In Africa we are developing commercially-viable solutions to transform lives and spark new business opportunities in key areas such as agriculture, healthcare, financial inclusion, education, energy, blockchain and cloud computing. In Brazil we are working on natural resources solutions, collaborating with leading energy companies to enhance energy exploration and develop new transformative technologies. In India we are bringing cognitive technologies to bear in solving problems within several industry domains such as financial services, education, human resources and fashion.
- We have been a home for creative scientific minds since 1945. Many of our researchers have become famous for their inventions and discoveries, earning an array of the most prestigious awards in science. These include six Nobel Prizes, 10 U.S. national medals of technology, five U.S. national medals of science and six Turing Awards, along with many other distinguished honors.
- IBM inventors received a record 9,100 patents in 2018, including more than 3,000 patents related to work in artificial intelligence, cloud and quantum computing. This marks the company’s 26th consecutive year of U.S. patent leadership, having crossing the 100,000-patent milestone the year before, in 2017. Our work in many of these began long before there were practical enterprise uses for the technology, and that spirit of research for the sake of discovery is what has propelled us to lead the field in patent grants for more than a quarter of a century. This year we led the industry with 1,600 AI-related patents, a number of which stemmed from our work around the language and machine learning techniques that power Project Debater. These innovations will help transform how we interact with AI and enhance our ability to use it as a tool to gain more meaningful insights. We’ve also explored how technology can improve our health and promote safety, patenting methods for smart wearables to communicate with electronic components embedded in prostheses – from hearing aids to prosthetic arms – potentially helping them quickly adapt to better suit the wearer’s needs. Another significant patent this year was for foundational blockchain technology that encrypts transactions as they are recorded and allows users to share transactions with an encryption key. Different keys can be provided to different user sets depending upon the type of data a user is authorized to access. For example, one key could be provided for access to medical transactions, and another key could be provided for financial transactions. IBM inventors also patented significant inventions in quantum computing, including a new way of miniaturizing components that improve the performance of quantum computers. This may allow the integration of discrete elements into a single quantum computing chip. The outstanding innovations from IBM researchers around the world who contributed to this record patent year are truly leading the way for the future – helping to make business and society smarter, safer and more sustainable.
- IBM Research collaborates across diverse disciplines, closely aligned with peers in the core fields of academia, to address some of the world’s most complex problems and promising opportunities. We believe that profound breakthroughs will come when businesses, governments, academic institutions and others work together to tap into diverse points of view and expertise. Collectively, we’re working to understand how systems are interconnected and the role technology plays within them.
- … and as they did we realized that they can have profound implications on our society. Once upon a time we used to build algorithms for accuracy, but that is changing. There is a realization that if they are going to be governing important aspects of our lives, they need to be more than just accurate – we need to trust them. So what does it take to trust and algorithmic decision?
- … and as they did we realized that they can have profound implications on our society. Once upon a time we used to build algorithms for accuracy, but that is changing. There is a realization that if they are going to be governing important aspects of our lives, they need to be more than just accurate – we need to trust them. So what does it take to trust and algorithmic decision?
