Introduction
The general consensus is that technological development has increased living standards globally. We are living historical times; never before has so few people lived under extreme poverty (26). It is almost miraculous considering the global growth of human population. Improvements in production efficiency that have resulted from new technologies have enabled the positive global progress.
One reason for the positive correlation between improved global prosperity is the usage of human labour. Capitalism (maximizing profits) and human needs, such as happiness and engaging relationships with others, are well aligned, as humans’ work performance is related to their wellbeing (2, 25). Especially companies whose business require solving difficult problems are increasingly focusing on employee satisfaction. In the future, those fundamental premises might not be as aligned as before.
Artificial intelligence’s importance is steadily growing in the economy. It is one of the key technologies offering competitive advantage to businesses who use it efficiently. The hype is very strong and companies are increasingly investing into AI and attracting research talent to join their ranks (9). As a result, the presence of algorithms in society is constantly increasing. Judging by the trend, we can argue with high confidence that AI is going to be one of the key drivers of economic development in the future.
Even when AI is used to refer to narrow intelligent agents capable of learning instead of self-aware intelligent agents capable of making independent decisions (Searle, 1980), it is probable that AI is going to be one of the key drivers of economic development in the future. Machine learning has been an important part of the latest AI development and the approach has been achieving learning and performance results, which were estimated to be impossible or far away to be achieved (24). Hence, we refer to the time horizon of the next few decades as the machine learning era.
The machine learning era can have many positive effects to humans and societies (24). However, the rise of artificial intelligence can also fuel unbalanced and unequal economic development. Several authors have discussed the topic, and especially the discussion about general intelligence has fuelled strong arguments for beneficial AI development (21). This view often omits short term consequences of the current technology trend.
It seems probable that during the machine learning era the most of the capital and income is concentrated to those who managed to first use artificial intelligence to get ahead of the competition. Many jobs are susceptible to computerization, especially low-skill and low-wage jobs could disappear already during the machine learning era (8).
In this article we will focus on three possible effects from the development of AI. These include data-driven centralization of businesses, the share between labour and capital, and economic impacts of algorithm biases. The goal of this article is to discuss their economic impact on the developed societies.
Data-driven centralization of businesses
Developing artificial intelligence requires processing power, data and tools. Arguably, processing power and computer capacity are rather cheap today. Similarly, most of the machine learning tools are open source. But the real asset, data, is increasingly concentrating into the hands of few major tech-giants; namely Google, Amazon and Facebook. The phrase “data is the new oil” was coined already in 2006 by Clive Humby (16), and it is today more true than ever (24). The importance of data only keeps on growing as more sophisticated algorithms are developed.
Data quality is one major factor in machine learning. Many organizations possess massive amounts of data sitting in data warehouses. However, it might not be very useful. Another classic phrase, “garbage in, garbage out”, catches the essence of data’s role in machine learning. The organizations who have patiently designed their data collection in a way that it is usable in machine learning are the winners of the machine learning era, as gathering useful datasets could take years.
Data is a very thought-provoking factor of production. Unlike other raw materials; it does not wear out. It can actually accumulate, as more information can be linked to the old datasets. The existing macroeconomic theories have hard time taking such a commodity into account, and this far we have not found any convincing models explaining the dynamics.
At the same time, the tech-giants have used digitalization to create new business models which massively benefit from economies of scale. They have become worrying monopolies in their own markets that lock users to their platforms. The network effects they have generated represent a form of demand side economies of scale, which means that each user makes the platform more attractive for new users. The success of their models are huge, as Google controls 90% of online searchers, Amazon 50% of digital sales and Facebook has soon 2 billion users (7).
The production costs are much lower than for the traditional firm as they need to employ a lot less workforce. The tech-giants already generate over three times more revenue per employee than the traditional companies and have ten times higher market capitalization (14).
Monopolies tend to stagnate and dissolve in face of competition. One theory is that as the amount of workforce grows, the communication gets too complex and it becomes impossible to steer the company enough quickly. The tech-giants are hiring new talent eagerly, but they are not dependent on the amount of employees similarly as traditional companies. Providing digital products, their customers are mainly interacting with software and algorithms. Therefore, they are not as subjective to the growing complexity. From production cost perspective, their biggest concern ought to be ever in-creasing complexity in their digital architecture. However, it seems that they are capable of handling it.
Establishing their position as tech-giants that defy the constraints of traditional businesses, these companies have extremely strong position in the machine learning era. Adding the amount of user data these companies have hoarded makes their status as tech-giants even stronger. Data being the most important asset in the near future, it is not farfetched to predict that these companies will keep on growing into unpreceded concentrations of equity.
