Technologies Demystified: Artificial Intelligence

Technologies Demystified: Artificial Intelligence

• 1.
  TECHNOLOGIES DEMYSTIFIED ARTIFICIAL INTELLIGENCE
• 2.
  TABLE OF CONTENTS 2 1.Introduction to Artificial Intelligence 4 - 11 1.0 Glossary of Key Terms 1.1 What is AI? 1.2 Fields of AI 1.3 Terminology 1.4 A Brief History of AI 1.5 The Road to Superintelligence 1.7 The Opportunities and Challenges of AI 2. Application Fields & Use Cases 12 - 14 2.1 Application Fields of AI 2.3 Austrian AI Startups 3. Outlook 15 - 18 3.1 Future Outlook
• 3.
  3 ABOUT Artificial Intelligence (AI)is one of the hottest topics in the tech and startup world at the moment. The field of AI and its associated technologies present a range of opportunities – as well as challenges – for corporates. Pioneers Discover focuses on emerging technologies like AI and therefore we wish to share our valuable insights so that you can introduce AI capabilities into your organization. As an advisory for startup-driven innovation, we can facilitate this by connecting you with our network of impressive AI startups that we’ve built over the years. Take a read and let us know how we can support you in the area of AI! 3
• 4.
  INTRODUCTION TO ARTIFICIALINTELLIGENCE 4
• 5.
  5 GLOSSARY OF KEYTERMS Souces: machinelearnings.co, 2017; Deep Learning: Methods and Applications, 2014 Neural Networks Autonomous Big Data Turing Test Natural Language Processing AI Technology Terms Singularity A computer system modelled on the human brain and nervous system. A term used in mathematics to describe a situation where normal rules no longer apply. In the context of AI, this expression is used when technology’s intelligence exceeds humans. A branch of AI concerned with understanding and analyzing the interactions between computers and human languages. A test to determine a machine’s capability to exhibit intelligent behavior equivalent or even indistinguishable from a human. A term used to describe data sets that are so large and complex that traditional data processing software applications are unable to deal with and analyze them. The ability to act independently of a ruling body. In AI, a machine or vehicle is referred to as autonomous if it doesn't require input from a human operator to function properly.
• 6.
  WHAT IS AI? AIis a broad and transformative technology, ranging from every day applications such as recommendation systems all the way to self- driving cars. It encompasses an entire branch of computer science, but can be split up into three critical categories: • Artificial Narrow Intelligence (ANI): Also referred to as Weak AI, ANI is AI that specializes in one particular area and lacks human consciousness. For example, a program that can tell you how to drive from point A to point B is usually incapable of challenging you to a game of chess. • Artificial General Intelligence (AGI): AGI or Strong AI refers to a computer that is as intelligent as a human across the board. That includes the capability to reason, plan, solve problems, think abstractly and learn quickly. • Artificial Superintelligence (ASI): Superintelligence is defined as an intellect much more intelligent than the best human minds in virtually any field including creativity, general wisdom and social skills. The concept of ASI is one of the main reasons the topic of AI is so relevant for our society. In this report, we’ll explain these categories in more detail and the road being taken to achieve ASI. Souces: Deloitte University Press, Waitbutwhy, 2015 6 “It is the science and engineering of making intelligent machines, especially intelligent computer programs.” - John McCarthy The point where computer performance, as shown in the graph, exceeds human’s is inevitable and not too far ahead. Performance
• 7.
  1. Intelligent systems:Rule-based software that consists of a database, an amount of fixed rules and a control system. Based on the current problem set the suitable rules are picked and applied. In this type of AI the computer is not able to learn autonomously. 2. Machine learning: Logic reasoning where rules are extended through logical terms and enlarge the database. These relations are produced through statistics. Machine learning aims to teach computers to learn from data and perform tasks without being explicitly programmed for it. 3. Deep learning: This subfield of machine learning makes use of multiple layers of information processing for feature extraction, pattern analysis and classification. Deep learning systems are modeled after neural-networks in the human brain and are running statistical methods, that take input and pass along output. 4. Machine thinking: A replication of the human brain in an anatomical and functional way through neural networks, which enables the ability to act in response to unforeseen situations and produce new ideas. Souces: Xephor Solutions, 2017 7 Intelligent systems Machine learning Deep learning Machine thinking FIELDS OF AI
• 8.
