ARTIFICIAL INTELLIGENCE by the Group Digital & IT Department

Transcription

1 ARTIFICIAL INTELLIGENCE by the Group Digital & IT Department

2 UMMARY AI : What is it about?...4 What s new?...6 The most mature applications...8 AI is already part of our daily lives...9 Fantasies and imagination...10 AI technologies...12 At the service of the energy transition...14 Its numerous challenges...16 To be continued Annex Annex Reading time : 20 min EDITORIAL Technologies are multiplying and many terms, with different meanings, exist to describe them. In order to understand their value, decrypt their challenges and go beyond emphatic promises, it is necessary to clearly explain them and to examine their applications in the real world: for whom? To do what? Why? The Group Digital & IT Department decided to produce a series of White Papers addressing these new technologies to answer four main questions: What are we talking about? What are the challenges and limits of these subjects? What are they used for? Their applications in the Energy sector, and in particular at ENGIE? This is the proposed approach for all the upcoming White Papers. This first edition gives you a look back at the fundamentals of Artificial Intelligence, then gradually explains what it already does and what value it could bring to the businesses, employees, to ENGIE and more globally to the Energy transition which we are part of. This open discussion continues on the Group Digital & IT Department s Yammer group: IT Hub. Why not join and share your point of view? Enjoy the deep dive! Contacts: Matthieu Pestel and Etienne Gehain (Group Digital & IT Department) 3

3 WHAT IS IT ABOUT? The world of Artificial Intelligence is characterized by its learning abilities and distinguishes itself from the world of Preprogrammed Computing: The media hype around the term Artificial Intelligence (AI) requires clarification. Artificial Intelligence is a science that emerged after the Second World War and the invention of the first electronic computers by mathematicians and software builders (Turing, Shannon, Wiener, Gödel...). The purpose of this science was twofold: to simulate human capacities to understand human intelligence better, and to replace humans in certain automatic and repetitive tasks. Since then, this science has seen numerous developments. Artificial Intelligence is commonly defined as follows: The purpose of the science of Artificial Intelligence is to have a machine accomplish the tasks otherwise accomplished by a human intelligence. PREPROGRAMMED COMPUTING 1STEP Code of a sequence of instructions to tell the program how to do it ARTIFICIAL INTELLIGENCE The program learns There are two limits to this definition: The concept of intelligence still Classical computer applications could be 2STEP The program executes that sequence of instructions When the program is questioned, it uses what it learnt to give an answer needs to be defined; included in that definition. ARTIFICIAL INTELLIGENCE AIMS TO SIMULATE, REPRODUCE OR EVEN SURPASS Thus, Artificial intelligence refers to learning technologies and underlying mathematical theories of neural networks. HUMAN INTELLIGENCE. 4 5

4 WHAT SNEW? In the last decade, Artificial Intelligence was able to grow thanks to two main factors: 1 Learning theories have matured. 2 Moore s Law made the computing They began to be taught and and memory capacities accessible communities were set up to share at a lower cost to unroll AI algorithms. their inventions ( open source ). Artificial Intelligence has a history filled with breakthrough moments of great excitement and dry patches: to to to to Information Creation Alan Turing s test Creation of First implementations of First dry patch Invention Second dry theory by Claude of the term definition the AI concept neural networks (Perceptron) period for the AI of dedicated patch period Shannon cybernetics by (Dartmouth conference First achievements: programming for the AI Norbert Wiener organized by John Experimental autonomous languages (Prolog, McCarthy) vehicle LISP...) to create Machine that plays chess expert systems, Conversational system inference engines in natural language or declarative systems october Work on neural Deep Blue (IBM) Announcement Watson (IBM) wins Announcement (contested) AlphaGo beats AlphaGo Zero Saudi Arabia grants networks beats Kasparov of the Google Car the Jeopardy game of Eugene Goostman s the Go worldwide becomes self- citizenship to the in chess successful completion champion Lee Sedol taught Sophia robot of the Turing test at the Reading University 6 7