The share between labour and capital
There is a wide discussion on AI and the future of work. It is unclear, whether machine learning and robotics are replacing or complementing human workers (23). Kim et al. (12) analysed that technological progress has had negative effect on employment during recent years. However, governmental interventions and policy changes might be effective in preventing similar development in future. Probably appreciable proportion of total future employment will consist of new technical jobs. Frey and Osborne (8) estimate that 47 % of total US employment is in high risk of being automated within next two decades. This includes large amount of occupations in the field of transportation, logistics, and production. It is also possible that service robots can replace human workers in certain tasks in the service industry. However, humans will probably continue to work in occupations, which need high level of social intelligence and creativity. Within these occupations machine learning will probably have complementary effect.
It is already possible to study how machine learning is changing the nature of work. In certain fields of business, such as transportation, digital platforms are largely responsible for matching service providers with the people who need services. For example, Uber is offering a digital platform for ridesharing which aims to connect drivers with people who need a ride (19). Unlike traditional taxi services, Uber does not employee drivers directly, instead the drivers are classified as independent contractors. For this reason every driver has possibility to decide themselves when and how much they want to work. At the same time they do not have same rights than drivers who are employed by a company offering rides for passengers. They are personally responsible for insurances, licences and maintenance of the vehicle. The quality of their work performance is automatically monitored by analysing the ratings given by the passengers, and if the rating is dropped too low, the driver will be deactivated from the Uber application. This actually means that the driver is fired even if they are not entitled to any redundancy pay. Similar developments have been detected in different sectors, such as food service and retail, where standard sifts are replaced by sifts scheduled on demand based on algorithmic predictions (5).
If many jobs will disappear within next two decades and working conditions within many occupations will decrease, it is likely that polarization of developed societies will continue. Future scenarios (e.g. 18) of societies with high unemployment, low paid service jobs and well paid knowledge workers seem now more and more probable. Such development would probably decrease the labour’s share of national and global income.
Economic impacts of algorithm biases
Machine learning and the algorithms do not only affect businesses, but also the everyday life of people living in modern societies. One reason for this is that algorithms are imperfect. Programming fairness to algorithms is difficult. Algorithms are not biased per se, but they are unfair because they judge individuals based on reference groups behaviour, while the standard way to perceive fairness would be judgement by individuals own actions. (6). The kind of algorithmic unfairness may have deep consequences from societal perspectives, as people in poverty are deemed to pay higher price for goods and services (even redlining) because of their reference groups higher default risk. (15) The lack of fairness feeds the vicious cycle and preserves the unfair statistical reality.
The phenomenon of algorithmic bias is built on forms of discrimination and has clearly negative consequences from societies’ perspective. Algorithmic unfairness has several ethical and social causes, which can affect great harm for minority groups and selected subpopulations. The phenomena is fairly recognized, but especially concrete and practical approaches to the economic impacts is missing. (5)
From the earlier academic discussions, it is possible to present algorithmic bias divided to four categories based on the underlying reasons where the bias is formed: 1) technical failure 2) worldview of the developer 3) research data biases 4) statistical based discrimination. Different forms of bias differs from each other based on the level of intentionality, scope and difficulty to solve. (5)
In addition to the social and ethical impact, algorithmic discrimination can have heavy economic impacts while algorithms are used for allocating private and public goods. Economic impacts of algorithmic unfairness can theoretically traced back to the discrimination theory in economics, which presents two main categories: theories of taste-based discrimination and theories of statistical discrimination. (10)
For the categorization for the concrete economic impacts of algorithmic unfairness we propose following: direct discrimination based on statistical attributes, direct discrimination from intentional favouring and indirect discrimination via advertising, recommendation and search (5). Besides the straight economic impacts, algorithmic unfairness might have psychological and cognitive effects, which should be researched further (17).
The concerns on algorithmic decision-making for public goods has been taken seriously (11), but link between algorithm unfairness and tech corporations needs to be constructed more carefully. It is unclear if the forms of algorithmic bias drives them more profits in some conditions (22).
Discussion
We have presented three emerging social and economic phenomena which characterize the machine learning era. The phenomena already exists in society to a certain extent. Many researchers have presented some empirical evidence, but technology is developing faster than ever, and scientists have hard time keeping up with the most recent social changes (1). According to our analysis, there are several reasons to believe that these effects on society will accelerate as machine learning is further adapted by organizations.
