The implementation of artificial intelligence from an economic perspective and its social-ethic challenges

Introduction

"Can machines think?" (Turing, 1950). With these pioneering words, and the questioning in the so-called imitation game whether it enables machines to think autonomously, Alan M. Turing developed one of the first fundamental ideas of machine learning and its significance for the future as early as the 20th century. Machine learning (ML) or also known as Artificial Intelligence (Al) is now no longer a dystopia that was only initially considered, it has evolved and is a tangible reality and immeasurable in all aspects of life. Due to the fundamental paradigm shift, massively marked by the emergence of digitalization and Industry 4.0, artificial intelligence, which is being used almost across the board, will play a key role in the coming decades. The applicability of artificial intelligence and increasingly complex algorithms continues to shape our everyday lives, from smartphones and search engines to the financial and healthcare sectors. Beyond this, the use of machine learning has a major impact on the economy and its future significant societal implications. Al is a highly topical issue because, despite promising technological progress, the use of Al is often controversial, especially with regard to whether the implementation of such technology is morally and ethnically acceptable and whether certain legal and normative boundaries are crossed. Therefore, the aim of this research paper is to analyze how far the implementation of artificial intelligence will continue to have a decisive impact on economic events and cause change. Irrespective of this, it is also important to analyze the social-ethnic effects of the use of Al.

Defining Al

Defining the term "artificial intelligence" precisely is still unclear, as there is no uniform explanation of the term due to the many different areas of application of Al, and it is also disputed how to define the word intelligence itself. (Buxmann & Schmidt, 2019, p.6) An authoritative definition that could endure for decades to come is that of Elaine Rich, whom she describes as "Artificial Intelligence is the study of how to make computers do things at which, at the moment, people are better. " Conversely, this definition makes it clear that machine information processing cannot exist without human cognitive thinking, which is why the boundaries between Al and neuroscience are becoming increasingly blurred these days. (Ertel, 2017, p. 2, as cited in Rich, 1983) The term artificial intelligence can therefore no longer be categorized in any precise field of research. Yet, artificial intelligence encompasses the field of computer science and aims to analyze problem findings with stable data sets. Moreover, the term Al is often equated with machine learning and deep learning, which in turn are a subfield of Al algorithms. (IBM Cloud Education, 2020). In particular, machine learning learns and makes decisions based on data that is already available, and when so-called neural networks are added, this is called deep learning. Deep Learning combines Al and Big Data (Menzel & Winkler, 2018) Symbolic Al follows a phenomenological approach of human behavior, i.e., the knowledge and behavior of humans are established in the intelligent systems, the learning process and the logic is very apparent. The approach was considered promising for a while, but neural Al is on the rise. The neural Al or neural networks manage in contrast to the symbolic large and unstructured data to evaluate, as well as precise clustering to operate, thus intricate processes can be mastered. (Dickson, 2019) Not all artificial intelligence is the same, there are key distinctions between weak and strong intelligence. Basically, an Al is described as weak which, with the help of a targeted algorithm, takes on individual tasks and separate problems. (James, 2019) Today, we are still in the realm of weak Al, because systems like voice assistants, facial recognition systems or even autonomous vehicles fall under the category of weak Al. (Labbe & Wigmore, n.d.) In contrast to weak Al, strong Al is still a theory of the future, as it would mean that strong Al would have the full and congruent cognitive abilities of a human. (IBM Cloud Education, 2020) The strong Al knows how to perfectly imitate the human being with all his abilities. In addition, the strong Al should go so far as to analyze knowledge transfer, future predictions based on previous knowledge, as well as adaptability in changing situations in order to provide solutions. (James, 2019) How questionable and fragmentary a super intelligence can be was already shown by an experiment from 2016:

Microsoft developed the Al chatbot Tay, which was placed on Twitter and had the task of learning from young people how they communicate. However, the premise here was that the bot had to be taken offline after only 24 hours because the bot was making Hitler comparisons, sexist remarks and Holocaust denials. (Hunt, 2016)