- How will COVID-19 effect the need for and use of robots in a service world with less physical contact? Will robots improve or harm livelihoods/jobs? Robots Rule Retail? Taking away jobs: https://www.forbes.com/sites/blakemorgan/2020/05/13/the-3-best-in-store-robots-and-why-they-work/#414ae0ca37b2 Employers prefer to deploy robots rather than people for routine jobs in retail, hospitality, travel, education, healthcare, government, manufacturing, farming, and then all industries. Telepresence Robot World? Adding more jobs: https://www.zdnet.com/article/best-telepresence-robots/ Show up when needed and where needed to help in situations of various skill levels – more job opportunities. Robots reduce need for stable identity of service provider, and thus increase the number of job roles that are cost effective for entrepreneurs to provide. Increase the staff I can afford to hire for a task. Better matching people to task. Robots at Home? Reduce need to have a job: https://home-automations.net/top-10-personal-robots-2020/ Don’t need to pay for gardner, chef, repairs, etc. – what if they barter with neighbor robots? Robots lower cost of living, lower need for high-pay jobs. ==== In the news…. Retail - How lockdown is changing shopping for good https://www.technologyreview.com/2020/05/25/1002168/retail-robots-save-local-store-business-lockdown-pandemic-coronavirus-economic-crisis/ https://medium.com/@jgwoodland/five-ways-in-which-artificial-intelligence-is-accelerating-the-development-of-new-medicines-5a1d31c2dfbb Finance - Millennials Prefer Robot Bankers to Humans, Nordic Data Show https://www.bloomberg.com/news/articles/2020-05-24/banker-bots-rake-in-nordic-wealth-business-and-reshape-finance Customer Service - Hello and welcome: robot waiters to the rescue amid virus https://www.washingtonpost.com/lifestyle/food/hello-and-welcome-robot-waiters-to-the-rescue-amid-virus/2020/05/29/59de1e06-a175-11ea-be06-af5514ee0385_story.html Technology - Covid-19 Makes the Case for More Meatpacking Robots https://www.wired.com/story/covid-19-makes-the-case-for-more-meatpacking-robots/ Who is an expert? A service research scholar who is studying Service Robots is Jochen Wirtz https://twitter.com/JochenWirtz/status/1204641608320184321 https://www.emerald.com/insight/content/doi/10.1108/JOSM-04-2018-0119/full/html https://scholar.google.com/citations?user=-_9L9P0AAAAJ&hl=en https://tinyurl.com/ya24mzy7 ==== Register for online event – June 2-3, 2020: https://pontsbschool.com/wp-content/uploads/2020/05/Third_Global_Conference_on_Creating_Value.pdf ==== Practical help: Creating Zoom breakout rooms Here is a link to a 3-minute instruction video. You only need to learn how to: create rooms (with 4 to 5 persons each) Automatically assign the participants (this is a default setting anyway) Set the time how long the rooms will be open (35 minutes). https://support.zoom.us/hc/en-us/articles/206476093-Getting-Started-with-Video-Breakout-Rooms
- URL: http://news.mit.edu/2017/ibm-mit-joint-research-watson-artificial-intelligence-lab-0907 URL: https://www.amazon.com/Master-Algorithm-Ultimate-Learning-Machine/dp/0465094279
- 1950 Nathaniel Rochester (IBM) 701 first commercial computer that did super-human levels of numeric calculations routinely. He worked at MIT on arithmetic unit of WhirlWind I programmable computer. Dota 2 is most recent August 11, 2017 as a super-human game player in Valve Dota 2 competition – Elon Musk’s OpenAI result. Miles Bundage tracks gaming progress: http://www.milesbrundage.com/blog-posts/my-ai-forecasts-past-present-and-future-main-post DOTA2: https://blog.openai.com/more-on-dota-2/
- What is beyond Exascale? Zetta (21), Yotta (24) Time dimension (x-axis) is plus or minus 10 years…. Daniel Pakkala (VTT) URL: https://aiimpacts.org/preliminary-prices-for-human-level-hardware/ Dan Gruhl: https://www.washingtonpost.com/archive/business/1983/11/06/in-pursuit-of-the-10-gigaflop-machine/012c995a-2b16-470b-96df-d823c245306e/?utm_term=.d4bde5652826 In 1983 10 GF was ~10 million. That's 24.55 million in today's dollars. or 2.4 billion for 1 TF in 1983 Today 1 TF is about $3k http://www.popsci.com/intel-teraflop-chip
- Source: http://service-science.info/archives/4741
