2. Artificial Intelligence in:

Henry Alexander Wittke

Artificial Intelligence, page 6 - 16

An Approach to Assess the Impact on the Information Economy

1. Edition 2020, ISBN print: 978-3-8288-4459-9, ISBN online: 978-3-8288-7480-0,

Tectum, Baden-Baden
Bibliographic information
economic sectors are affected, what implications may arise for the entire information society and for the political sphere? Before discussing op tions for such actions, chapter seven will seek to further establish the va lidity of the action claim put forward by looking at already occurring impacts of AI. These short elaborations seek to foster the rather theoreti cal argument to Machlup’s framework with current evidence. Then, chapter eight discusses areas for actions to be taken, here proposing the need for further education for so-called knowledge workers, retraining the workforce of the future or a basic income. Finally, chapter nine will do a short critical review of the research so far, and then the conclusion will finish this thesis. 2. Artificial Intelligence The beginning of the following chapter will concentrate on creating a definition of artificial intelligence and describes briefly what makes this emerging technology so powerful. Nowadays artificial intelligence is regularly used by people as shorthand to talk about everything from building robotic process automation tools, to chatbots, neural networks or deep learning (cf. Donovan 2017). The reappearing boom of AI in recent years in all things artificial intelligence catalyzed by break throughs in the area of machine learning. Machine learning can be de fined as a set of methods that can automatically detect patterns in data, and then use the uncovered patterns to predict future data, or to perform other kinds of decision making under uncertainty. It involves training computers to perform tasks based on examples, rather than by relying on programming by a human. In short, it is a self-teaching computer system (cf. Murphy 2012: 32; Piovesan/ Ntiri 2018: If.; Wired 2018; Fraser 2017). The most relevant types of machine learning are supervised, un supervised and reinforcement learning. The details of each of these methods are beyond the scope of this paper. Today these methods can be combined with a deep learning architecture, which makes machine learn ing approaches ways more powerful. Deep learning operates by using artificial neural networks. They contain layers of nodes that in some ways mimic the neurons in the human brain, to train computer systems on large quantities of data to recognize patterns in digital representations of sounds, images, and other data. Every single layer of neurons takes the data from the layer below it, performs a calculation, and provides its 6 output to the layer above it. This architecture can be combined with an unsupervised process to learn the features of the underlying data, such as the edge of the face, and then provide that information to supervised learning algorithms to recognize features as well as the final result, which in the example of a photo of a human face correctly identifies a person in the picture (visual recognition). Besides these combinations, the most promising method is reinforcement learning combined with deep learning, so-called deep reinforcement learning systems (cf. Bu chanan et al. 2017: 6f f , Murphy 2012: 32; Piovesan/ Ntiri 2018: If.). This combination is a powerful set of techniques used to generate control and action systems whereby autonomous agents are trained to take ac tions given an environment state to maximize future rewards. Though nascent, recent advances within this area are impressive. In addition to its recent victories in the game of Go, the software company Google DeepMind has achieved superhuman performance in several Atari games (cf. Fortunato et al. 2017). These short mentioned successful use-cases are notable technological milestones. Amongst other use-cases they have the ability to change the economic landscape, creating new opportunities for business value creation and cost reduction (cf. Esteva et al. 2017; Brynjolfsson et al. 2017: 3). Intending to the potential of AI, in combination with recent improve ments in big data, cloud or connected devices and the support of possible future technologies, such as quantum computers, this paradigm shift could be further stimulated. By that it is important to realize that the ear ly leading adaptors of these emerging technological possibilities in the field of AI also take responsibility for their actions, even to make sure that everybody has a common idea what AI means exactly for individu als, organizations, companies, and societies. The following chapter is intended to define artificial intelligence with the help of the definition of intelligence. After that, a differentiation of AI into weak and strong AI will be introduced. Then, a definition of AI for the following research should be made. The last part of this chap ter will look at the current state of deployment of AI. It will be discussed that economic actors actively and strongly already pursue the introduc tion of AI, seeking efficiency gains and higher profit margins. 