Applications of Al

The application of artificial intelligence is considerable and steadily increasing. Al is becoming commonplace. The most common keywords associated with Al are robotics or autonomous driving. However, sectors such as healthcare could benefit from the technology by improving diagnostics and optimizing medical interventions. Al could also contribute to sustainable optimization in the agricultural food industry, by analyzing data to minimize the use of fertilizer or water. Even in administrations and municipalities, Al could act as an early warning system of disasters. (Europäisches Parliament, 2020) The use of intelligent technology is also finding its way into industries and services. According to McKinsey (2020), Al is used significantly in product development and marketing, followed by personnel management and manufacturing. Figures also reflect usage, with 50% stating they have de-embedded Al in one of their business functions, and 22% of respondents reported that the use of Al accounted for 5% of their EBIT margin. (McKinsey, 2020)

Effect of Al on the Economy

The advancement of Al-based technologies is not diminishing, but accelerating, if not exponentially. In economic terms, Al is associated with the term general purpose technology, which can be understood as "there are a handful of "generic", or "general purpose" technologies (GPT' s) characterized by their pervasiveness (i.e., they can be used as inputs in a wide range of downstream sectors), and by their technological dynamism." (Bresnahan & Trajtenberg, 1992) As a result of the dramatic acceleration of Al, the implementation is also reflected in particular in the increase in financial resources. As a result of the dramatic acceleration of Al, the implementation is also reflected in particular in the increase in financial resources. Thus, the global venture capital of Al start-ups rose from 3 billion US dollars in 2012 to 75 billion US dollars in 2020. About 80% of these investments in 2020 are from China and the US, compared to only 4% from the EU. The majority of investments were made in the automotive and transportation sectors, with China and the United States accounting for almost 98%.(OCED, 2021) According to one estimate, the implementation of Al could increase the value of global GDP to $15.7 trillion by 2030. (Firth- Butterfield et al., 2022) In particular, China's immense investments in Al technologies are remarkable. China's acceleration in Al technology is not just coming from nowhere, its regulations on data protection are weak compared to the European market, and the use of Al-based facial recognition programs has accelerated the Chinese market. ( Li et al., 2021) In Germany, it is assumed that only the adoption of Al technologies could increase the GDP by 11.3% by 2030, which corresponds to 430 billion euros. (PWC, n.d.) Apart from the high investments made by strong economic nations, the impact of Al on economic growth is debated repeatedly. The assumption is that there are two main growth factors, one is direct GDP growth from companies producing Al technology and the other is indirect GDP growth from sectors using Al to increase productivity. (Chen et al., 2016) In the 1980s, the well-known economist Robert Solow put forward the thesis of the productivity paradox. He argued that the age of information technology has no discernible connection with a country's growth, but rather leads to stagnation. (Dudley, n.d.) Solow's assumption may be confirmed, because the OCED also declares a decline in productivity in the past years. The reasons for a slowdown in productivity growth are more far-reaching, because on the one hand the effects of the financial crisis in 2008 are still having an impact, and on the other hand the structural shift of the individual sectors influences growth, yet with the emergence of the first commercially available computers at the end of the 1990s, the growth then associated with them was declared insufficient. This in turn reflects Solow's assumption of the complementarity of new technologies and productivity growth. (OCED economic outlook, 2019) From a macroeconomic perspective, artificial intelligence could be another link in the so-called production function. (Vöpel, 2018). The production function, which is composed of human capital, real capital, natural resources and technological knowledge, describes productivity. (Mankiw & Taylor, 2016, p.679 - 681) Artificial intelligence increases the substitution function to human capital, because the knowledge of the Al and its predicted abilities eliminate the knowledge and qualification of the individuum. Due to the learning algorithms, the knowledge of the Al is unbound, which illustrates the contrast to the bound human capital, which requires individual learning and qualification. (Vöpel, 2018) If Al is included in the production function in combination with Big Data, it would certainly be economically advantageous because returns could also be skimmed off with it. (Schneider, 2021) Moreover, economists such as Brynjolfsson argue that, despite all counterarguments, Al offers a way out of the productivity paradox in the long run, precisely because Al acts as an additional intangible capital input. Hence, the assumption that the handling of education and labor with respect to Al will no longer appear to be efficient in the future and conventional measurements such as GDP will need to be adjusted. (Brynjolfsson, et al., 2017) Furthermore, which is considered an economic and social controversy, is the question of how the advancement of new technologies and the general aspect of automation endangers jobs. From an economic point of view, it is very difficult to make a clear statement here, due to several factors such as the rapid progress and changes in Al-based technologies, as the algorithms and data volumes become increasingly powerful and large. (Buxmann & Schmidt, 2019, p. 30). According to a study by the OCED (2021), only 14% of jobs are currently affected by automation, which is also in contrast to the 12% increase in the labor market between 2012 and 2019. Moreover, people with lower qualifications tend to be more affected by automation than those with higher qualifications. (OCED, 2021) It is also estimated that about 85 million jobs will be replaced by the year 2025 due to the shift in the labor market, but in turn 67 million new jobs will be created through the interaction of humans and the algorithm. (World Economic Forum, 2020) The insertion of Al into the production function shifts the relationship between labor and capital, consequently weakening work with low skills, an example here would be the transport sector, because the capital is there, but a significant reduction in labor. A capital less market entry of new competitors would be more efficient, because the data do not depend on capital, market entry barriers are lowered. (Vöpel, 2018) The adaptation of Al is synonymous with efficiency and cost savings in a wide range of industries. In fact, the use of Al is creating new economic orders, not only in the entirely discussed showdown between machines and humans on the labor market, but also in a wide range of industries. Economically, Al can be an asset and a driver of growth, labor and capital, but any innovation is fraught with controversy, especially how to account for the adaptation of Al ethnically and socially.