  8 A (VERY) BRIEFHISTORY OF AI - PART 1 1955: The term “Artificial Intelligence” is coined by John McCarthy as the topic for the first conference on the subject. 1961 Unimate, the first industrial robot, starts working on a General Motors assembly line. 1986 A Mercedes-Benz constructed in Munich becomes the first driverless car, driving up to 80 km/h on empty streets. 1993 Vernor Vinge publishes “The Coming Technical Singularity”. He predicts that within 30 years superhuman intelligence is possible and that shortly after the human era will end. 1979 The Stanford Cart becomes one of the earliest examples of autonomous driving by crossing a chair-filled room without human intervention. 1997 IBM’S Deep Blue beats the reigning world chess champion, becoming the first computer chess-playing program to do so. 1990s Web crawlers and similar AI-based information extraction programs become fundamental in the general use of the internet. Souces: Gil Press; Forbes, 2016; MissQT Fintech, 2015; AITopics, 2016; The History of Artificial Intelligence, University of Washington, 2006 2000 MIT builds Kismet, the first robot to recognize and simulate human emotions.
• 9.
  9 A (VERY) BRIEFHISTORY OF AI - PART 2 2000 Honda’s artificially intelligent humanoid robot, ASIMO, is able to walk as fast as humans and deliver trays in a restaurant. 2006 “Learning Multiple Layers of Representation”, published by Geoffrey Hinton, shapes the new approach to deep learning. 2011 IBM Watson competes on live television on Jeopardy! and defeats two former champions. 2012 Applications like Apple’s Siri, Google Now and Microsoft’s Cortana are able to answer questions, give recommendations and perform actions that use natural language. 2009 Google starts the secret development of completely driverless cars. 2016 The two Google Neural Networks, Bob and Alice, are able to create an unsupervised encryption method without human intervention. 2014 The Chatbot “Eugene Goostman” is the first computer program to ever pass the Turing Test. Souces: Gil Press; Forbes, 2016; MissQT Fintech, 2015; AITopics, 2016; The History of Artificial Intelligence, University of Washington, 2006; Wikipedia Timeline of artificial intelligence, 2017 2017 Google’s AI is able to detect cancer faster than human doctors and with an accuracy of 92.4%
• 10.
  THE ROAD TOSUPERINTELLIGENCE Building human-level intelligence is an incredibly challenging task for a number of reasons. First, it is important to understand that the human brain is the most complex object in the known universe and still not fully researched. Interestingly the hardest parts of building true AI are not what they seem to be. Things like calculus or language translation have been solved already, but vision, motion and perception are extremely difficult for a computer. This may seem paradoxical, but it is closely connected to human evolution. Over millions of years humans were able to optimize those skills and “program” the software in our brain accordingly. On the other hand mathematics or playing chess are new to us and therefore have not been evolved. For a computer to truly reach human-level AI, it would be necessary to recognize facial expression, understand context or learn by itself. To reach this level, one of the most important concepts is recursive-self- improvement, where an AI system is programmed to improve itself and soar upwards along the intelligence scale, overtaking humans. This is what we call an intelligence explosion, and there’s no real way to predict what ASI will be able to do once this is achieved. Souces: Deloitte University Press, Waitbutwhy, 2015 10 “AI has by now succeeded in doing essentially everything that requires ‘thinking’ but has failed to do most of what people and animals do ‘without thinking.” - Nils Nilsson FROM ANI TO ASI Increase Computational Power For an AI system to be as intelligent as the human brain (AGI), it has to match the brain’s computing power. According to Moore’s Law, a reliable rule that the world’s maximum computing power doubles approximately every two years, advancement in this area is growing exponentially. Make it Smarter This is the hard part of building human-level AI. Although a couple of methods seem promising nobody truly knows how to make machines smart. Possible scenarios involve emulating the human brain, using evolution and self-learning techniques. How do we get there? So, When Will It Happen? The question at this point is not if it is going to happen, but rather when. The median year on a survey of hundreds of scientists revealed the year 2040 for AGI and shortly after that ASI.
• 11.