- +3 from original estimates, getting video understanding (verbs and nouns and context) and episodic dynamic memory for learning events and expectation violations and importance is taking longer than expected… Expert predictions on HMLI: URL https://arxiv.org/pdf/1705.08807.pdf 2015 Pattern Recognition Speech: URL: http://spandh.dcs.shef.ac.uk/chime_challenge/chime2016/results.html 2015 Pattern Recognition Images: URL: http://www.image-net.org/ 2015 Patten Recognition Translation: URL: http://www.statmt.org/wmt17/ 2018 Video Understanding Actions: URL: http://www.thumos.info/home.html > Also UCF101 http://crcv.ucf.edu/data/UCF101.php 2018 Video Understanding Context: URL: http://visualqa.org/challenge.html 2018 Video Understanding DeepVideo: URL: http://cs.stanford.edu/people/karpathy/deepvideo/ 2021 Memory Declarative: URL: https://rajpurkar.github.io/SQuAD-explorer/ Also Allen AI Kaggle Science Challenge https://www.kaggle.com/c/the-allen-ai-science-challenge 2024 Reasoning Deduction: URL: http://www.satcompetition.org/ 2027: Social Interaction Scripts: URL: https://competitions.codalab.org/competitions/15333 2030: Fluent Conversation Speech Acts: URL: http://convai.io/ 2030: Fluent Conversation Intentions: URL: http://workshop.colips.org/dstc6/ 2030: Fluent Conversation Alexa Prize: URL: https://developer.amazon.com/alexaprize 2033: Assistant & Collaborator Summarization: URL: http://rali.iro.umontreal.ca/rali/?q=en/Automatic%20summarization 2033: Assistant & Collaborator Debate: URL: http://argumentationcompetition.org/2015/ 2036: Coach & Mediator General AI: URL: https://www.general-ai-challenge.org/ 2036: Coach & Mediator Negotiation: URL: https://easychair.org/cfp/AT2017
- ROW – Rest of World Who is winning: https://www.technologyreview.com/s/608112/who-is-winning-the-ai-race/ Leaderboards and reproducibility: Hugo Larochelle (Google Brain) (@hugo_larochelle) 8/21/17, 7:36 AM My slides for my talk at ICML 2017 Reproducibility Workshop, on incentives for open source and on open research: https://drive.google.com/file/d/0B8lLzpxgRHNQZ0paZWQ0cTcxMlNYYnc0TnpHekMxMjVBckVR/view Slide 20: Conclusions: "Open source is the key to better reproducibility"
- URL: https://www.slideshare.net/AccentureTechnology/ai-and-the-economy/1 URL: https://www.accenture.com/no-en/insight-artificial-intelligence-future-growth Plastino E, Purdy M (2018) Game changing value from Artificial Intelligence: eight strategies. Strategy & Leadership. 2018 Jan 15;46(1):16-22. URL: https://www.slideshare.net/AccentureTechnology/ai-and-the-economy/1 URL: https://www.accenture.com/no-en/insight-artificial-intelligence-future-growth URL: https://www.researchgate.net/profile/Eduardo_Plastino/publication/323126961_Strategy_Leadership_Game_changing_value_from_Artificial_Intelligence_eight_strategies_Article_information/links/5a81ae20a6fdcc6f3eacfe35/Strategy-Leadership-Game-changing-value-from-Artificial-Intelligence-eight-strategies-Article-information.pdf Someday we will have AI's that help write the first draft of our chapters - see Narrative Sciences: How the future gets written - https://narrativescience.com/ Alex when to grad school with Kris Hammond who founded Narrative Sciences... and one of Alex's and my office mates Natalie Dehn was working on this topic for her dissertation at Yale - see: https://nil.cs.uno.edu/publications/papers/dehn1981story.pdf Circa 1984.... about 34 years ago... Writing papers will get much easier....
- URL: https://www.wsj.com/articles/automation-makes-us-dumb-1416589342 URL: https://maliciousaireport.com/
- The nature of reality changes when there is more than one intelligent species, and we are not the smartest. The nature of reality also changes when the cost of exploring alternate experience pathways are made less risky – the notions of time and identity changes as a result. Mitigate risks and harvest benefits of existence, by learning to evermore efficiently and rapidly rebuild from scratch to higher states of value and capability of entities. The evolving ecology of service system entities their value co-creation and capability co-elevation mechanisms, as well as their capabilities, constraints, rights, and responsibilities at each stage in time. Human progress as well as the development of individuals, and the arc of institutions can be viewed in this way. Entities exist as individuals and populations. Generations of entities, generations of species (populations), generations of individuals (cohorts).
- By 2036, there will be an accumulation of knowledge as well as a distribution of knowledge in service systems globally. We need to ensure as there is knowledge accumulation that service systems at all scale become more resilient. Leading to the capability of rapid rebuilding of service systems across scales, by T-shaped people who understand how to rapidly rebuild – knowledge has been chunked, modularized, and put into networks that support rapid rebuilding.