2.1. Intelligence and Artificial Intelligence Starting with a definition attempt of artificial intelligence, it helps to dif ferentiate AI from the basic concept of intelligence. So far, there are sev 7 eral different types of definitions, because intelligence exists at different levels and there is no consensus on how to distinguish them precisely. However, similarities can be detected under different definition attempts. In essence, it describes a general mental ability that includes the ability to discern rules and reasons, think abstractly, learn from experience, de velop complex ideas, plan and solve problems. Artificial intelligence, in turn, is supposed to reproduce the aspects as mentioned earlier of human behavior to be able to act humanly in this way, without being part of a sentient organism. Thus it is intelligence that is artificially made. It in cludes qualities and abilities such as solving problems, explaining, learn ing, understanding speech as well as the flexible reactions of a human being (cf. Gentsch 2018: 17; Marwala/ Hurwitz 2017: 9). Since the mid dle of the 20th century, AI forms a field within computer science. The aim is to find methods by which human abilities such as the conclusion of experts, mathematical proofing, the recognition of images, the under standing of the natural language or the targeted optimization in any envi ronment can be simulated on computers. These systems should be adap tive — for example, they could query an unknown fact to the user and to save the gained information for reuse. The stored programs and data can be represented in the form of rules (cf. Wedde et al. 1990: 280). Since it is not possible to find the only one universal definition of Artificial Intelligence, the following definition by Elaine Rich seems to be the most appropriate for this research contribution: "Artificial Intelligence is the study of how to make computers do things at which, at the moment, people are better.” (Ertel 2016: 2). It briefly characterizes what scientists have been doing in the field of AI for the last fifty years and what they will do in the future. Humans are still more capable and suitable in most of the fields, but computers and algorithms already offer enormous advantages. Their ability being to dominate more fields in the existing society will increase. Further abili ties of AI could evolve exponentially (cf. De Waele 2016: 1). Other sci entists say that AI is more like a theory and development of computer systems able to perform tasks usually requiring human intelligence. Al ternatively, AI is an activity devoted to making machines intelligent, and intelligence is that quality that enables an entity to function appropriately and with foresight in its environment. 8 2.2. Different Types ofA l Overall, AI is a sophisticated technology with an equally complex conceptional understanding. There is no universal definition of the term it self. In part, determining what AI is, it is only possible in its context. From a technical point of view, the types of AI are also enormously con troversial and cannot be differentiated at first glance. For this purpose, the specific artificial intelligence must be closely examined and tested for their abilities. So far, extensive tests focus on comparing human intelligence with arti ficial intelligence using, for example, Alan Turing's so-called Turning Test, widely used since 1950. He tests a machine's ability to exhibit in telligent behavior equivalent to, or indistinguishable from, that of a hu man (cf. Puget 2017). An established AI Index or different types of distinguishing levels of AI itself is underrepresented. There are no clear procedures that could dif ferentiate several AI algorithms with the help of specific scalability. In order not to place the previously outlined technical complexity of today's AI systems in the focus of this chapter, the various differentia tion and types of AI will be fundamentally differentiated into two types for this work: Weak and strong AI. 2.2.1. Weak AI vs. Strong AI Weak AI or rather narrow AI describes an interpretation of AI according to which conscious awareness is a property of specific brain processes and whereas any physical behavior can, in principle at least, be simulat ed by a computer using purely computational procedures; computational simulation cannot in itself evoke conscious awareness (cf. Colman 2015). According to this view, the weak artificial intelligence is usually entrusted with concrete application problems, which do not depend on logical thinking, decision making or even not based on consciousness. It primarily serves humans as an information provider on which it bases their decisions. These include, for example, the following areas of re sponsibility: Expert systems, navigation systems, voice recognition or the automatic correction suggestions for digital search functions (cf. van der Touw 2016). Some common examples of weak AI are in consumer applications such as Apples Siri or Google Maps. It is combining several narrow AI techniques plus access to extensive data in a cloud (cf. 9 Greenwald 2011). In shorthand, weak Al can be described as: „devices and applications that do specialist tasks much better than we ourselves could do them, mostly because they number-crunch in ways we can't.