5 THE MOST MATURE APPLICATIONS AI IS ALREADY PART OF OUR DAILY LIVES Artificial Intelligence is already part of our daily lives and is used in most of the following activity sectors: ONE EXAMPLE: THE SMARTPHONE. Voice search Geo-tracking Image recognition (e.g.: facial recognition) Video games Translation Predictive text Voice recognition (e.g.: speech to text ) Assistance in medical diagnosis filtering Itinerary suggestions Vocal assistant There are some limits to these applications: Treatment of natural language (e.g.: translation software) Predictive industrial systems (e.g.: predictive maintenance, ) 1 The current systems are still very specialized (AlphaGo can only play Go); 2 Like any computer system, these algorithms are not infallible and errors can occur; 3 It is very difficult to validate and verify a learning system. As much as a conventional computer program can be controlled, a learning system is like a black box for the human understanding. In a certain Conversational agents / ChatBots Personal assistants on smartphones (e.g.: Siri) Semantic search (e.g.: search Personalized recommendations for products, music, etc. way, humans cannot reconstitute the logic, the causal path of the machine (more exactly the system remains deterministic but at a level that far exceeds the human capacity to control it. The tactics used by the machines to win in chess or Go have challenged the usual strategies of these games. HUMAN UNDERSTANDING CAN NO LONGER CONTROL THE INTERNAL LOGIC OF THE ARTIFICIAL INTELLIGENCE SYSTEMS. engines) 8 9

6 FANTASIES AND IMAGINATION AI IS ALSO A SUBJECT OF POLITICAL PHILOSOPHY, ETHICS AND LAW. Artificial Intelligence also refers to how humans have imagined robots. Its history can be traced back to ancient times: 4th century BC The animated tools of Aristotle (La Politique) 8th century BC Tripods of the god Hephaestus that move automatically (Homer, L Iliade) 1646 Humans have often given Artificial Intelligence a human face. They project themselves onto the machines, and for the first time in the history of humanity, feel overwhelmed by another form of intelligence. Humans knew that they were being overtaken by certain animals in other fields (power, speed, perception...), but learnt to compose, master or tame. It is the end of a monopoly and a source of astonishment, vertigo and anxiety. Which is probably why this subject of reason triggers so much passion. But are we really talking about the same intelligence? 16th century The Animal-Machine of René Descartes This question led the philosopher John Searle to identify a weak AI and a strong AI. The Kabbalistic tradition The weak AI The strong AI, of Golem (originates from Prague, 16th century) 1818 refers to the development of expert and learning systems on the other hand, refers to the idea of a machine that that are based on mature has a sentimental consciousness, 1920 Invention of the word robot (from Rabota: work, chore) by the Czechoslovak writer Karel Čapek The novel Frankenstein or the modern Prometheus of Mary Shelley 1968 or emerging technologies using societal applications. This AI is fitted with no intelligence, no emotions, nor intentions or will identical to humans. exercising its will or free will with its own identity. This strong AI remains very utopian to this day, even if theories are developing around the concept of technological The film Blade Runner and its replicants 1982 The film 2001, The Space Odyssey and the HAL computer Fundamental questions are being raised by Artificial Intelligence and essential concepts are being discussed: consciousness, free will, common sense, intention, intuition, emotion... These questions could also be raised in the animal world with a triptych Man vs Animal vs Machine