- URL: https://www.technologyreview.com/s/601641/a-big-leap-for-an-artificial-leaf/
- URL: https://www.independent.co.uk/news/science/world-hunger-food-electricity-carbon-dioxide-ingredients-solve-climate-change-scientists-finland-a7869316.html
- URL: https://www.technologyreview.com/s/601420/the-elderly-may-toss-their-walkers-for-this-robotic-suit/
- The weakest link is what needs to be improved – according to system scientists. Accessing help, service, experts is the weakest link in most systems. By 2035 the phone may have the power of one human brain – by 2055 the phone may have the power of all human brains. Before trying to answer the question about which types of sciences are more important – the ones that try to explain the external world or the ones that try to explain the internal world – consider this, slide that shows the different telephones that I have used in my life. I grew up in rural Maine, where we had a party line telephone because we were somewhat remote on our farm in Newburgh, Maine. However, over the years phones got much better…. So in 2035 or 2055, who are you going to call when you need help?
- Today’s talk will explore two questions What should we know how to make? What might programming education become? If we look at history we see a time when people could make only simple things, and often a single person could make them. Would it ever be possible for a single person to know and make complex things? And what role might programming education play? Will the cognitive era – the coming era of smart machines – make people more capable or less capable to know and make complex things?
- In the 1940’s IBM started teaching computer science at Columbia. My first program – punch cards 1972.
- Wendy Murphy’s dog – hard for AI to recognize in 2016, easy in 2018…
- Visit IBM Research – Almaden, San Jose, CA USA 05120 – instructions: http://service-science.info/archives/4679 Join ISSIP.org – it’s free for individuals to join and get monthly newsletter: http://service-science.info/archives/4901 Contribute a short book to our series – blog compilations welcomed - http://www.businessexpertpress.com/product-category/service-systems-and-innovations-in-business-and-society/ We are trying to make complex servce systems things simpler – but not too simple. Wise innovation increase resilience with abundant opportunities for all.
- Visit IBM Research – Almaden, San Jose, CA USA 05120 – instructions: http://service-science.info/archives/4679 Join ISSIP.org – it’s free for individuals to join and get monthly newsletter: http://service-science.info/archives/4901 Contribute a short book to our series – blog compilations welcomed - http://www.businessexpertpress.com/product-category/service-systems-and-innovations-in-business-and-society/ We are trying to make complex servce systems things simpler – but not too simple. Wise innovation increase resilience with abundant opportunities for all.
- it used to be that computers couldn’t understand images
- Modha’s Brain - Goal 1KW and 2 Litres…. Dharmendra Modha and his design for a brain chip playing pong: https://www.youtube.com/watch?v=gQ3HEVelBFY https://www.youtube.com/watch?v=tqeINGOzIZo https://twitter.com/dharmendramodha/status/545693986149511168
- URL: https://en.wikipedia.org/wiki/History_of_artificial_intelligence URL: http://www.businessinsider.com/infographic-ai-effect-on-economy-2017-8 Today’s infographic comes from the Extraordinary Future 2017, a new conference in Vancouver, BC that focuses on emerging technologies such as AI, autonomous vehicles, fintech, and block http://extraordinaryfuture.com/e/extraordinary-future-2017-71chain tech. Nathaniel Rochester: In 1948, Rochester moved to IBM where he designed the IBM 701, the first general purpose, mass-produced computer. He wrote the first symbolic assembler, which allowed programs to be written in short, readable commands rather than pure numbers or punch codes.