“ (Souter 2018: 1). However, it is still supervised programming, which means there is a programmed output or action for given inputs. So weak Al might behave like a robot or manufacturing line is thinking on its own. Compared with that, Strong Al or Artificial General Intelligence (AGI) is a more complex approach that might change output based on given goals and input data. “An interpretation o f artificial intelligence according to which all thinking is computation, from which it follows that conscious thought can be explained in terms o f computational principles, and that feelings o f conscious awareness are evoked merely by certain computations carried out by the brain or (in principles at least) by a computer.” (Colman 2015b). So a program could do something that it was not programmed to do when it detects a pattern and determines a more efficient way to reaching the goal it was given (cf. Kerns 2017: 2). The strong Al wants to create the most powerful computer systems imaginable. However, the utmost efficiency imaginable is the reproduction of human intelligence. In this way, strong Al can be described as the highest level of sapience or as machines with human-like intelligence and beyond (cf. Sarkar 2018: 1; Alpcan 2017: 2). „Strong“ is not this output of strong Al, what it already has created, which is in their own perspective rather "weak", but what it promises to deliver (Sesink 2012: 3f.). Regardless of this, one often speaks of strong artificial intelligence as when a machine has the same intellectual abilities as a human or even surpasses it. A perception like this would also mean that no longer only people are competing for a par ticular job with each other, but also the with strong Al itself as if it were a human worker. Current examples for the early beginning of strong Al could be seen in deep reinforcement systems playing chess or go, soar cognitive architec ture, autonomous vehicles or Al systems like Google Duplex. However, strong Al has not achieved his potential yet. Based on Al and its often quoted exponential character it can be assumed that strong Al might be relevant in the near future. However, it is still controversially discussed. While some argue the opposite that the full potential of strong Al will be a long-term achievement, others argue that it will never be reached. 10 Based on a few studies, some scholars predict that computers will reach human intelligence around 2029, because AI is an exponentially increasing technology and undergoing a massive acceleration driven by an immense growth especially in available data and the rapid evolution of algorithms. The full potential of strong AI, so-called real singularity or superintelligence of machines when they are more intelligent than humans at all, will come by 2045 (cf. Makridakis 2017: I lf ; Kurzweil 2005). Furthermore, other scholars asked 60 experts of an Artificial General Intelligence Conference (2011) to answer the question, if they believe that AGI will be effectively implemented in the following timeframe. Of the experts surveyed, 43.3 percent estimated that it would be before the year 2030, 25 percent estimated between 2030 and 2049, 20 percent said between 2050 and 2099, 20 percent said after 2100, and 1.7 percent said never. In result, more than two-thirds of respondents predicted that AGI would occur before 2050. A more recent survey con ducted in 2016 by Etzioni et al. was related to superintelligence, as an often mentioned ability of strong AI. Etzioni's question was based on Nick Bostroms book, which defined superintelligence as an intellect that is much smarter than the best human brains in practically every field, including scientific creativity, general wisdom, and social skills. Etzioni asked the experts when they think superintelligence could be a reality. The answers of the 80 responders are summarized as follows: Nobody thought in the next ten years, 7.5 percent estimated between the next ten and 25 years, 67.5 percent in more than 25 years, 25.0 percent decided to say never. These results put the start of superintelligence later than that of Kurzweil survey but still a realistic scenario (cf. Makridakis 2017: Ilf.). To return the perspective of this work and a possible qualitative investigation of strong AI, all in all, there are five areas that are often associated with strong AI: Awareness, self-perception, sensibility, wis dom, and feelings. However, it is still unclear to what extent these capa bilities are interrelated. For example, if awareness is necessary to think logically (cf. van der Touw 2016). Based on these basic descriptions strong AI should be seen in the course of this research with the following characteristics: Logical thinking, nat ural language, self-learning, self-structured planning and decision making in uncertainty. 11 In conclusion, there are significant differences between the two types of AI. The term Strong AI refers to the type of AI that has specific capabilities similar to those of human intelligence. This concept differs significantly from the existing computing systems and mainstream AI algorithms, already described as weak AI. Weak AI-based systems are successful in solving specific problems when the problem context is pro vided directly by a human programmer. In contrast, strong AI is much deeper and broader in its ambition as it aims for human-like flexibility, understanding, and creativity (cf. Alpcan 2017: If.). 