7 AI TECHNOLOGIES The planet is facing the challenge of energy transition. By increasing management capabilities, Artificial Intelligence makes it possible to implement new and optimized energy systems: nonexhaustive SEMANTIC ANALYSIS IMAGE RECOGNITION CONVERSATIONAL INTERFACE EMPATHY KNOWLEDGE MANAGEMENT PREDICTION / PRESCRIPTION Generic use cases Generic use cases Generic use cases Generic use cases Generic use cases Generic use cases Automatic translation Facial recognition Chatbots Social robots (Pepper, Automatic consolidation Products / services (Reverso) Product search using Personal assistants Matilda, Paro) of information recommandation Contextual analysis of data (text) photos (Blippar) Energy use cases (Hey Google, Alexa) Energy use cases Energy use cases Automatic segmentation Detection of anomalies (fraud) Energy use cases Monitoring of renewable Vocal control of comfort of customers Energy use cases Customer satisfaction production assets (smart home) or services Predictive maintenance measurement Thermal mapping of (smart building) of production/storage assets Interoperability of buildings Prediction of energy production, storage and production (linked energy consumption assets to weather forecast) Prediction of consumption ENGIE USE CASES ENGIE USE CASES ENGIE USE CASES ENGIE USE CASES ENGIE USE CASES ENGIE USE CASES CUSTOMER RELATIONSHIP Identification of the main areas of concern for customer service phone calls Prospect identification for energy efficiency offers (Accent project) Chatbot Promotional campaign with the Pepper robot at the entrance of department stores Customer churn reduction OPERATIONAL EFFICIENCY Interoperability of production, storage and energy consumption assets (Contextual exchange of data between different IoT- SEAS project) Monitoring of renewable energy production assets Thermal mapping of buildings BENGIE (purchasing) Prediction of site consumption (sourcing optimization, flexibility aggregation, Green Charge Networks) PREDICTIVE MAINTENANCE Detection of hot spots on photovoltaic panels, small cracks on wind turbines, etc. Analysis of vibrations in alternators (electricity production, Laborelec LVMS) NEW OFFERS Smart thermostat control Homecare for elder people Customization of B2C offers based on automatic criteria 12 13

8 AT THE SERVICE OF THE ENERGY TRANSITION SMART HOME Management of the home automation hub Optimization of energy consumption with smart meters and connected thermostats Learning of the rhythm of daily life Optimization of the recharging of electric cars Optimization of solar panels production SMART BUILDINGS & FACTORIES Optimization of resources and building fluids (electricity, gas, lighting, heating, air conditioning, water, etc.) Occupant services (information, security...) SMART CITIES Optimization of a city s resources (lighting, water...) Optimization of flows (transport...) Optimization of services (security...) SMART GRIDS Optimization of distribution and transmission networks Equilibrium management Energy Communities ELECTRIC MOBILITY CENTRALIZED PRODUCTION DECENTRALIZED PRODUCTION Optimization of production DATACENTERS Optimization of the energy Optimization of storage / recharging of electric vehicles Optimization of production systems (predictive systems (predictive maintenance...) Intermittency management consumption of Artificial Intelligences, blockchains... maintenance...) 14 15

9 ITS NUMEROUS CHALLENGES THE CHALLENGE OF VIRTUAL IDENTITIES AI enables the development of virtual identities. The mathematician Alan Turing created a test in which a human being communicates with a machine and has to recognize whether it is a machine or a human being. The machine passed the test if the human being did not realize that he/she was communicating with a machine. Today, some machines passed the test. In the future, the distinction between human vs. virtual identities will become an issue for our societies, particularly in terms of security. THE CHALLENGE OF AUGMENTED WEAPONS AND NATIONAL SOVEREIGNTY Like many technologies, it is clear that the AI will also be used for military purposes (parallel to the development of theoretical physics at the beginning of the 20th century and the invention of the atomic bomb). This raises the question of military drones and other systems. What limits should be set? The AI will be at the heart of military arsenals. Its mastery is therefore a matter of national sovereignty (particularly for the United States and China, who are ahead of the game). THE CHALLENGE OF HUMAN RESPONSIBILITY AND MACHINE AUTONOMY Two situations may arise with a machine: It is at the service of the person who has the final decision (enlightened or augmented human intelligence) It is autonomous and acts according to its conception. The responsibility issues in both cases are not trivial. In the first case, are humans still able to make the right decisions? Are their decisions always as rational as a machine could be? In the second case, in case of failure, error or fault, who should be held responsible: the designer? the owner? the AI teacher?... What should we do with the machine: Destroy it? Fix it at the risk of recidivism? Will there be a need for the same approval procedures used for drugs?... It should be noted that these questions do not only refer to the domain of AI alone. Machines or algorithms can be autonomous without using learning systems (e. g. APBA THE CHALLENGE OF INDIVIDUAL FREEDOM AND DIGITAL TRUST THE CHALLENGE OF FRENCH AND EUROPEAN DIGITAL LEADERSHIP France owns many assets in the digital world: quality of engineering training, French school of mathematics (13 Fiels medals including Cédric Villani), a rich ecosystem of start-ups... But these assets have not yet translated into the emergence of one or more digital champions (Blablacar remains our most successful). Caught between America and its GAFAMs (Google, Amazon, Facebook, Apple, Microsoft and NATU (Netflix, Airbnb, Tesla and Uber) on one hand, and China and its BATXs (Baidu, Alibaba, Tencent and Xiaomi) on the other, France and Europe must find their own place. It s about scaling up and creating digital giants. THE CHALLENGE OF TALENT AND THE IMPACT ON JOBS In the midst of the exponential development of AI, talent has become rare and is taken away by some major players in the digital world, such as Google, Apple, Facebook, Amazon, Microsoft, IBM, NVIDIA... Added to that, the lack of teachers slows down the training of talent in that field. Also, questions around employment are getting the attention of trade unions and political parties. But the impact is still unknown, as many jobs will have to adapt, and not only the working class jobs. Artificial Intelligence is a big consumer of data, particularly personal data. In the context of the digital transformation of our society, the balance needs to be found between freedom of entrepreunership, the development of uses and the regulation of personal data. A certain trust with each citizen and consumer must be established and preserved. THE DEMOCRATIC CHALLENGE OF LEARNING In a world that is increasingly being dominated by computers and algorithms, learning about them has become a condition of freedom and democracy, in order to understand them and not just be an IT consumer