  11 THE OPPORTUNITIES ANDCHALLENGES OF AI The rapidly developing field of AI presents a series of opportunities and challenges for corporates looking to transform their business. “Pay close attention to the development of AI, we need to be very careful in how we adopt AI.” - Elon Musk, CEO of Tesla Souces: information-age, 2017; futureoflife.org 2016; Harvard Business Review, 2017 AI systems with real human-level intelligence will provide a massive challenge for political systems around the world. When we get to that point, rules and regulations have to be in place to control these systems and the companies owning them. Ethics and Politics One of the most important challenges will be to program AI systems according to human morals and beliefs. Since AI systems will be extremely good at achieving its goals, it is crucial to prevent such misalignment. Misaligned AIOpportunities AI systems will inevitably be able to perform jobs and tasks better than humans, creating possibilities for corporates to cut costs and increase productivity. Tasks like document or process management and data analysis can be automated by AI and its foundation technologies. While job loss is a factor, AI systems will rather augment workers than replace them. AI Automation ü Detecting and deterring security intrusions ü Resolving users’ technology problems ü Anticipating customer behavior and presenting offers ü Financial trading ü Automating call distribution ü Monitoring social media ü Gauging internal compliance Challenges
• 12.
  APPLICATION FIELDS &USE CASES 12
• 13.
  13 APPLICATION FIELDS OFAI Health Care Topics Cancer detection Health assistants Dynamic care Better pattern recognition and access to huge datasets will allow the healthcare sector to thrive with AI. Google developed an algorithm that can detect breast cancer at an early stage and will help radiologists to detect tumors, which could save millions of lives and dollars. Mobility Topics Self-driving cars Connected cars By the year 2025, self- driving cars will create a $42 billion market that will allow freight companies to save up to $70 billion annually. More incredibly, around 90% of all traffic fatalities will be reduced. Logistics Topics Robotics Drones 3D printing A leading example, Amazon was able to cut warehouse operation costs by 20% by using robots. AI-automation will soon hit other industries operating in the logistics field such as waste management reducing costs and even helping the environment through more efficient disposal of waste. White-collar Industry Topics Virtual assistants Process automation Knowledge workers spend half of their time doing research, arranging meetings or coordinating other people. AI- automation is able to replace these tasks and augment work. Sales, consulting and law are examples of industries that will be most significantly impacted. Souces: machinelearnings, 2016
• 14.
  14 AUSTRIAN A ISTARTUPS LOCATION Vienna FOUNDED 2011 IN BRIEF Cortical.io develops Natural Language Understanding (NLU) solutions based on proprietary Semantic Folding technology inspired by the latest findings on the way the human cortex works. The central technology component encodes words in the same way as sensorial information is fed into the brain. PRODUCT/TECHNOLOGY Based on the input data, representations are created which encode the meaning of language and measures semantic similarities. Its easy-to-use platform allows for fast semantic search, classification and filtering. LOCATION Vienna FOUNDED 2012 IN BRIEF AGI-based software system that is able to overtake tasks, where creativity and the generation of new ideas are needed. After a period of training the system works completely autonomously and performs tasks, where nowadays a highly- skilled workforce is needed. PRODUCT/TECHNOLOGY The real-time software system is developed from scratch with AI, high performance computing (HPC) and a strong encryption. It is a replication of the human brain and its funtionality, which allows for functions like filtering, predicting, pattern association and pattern recognition to be done. Souces: xephor-solutions, 2017; cortical.io 2017
• 15.
  PIONEERS DISCOVER OUTLOOK 15
• 16.
  FUTURE OUTLOOK Souces: Waitbutwhy,2015 ASI AGI Outlook 2040 In a study conducted at an annual AGI Conference, participants concluded we are likely to achieve human-level AI by the year 2040, only 23 years from now. Outlook 2060 The timeframe from AGI to ASI is very hard to predict and could potentially happen very quickly with recursive self-improvement. A realistic estimate expects the arrival in the year 2060. Robotics AI for robotics will enable humans to take on challenges such as the increasingly aging population and natural disasters. Climate change Super-intelligent AI systems could potentially halt CO2 emissions by developing a better way to produce energy and remove fossil fuels. Human enhancement Multiple experts predict that humans and computers will become one tightly-coupled cognitive unit. Startups like Neuralink are already developing an ultra-high bandwidth brain-machine interface. Autonomous driving By the year 2025, the market for self-driving cars will be worth $83 billion, with the highest investment opportunities in AI. 16
• 17.
  17 Happy to help! Feelfree to contact us in case you have questions about AI technology and how your company might benefit from it! Christoph Fuchsjäger Consultant Christoph.fuchsjaeger@pioneers.io www.pioneers.io/discover/about
• 18.
  Technology needs visionariesto make it fly. Be Bold. Be a Pioneer. pioneers.io/discover/about @pioneers @pioneersfestival pioneers.io