2.2.2. Al Definition for the Analysis The following part of this research will focus on the early form of strong AI. Although strong AI is not being used today as predicted by its poten tial, some scholars expect that the necessary technical developments will evolve rapidly in the next years. By looking into empiricism, tendencies and the early beginning of strong AI could already be identified. Exam ples are deep reinforcement learning systems winning in chess or go against the human world champions, AI systems like Google Duplex or autonomous vehicles. Nevertheless, against the background of the criti cal literature as well as the current technical limitations of strong AI, a weak form of strong AI should be used for the analysis of this research. The core of this definition will thus be formed concerning the previous chapters and by its current state of the art. So the basic understanding of strong AI for the following analysis within this research will be de scribed as follows: In general, build on weak AI because it is still working in the economy. But the AI system has also be described at least with one of the following characteristics of strong AI: Logical thinking, natural lan guage, self-learning, self-structured planning and decision-making in uncertainty. Thus, the definition would theoretically be located between weak and strong AI, but by describing it with one of the mentioned char acteristics should mean the system is already an early stage of strong AI. The more of the mentioned characteristics would be fulfilled the more advanced would the strong AI become. For example, deep reinforcement learning systems winning in chess or go against the human world champions, because of such increasing suc cesses, at least the self-learning characteristic would be evaluated as re alistic. Looking at systems like Google Duplex, the natural language 12 characteristic could be considered as given. In the case of autonomous vehicles, however, several properties of strong AI could already be con sidered. Both self-learning, self-structured planning and decision-making in uncertainty would be conceivable. Thus, autonomous vehicles would undoubtedly be a showcase example of strong AI within the mobility sector. Based on these examples, it can also be argued that they already go well beyond the complexity measure of weak AI and can, therefore, be seen as the early form of strong AI. 2.3. Place Value o fA l in the Economy Andrew Ng, Professor of Stanford University, coined the phrase that AI seems to be the new electricity. From his perspective, AI will revolution ize all industries of the economy in a similar way to the electrification process of the world (cf. Tomer 2017; AI100 2017: 54). This metaphor is symbolic of current developments and potentials in the field of AI. Arti ficial Intelligence is undergoing a massive acceleration driven by an im mense growth especially in available data and the rapid evolution of al gorithms. The following chart shows most of the different industry sectors. It measures, on the one hand, the current AI adoption, and on the other hand, the future AI demand trajectory: Fu tu re A I d e m a n d tra je c to ry 1 A ve ra ge es tim a te d % cha n g e in A I spe nd ing , n e x t 3 y ea rs , w e ig h te d b y f irm s ize2 13 12 11 10 9 8 7 6 5 4 3 2 1 0 0 2 4 6 8 10 1 2 14 16 18 2 0 2 2 2 4 26 2 8 30 32 C u rre n t A I ad o p tio n % o f f irm s a d o p tin g o n e o r m o re A I te c h n o lo g y a t sca le o r in a co re pa rt o f th e ir B u s iness, w e ig h te d b y firm size® 1 Based on the midpoint of the range selected by the survey respondent. 2 Results are weighted by firm size. See Appendix B for an explanation of the weighting methodology. Figure I: Sectors leading in A I adoption today also intend to grow their invest ment the most. Source:McKinsey 2017:19. T rave l an d to u rism « H ea lth ca re • • P ro fess ion a l se rv ices R e ta il • C o n su m e r p a cka g e d g o o d s • • E d uca tion * C on s truc tio n L e a d in g s e c to rs F ina nc ia l s e rv ice s • H ig h te c h a nd te le com m un ica tio ns * T ransp o rta tion a n d log is tics A u to m o tive an d asse m b ly E n e rg y an d re so u rce s • ^ • M ed ia an d en te rta inm en t F a l l in g b e h in d 13 Financial services, high tech and telecommunications are the leading ear ly adopters of Artificial Intelligence and will be the leading industries that adopt AI in the next two years. In addition to this chart, it is also noteworthy that financial services1 and healthcare2 seeing the highest increase in their profit margins as a result of AI adoption (cf. Forbes 2017). Within ten years AI and robotics are expected to create an esti mated annual so-called creative disruption impact of up to 33 trillion dollars globally. It includes eight trillion to nine trillion of cost reduc tions across manufacturing and healthcare, nine trillion in cuts to em ployment costs due to AI-enabled automation of knowledge work and 1.9 trillion in efficiency gains via autonomous cars and drones (cf. Me dium 2017). It is predicted that AI may power the next generation of ef ficiency tools (cf. Motte 2017). The leading technology firms are in a race to build the best AI and capture a massive market. Thus IBM is working on its Watson, Amazon is banking on Alexa, Apple has Siri. At the same time Google, Facebook, and Microsoft are devoting their research labs to AI and robotics. All of these companies show their public attitude and are highlighting their enthusiasm for Artificial Intelligence (cf. Kelnar 2016; Bajpai 2017; The Guardian 2017; Maney 2016: 1). At the same time, invest ments in the field of AI are rising considerably. For example tech giants including Baidu and Google spent between 20 billion to 30 billion dol lars on Artificial Intelligence in 2016 alone. It includes 90 percent on R&D and deployment, ten percent on Artificial Intelligence acquisitions (cf. Data Collective 2017; Medium 2017; Forbes 2017). “The last 10 years have been about building a world that is mobilefirst. ( ...) In the next 10 years, we will shift to a world that is AIfirst.” (Rieber 2017: 13). Practitioners place a high expected future value on Artificial Intelli gence. The current age of Artificial Intelligence has the potential to dis rupt the world’s largest industries and is expected to contribute 15,7 tril lion us dollar to the global economy by 2030. That is one reason why AI is often called a new wave of innovation or the world’s next industrial revolution (cf. Hindu Business Line 2017; Piovesan/ Ntiri 2018; Baron 2016: 1). AI and the widespread adoption of cognitive systems across a 1 AI and advances in robotics will likely disrupt millions o f workers within these sectors globally, generating up to 50 percent in productivity gains. 