10 TO BE CONTINUED ANNEX GLOSSARY The Artificial Intelligence world represents our new horizon. We are heading towards an Augmented Society, in fact we are already there. This movement is irrepressible and irreversible. Artificial Intelligence mixes many concepts and terms that need to be defined in order to become familiar with this technological universe. Some of these are explained below to help you. If you need more information, do not hesitate to contact Matthieu Pestel and Etienne Gehain from the Group Digital & IT Department. Although ethical questions and fears arise, it is above all a matter of embracing this new field of development and growth opportunities with confidence and daring. A new world is to be invented. France and Europe still need to find a place in it. We need strong ambition and the emergence of digital leaders. And ENGIE will play its part in the field of energy transition. BIG DATA (MEGADATA, MASSIVE DATA) A very large volume of structured or unstructured data which requires adapted analysis tools (Official Journal of the 22nd of August 2014). The big data phenomenon is linked to new uses, such as social networks, online videos or connected objects, which create billions and billions of bytes of data every day and can no longer be stored, processed and analyzed with conventional tools. BLOCKCHAIN (TRANSACTION REGISTRY) A technology that makes it possible to keep track of a set of transactions in a decentralized, secure and transparent way, that takes the form of a chain of blocks. Blockchain is used in particular to make cryptographic currencies, such as bitcoin. DATA MINING A technique of analyzing data using different perspectives and transforming this data into useful information by establishing relationships between data or identified patterns. This information can then be used by companies to increase sales or reduce costs. It can also be used to better understand customers in order to establish relevant marketing strategies. DATA SCIENTIST Engineer in charge of managing, analyzing and exploiting massive data within a company