Social Challenges of Al

Due to the dynamic nature of Al technology, the controversy surrounding Al must be examined from a social perspective. Although Al will only lead to a deferral of work and not to outright mass unemployment, the social controversy surrounding technology and work centers on the fact that it fosters the growth of social inequality. Thus, according to a scientific study, Al in the U.S. is responsible for a reduction in wages of about 50% to 70%. The increasing inequality in wealth and income is therefore brought with the automation, because while the salary of people with a higher degree increased, in contrast, the salary of less qualified people decreased. (Kelly, 2021) In the context of the inequality debate, human enhancement technologies play an essential role. The assumption here is that intelligence is measurable. People with a higher income are measured with a higher intelligence, a higher productivity, again the resulting dilemma is that the rest with a lower income represents the opposite. Ultimately, incentives are created for the solvent part of society, improvements can be easily bought, moreover, they are in competition with other wealthy people. The fact is, there is a dependency between income and well-being. The majority of society with low incomes who do not have access to the latest technologies are left behind. As a result of the increasing dynamic of innovation growth, a void is emerging between the best technologies and what is publicly obtainable for the poorer part of the population. (Korinek & Stiglitz, 2017) Moreover, Al or rather the sub-discipline Machine Learning can contribute to decision-making, but there is a risk that decisions will be cognitively biased. (Silberg & Manyika, 2019) As a result, the bias of the algorithm leads to major misunderstandings and even discrimination. Thus, during the ML learning phase, certain and already existing social biases are reinforced and even replicated. (Garcia-Gathright, et. al., 2018) In the past, large companies such as Amazon.com used Al algorithms in the recruitment process to make the review of individual CVs more efficient, but the algorithm caused a bias where women were treated less favorably in the selection process. (Dilmegani, 2022) In addition, as with any new technology, there is of course the question of the extent to which transparency and data protection are affected. To ensure safe handling, some countries, including the European Union (2021), have already taken legislative measures with reference to the emerging risks of Al. The framework is intended to be compatible with EU fundamental rights and impacts both individuals and organizations. However, the current non-transparency, otherwise known as the black box of Al, could further fuel the criticism, because the so-called black box paradox is understood as the impenetrable system of any Al, in short, inputs and executions are not visible. Due to this impenetrability of Al, biases also arise. (Caussauwers, 2020) The extent to which Al affects privacy is also still controversial. China uses Al to analyze the data collected from its citizens and ultimately to build a rating system for their behavior. Here, transparency around Al receives a completely new dimension (Buxmann, 2018) Al literally embodies the keywords transparency and above all privacy, so one often speaks of the so- called Al dilemma. The dilemma is that Al feeds large amounts of data and the larger the amount of data, the greater the risk of misuse. The original purposes are disregarded, data could be sold to third parties or stolen. (Sträter & Lundbaek, 2021) Thus it is already in such a way that for the personnel selection algorithms are used which act after the Black box principle, like which selection criteria were met, remain thus obscure. (Buxmann, 2018) Weak Als such as smart home devices like Alexa or Google come under criticism frequently. It has long been known that voice assistants record what their users say, some of it confidential, which is then stored in a cloud. Therefore, it happens that in more than one out of ten protocols, Alexa accidentally went on.

[...]