2 AI-based services will play a growing role in automated support, diagnosis, and advice in healthcare (cf. Medium 2017). 14 broad range o f industries will drive worldwide revenues from nearly 8 billion dollars in 2016 to more than 47 billion dollars in 2020 with bank ing named as one of the top two industries to lead the charge (cf. IDC 2016). This estimate and assumption based on several parallel develop ments, which will not be discussed in detail in this research. However, to name a few, the following illustration helps. Enormous Growth in Data Availability of Cheaper Smarter Algorithms Cloud Computing Connected Devices (loT) Faster Internet (Big Data) Computing Resources Architecture Figure II: Infrastructure Development. Source: Own graphic. These technological achievements, increasingly powerful hardware, evolving machine learning approaches, and vast new data sets are fuel ing a transformation of these major global industries such as financial services, healthcare, retail, education, manufacturing and supply chain, mobility or the security sector (cf. Data Collective 2017; Medium 2017). All in all, the potential of Al in the business sector can be seen in the increase in profit margin and productivity. Furthermore, Al is seen as a productivity-enhancing factor. A Bank of America Merrill Lynch report predicted that adoption of robots and Al could boost productivity by 30 percent in many industries while cutting manufacturing costs by 18 to 33 percent (cf. Medium 2017). By that, the McKinsey found out: “companies who benefit from senior management support for Al in itiatives have invested in infrastructure to support its scale and have clear business goals achieve 3 to 15 percentage point higher profit margin.” (Forbes 2017). In addition, the following figure shows the economic impact of artificial Al in developed countries that currently make up almost 50 percent of global gross domestic product (cf. Accenture 2016). More precisely it estimates the increase in labor productivity with Al, more precisely the percentage difference between a baseline without Al improvements and an Al steady state in 2035: 15 Country Percentage Increase in Labor Productivity with AI Sweden + 37% j | Finland + 36% § H § USA + 35% £ | Japan + 34% ^ Austria + 30% Germany + 29% Netherlands + 27% g l i UK + 25% France + 20% Figure III: Projected Increase in Labor Productivity with AI. Source: Own graphic. This comparative country illustration, based on analysis by Accenture and Frontier Economics, underlines that AI can be seen as a robust productivity-enhancing factor (cf. Medium 2017). With much of today's businesses making use o f AI for the task of increasing their productivity, even a new job title of a “Chief AI Officer” has been introduced to man age the use of AI in the most efficient way (cf. Forbes 2017b; Schrage 2017: 1). In this context, it underlines the importance of using AI as a productivity-enhancing technology with optimal impacts in each compa ny. In the healthcare sector, an architecture using deep neural networks was tested against 21 board-certified dermatologists and matched their per formance in diagnosing skin cancer (cf. Esteva et al. 2017). In the soft ware sector, the social networking company Facebook uses their neural networks for more than 4.5 billion translations every day. For example, image recognition is one reason amongst others, why an increasing number of companies have responded to these opportunities (cf. Brynjolfsson et al. 2017: 3). Another common use-cases are natural lan guage processing, which refers to a system’s ability to process and un derstand human language to convert it into full representations. It holds enormous benefits as well and is for example used in chatbots, real-time text translation or new systems like google duplex (cf. Murphy 2012: 32; Piovesan/Ntiri2018: If.). In conclusion, it can be assumed that AI has a revolutionizing im pact. However, the whole development of AI in the economy and this desired increase in productivity and profit margin of the companies has far-reaching consequences, which will be analyzed in a later part of this research. 16

Chapter Preview



The ongoing and seemingly unstoppable digital transformation brings forth new options, opportunities but also challenges to individuals, organizations, companies and societies alike. Governments are alarmed, realizing the potential consequences on the workforce, while also being apparently helpless against uncontrollable and powerful digital players such as Google or Facebook. As Henry Wittke shows, recent breakthroughs in the field of machine learning increase the potential of Artificial Intelligence to disrupt the world’s largest industries. Wittke attempts to provide a basic framework of what constitutes AI as well as to assess its impact on the Information Economy. What happens in case of rising mass unemployment or social inequality? What will be the effect on labor as a value system for today’s societies? Could the entire notion of capitalism be questioned in the wake of AI? The book aims to draw conclusions and give recommendations to policymakers.