- URL: http://www-formal.stanford.edu/jmc/history/dartmouth/dartmouth.html URL: https://en.wikipedia.org/wiki/Nathaniel_Rochester_(computer_scientist)
- URL: https://www.ted.com/talks/noriko_arai_can_a_robot_pass_a_university_entrance_exam
- O*NET Online is the occupation network online, started by the US Dept of Labor in the 1990’s – it now represents one of the most comprehensive lists of occupations along with a great deal of information about each occupation, including skills, tasks, certifications, demand for these jobs, etc. O*NET lists about 1000 occupations from Accountants to Zoologists – and many job families in between. O*NET updates the descriptions of the occupations as well as adding new occupations over time. Source: http://www.onetonline.org/find/family?f=0
- GitHub – open source code – http://github.com Kaggle – data and competitions – http://Kaggle.com Leaderboard – AI an competitions - https://www.slideshare.net/spohrer/leaderboards-80909263 Figure Eight – label data - https://en.wikipedia.org/wiki/Figure_Eight_Inc. Open Source Guides – reader, contributor, committer, governance - https://opensource.guide/ GitHub is to knowledge in action (writing code) as Wikidedia is to knowledge in text (writing text)
- URL: Why software is eating the world – see https://www.wsj.com/articles/SB10001424053111903480904576512250915629460 URL: Microsoft acquiring GitHub – see https://blogs.microsoft.com/blog/2018/06/04/microsoft-github-empowering-developers/ URL: Models will run the world – see https://www.wsj.com/articles/models-will-run-the-world-1534716720 URL: John Marks, “Why Open Source Failed” https://medium.com/@johnmark/why-open-source-failed-6cae5d6a9f6 First, the good news, which is actually bad. In a 2016 survey from Blackduck, 96% of software products developed that year used open source software. That number is likely higher now. In the software world, particularly software that runs the computing infrastructure of the internet, open source is ubiquitous. One could claim, without any exaggeration, that our current world runs on open source software or that our modern world would not exist in its current form without open source software. I don’t know how to calculate the total value of open source software to the world, but I do know that if open source software suddenly went away, the results would be catastrophic, an existential crisis for humanity. So when I write that “open source has failed” I’m obviously not writing from a technology perspective, where it was been a clear-cut winner and the foundation of an endless supply of business models, products, and services. To say that open source contributed to the overall innovation of the world would be a shameful understatement. Better would be to say that the world’s computing innovations owe their existence to the triumph of open source development. If you think this all sounds pretty terrific, read on to find out what I left out. In the context of this essay, “failure” refers not to any technical achievements but rather to the lack of social ones. When we were but wee lads and lasses on the forefront of this thing we called free software and eventually open source, we knew that this was dangerous stuff. It was destined to set fire to an entire industry, undermining entrenched monopoly powers and establishing a more equitable approach to building wealth around the tools that would power humanity in the 21st century. It was about the democratization of software and would smash what we then called the “digital divide”. That premise was entirely false. The crux of this essay is thus: not only did open source not stem or stall the redistribution of wealth and power upwards, but rather it aided and abetted the redistribution of wealth and power upwards. To be an open source proponent at this time without acknowledging this very real and most unfortunate consequence is to be a cog in a much larger machine; a stooge; a very useful idiot. When considering the role of open source in redistributing wealth upwards, it’s instructive to consider the example of Microsoft. Not because I enjoy picking on them or think they’re evil — I don’t; Microsoft as a publicly traded company is no more or less evil than any other company. Rather, I like to single them out because their public stance towards open source has changed much over the years and is a useful measuring stick for the points I’m trying to make. Did you ever wonder *why* their public stance towards open source shifted so much over the years, from “Linux is a cancer” to “use our open source software”? Could it be because, unlike the company’s predecessors in 2000, current executives now understand that open source software forms the building blocks of modern capitalist behemoths?
- URL: https://www.nytimes.com/2017/12/27/business/the-robots-are-coming-and-sweden-is-fine.html
- AI Fairness 360 toolkit (AIF360) is a comprehensive open-source toolkit of metrics to check for unwanted bias in datasets and machine learning models, and state-of-the-art algorithms to mitigate such bias. Machine learning models are increasingly used to inform high-stakes decisions about people. Although machine learning, by its very nature, is always a form of statistical discrimination, the discrimination becomes objectionable when it places certain privileged groups at systematic advantage and certain unprivileged groups at systematic disadvantage. Bias in training data, due to either prejudice in labels or under-/over-sampling, yields models with unwanted bias. The initial release of the AIF360 Python package contains nine different algorithms, developed by the broader algorithmic fairness research community, to mitigate that unwanted bias. The goal of the package is not only a way to bring researchers together, but also to translate collective research results to data scientists, data engineers, and developers deploying solutions in a variety of industries. AIF360 is a bit different from currently available open source efforts due its focus on bias mitigation (as opposed to simply on metrics), its focus on industrial usability, and its software engineering. AIF360 is relevant to credit scoring, predicting medical expenditures, and classifying face images by gender, among many other areas.