11 DEEP LEARNING A learning technique that enables a program, for instance, to recognize the content of an image or understand spoken language. The deep learning technology learns to represent the world, how the machine can represent speech or image for example (Yann Le Cun). MACHINE LEARNING A process that allows a computer or an artificial intelligence to learn from its experience or via the analysis of examples, rather than using specific programming. This automatic learning was used in particular for Alpha- Go, the computer program that defeated the South Korean Go game champion Lee Sedol in March SINGULARITY The technological singularity is the hypothesis that the invention of Artificial Intelligence would trigger a boom in technological growth which would lead to unpredictable changes in human society. Beyond this point, progress would no longer be the work of Artificial Intelligences or a superintelligence that would improve itself. But, with new and increasing intelligent generations appearing faster and faster, building an explosion of intelligence, it would ultimately create a powerful superintelligence exceeding human intelligence, in quality, by far. The risk would be the loss of human political power over our destiny. Such consequences were debated in the 1960s by I. J. Good. According to Ray Kurzweil, co-founder of the Singularity University, this notion of technological singularity was first considered by John von Neumann in the 1950s. The possibility and timing of this hypothetical event raises debate among scientists. Several futurologists and transhumanists await its happening for the third decade of the 21st century. TRANSHUMANISM Transhumanism is an international cultural and intellectual movement advocating the use of science and technology to improve the human condition, particularly by increasing the physical and mental characteristics of human beings. Transhumanists consider certain aspects of the human condition such as disability, suffering, illness, ageing or death to be unnecessary and undesirable. Transhumanism shares many elements with humanism, including respect for reason and science, a willingness to progress and the enhancement of human (or transhuman) existence. It differs, however, in recognizing and anticipating the radical changes that may result from emerging technologies. The transhumanist movement is concerned with the dangers and benefits of such developments. 02 SOME BENCHMARKS AND REFERENCES IN ARTIFICIAL ANNEX EVOLUTION OF THE AI APPLICATIONS INTELLIGENCE The first applications of Artificial Intelligence were developed for strategy games (checkers, chess, go). SPAM Anti-spam features in personal mailboxes are a classic example of machine learning. Image recognition or speech recognition are classic examples of deep learning. Source: Sources Wikipédia ALAN TURING S TEST (1950) The Turing test is an artificial intelligence test based on the ability of a machine to imitate the human conversation. Described by Alan Turing in 1950 in his publication Computing machinery and intelligence, this test consists of putting a human in a blind verbal confrontation between a computer and another human

12 JOHN SEARLE S CHINESE ROOM (1980) The term Chinese chamber refers to a thought experiment imagined by John Searle around Searle wondered whether a computer program, no matter how complex, would be enough to give a system a spirit. This thought experiment aims to show that an artificial intelligence can only be a weak artificial intelligence and can only simulate a consciousness, rather than possessing genuine mental states of consciousness and intentionality. It also aims to show that the Turing test is insufficient to determine whether or not an AI has these mental states. In this thought experiment, John Searle imagines a person who has no knowledge of Chinese (in this case, himself) locked in a room. This person is provided with a catalogue of rules that helps answering Chinese sentences. These rules are perfectly clear to the operator and their application is based solely on sentence syntax. A sentence of a certain syntactic form in Chinese is correlated with a sentence of another syntactic form. The operator locked in the room receives the sentences written in Chinese and, by applying the rules that he has at his disposal, produces other sentences in Chinese which are in fact answers to questions asked by a true Sinophone located outside of the room. From the speaker s point of view, the person locked in the room behaves like a real Chinese speaker. But, in this case, the latter has no understanding of the meaning of the Chinese sentences he transforms. He is just following predetermined rules. Ironically pointing towards the Turing test procedure, which is supposed to show that a sophisticated computer program can be described as intelligent, John Searle imagines that the program determining the answers given to the Chinese-speaking interviewer becomes so sophisticated, and the non-chinese person answering the questions becomes so skilled at manipulating symbols, that at the end of the experiment, the answers to the questions cannot be distinguished from a true native Chinese speaker. Although according to John Searle, the person we imagine locked up in the room still does not understand a word of Chinese. This experience of thinking shows that it is not enough to be able to reproduce exactly the linguistic behaviors of a Chinese speaker to speak Chinese, because speaking Chinese, or any other language, is not just saying the right things at the right time, it is also signifying or wanting to say what we want to say: a controlled use of language is thus coupled with an awareness of the meaning of what we say (intentional awareness). And the artificial reproduction, even perfect, of a linguistic behavior is not enough to reproduce such conscience. ISAAC ASIMOV S THREE LAWS OF ROBOTICS (1942) The Three Laws of Robotics, written by the science fiction writer Isaac Asimov, are rules that all robots must obey. The rules were introduced for the first time in his 1942 short story Runaround and they say that: 1 A robot may not injure a human being or, through inaction, allow a human being to come to harm. 2 A robot must obey the orders given it by human beings except where such orders would conflict with 3 the First Law. A robot must protect its own existence as long as such protection does not conflict with the First or Second Laws. 22

13 Contacts: Matthieu Pestel and Etienne Gehain (Group Digital & IT Department) ENGIE 1 place Samuel de Champlain, Faubourg de l Arche Paris La Défense CEDEX, France T +33 (1) F +33 (1) engie.com