Many of the presented factors are not new. For example, there has been historical discussion whether technology will make jobs disappear since the industrialization and whether it will increase the general population’s living standards (13). Our claim is not that there will be massive unemployment as all low-skill low-wage jobs disappear. However, it is evident, that if the pace of change is as fast as it is now, there will be a lot of structural unemployment as it will be difficult to re-educate people to match needs of the businesses. It is simply based on the notion that the 4th industrial revolution is happening in unpreceded pace unlike the earlier industrial revolutions (1).
Similarly wealth and knowledge has been concentrated into the hands of few before, and individuals’ fate has been decided based on arbitrary reasoning. It is even happening today in many countries. The difference is that in the future the decision making is more and more computerized and the businesses, such as the tech-giants, are wealthier than majority of worlds nation states. Simultaneously media, employment, politics, entertainment, and every major building block of society are ongoing drastic changes.
The new world order brings about very concerning risk scenarios. By controlling huge amounts of data the companies have a massive surveillance system. They most likely already know you better than you know yourself. Simultaneously, they are growing in economic power. One interesting question is that what if one of them is actually evil? Or what if one of them turns evil and starts to exploit their position? China is planning dystopian a social credit system to be launched by 2020 which would give every individual a social credit score, run by private companies and the state of China (3). While the developed values disassociates us from big brother systems, one could emerge from the private sector nevertheless. Large monopolies who everyone is dependent on can already now ruin individual’s life.
The machine learning era is not all negative. There is huge potential for increasing global living standards. It has been presented that the global wealth will follow power-law, where few rich control most of the global equity (4). But if the productivity increases enough, it could mean that the curves base y value (minimum living standard) keeps on rising to levels where everyone has shelter and food. It could even result in a global basic income. In such scenario, many risks related to inequality are solved.
The discussion here brings forth the important question; what can we do to reach the good outcomes? We believe that studying the factors presented in this analysis further and finding empirical evidence on their existence is the first step. The regulation should also be updated actively to stay in par with technological change. The algorithm bias can also be turned around by using algorithms for example to intervene instead of judge. For example, young males being left outside of the society in Finland are estimated to cost the society one million euros each. With better algorithms, it could be possible to detect those in risk early and help them, instead of marketing bad loans to them.
In this article, we have aimed to compose a big picture of the social and economic phenomenon during the machine learning era. However, the scope of this article should be narrowed down to specific context to study the impact of machine learning through empirical methods. Ideas for forecasting the upcoming changes could be found within the research tradition of future studies. Another possibility would be netnographic study on the opinions and foresight of people who are involved in the development of machine learning.
References
Aghion, P., Jones, B. F., & Jones, C. I. (2017). Artificial intelligence and economic growth. (No. w23928).
Alimo‐Metcalfe, B., Alimo‐Metcalfe, J., Bradley, M., Mariathasan, J., & Samele, C. (2008). The impact of engaging leadership on performance, attitudes to work and wellbeing at work: A longitudinal study. Journal of Health Organization and Management, 22(6), 586-598.
Botsman, R. (2017). Big data meets big brother as china moves to rate its citizens. Wired, October 21, 2017.
Brynjolfsson, E., McAfee, A., & Spence, M. (2014). Labor, capital, and ideas in the power law economy.
Campolo, A., Sanfilippo, M., Whittaker, & M., Crawford, K. (2017). AI Now 2017 Report. Edited by A. Selbst. Available online https://ainowinstitute.org/reports.html. Accessed 30.1.2018.
Corbett-Davies, S., Pierson, E., Feller, A., Goel, S., & Huq, A. (2017). Algorithmic decision making and the cost of fairness. KDD '17 Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, Halifax, NS, Canada, August 13-17.
The Economist. (2018). How to tame the tech titans. Retrieved from https://www.economist.com/news/leaders/21735021-dominance-google-faceboo...
Frey, C. B., & Osborne, M. A. (2017). The future of employment: How susceptible are jobs to computerization? Technological Forecasting & Social Change 114, 254–280.
Gibney, E. (2016). AI talent grab sparks excitement and concern. Nature, 532, 422-423.
Goodman, B. W. (2016). Economic Models of (Algorithmic) Discrimination. 29th Conference on Neural Information Processing Systems (NIPS 2016), Barcelona, Spain.
Goodman, B., & Flaxman, S. (2016). European Union regulations on algorithmic decision-making and a "right to explanation". ICML Workshop on Human Interpretability in Machine Learning (WHI 2016), New York, NY.
Kim, Y. J., Kim, K., & Lee, S. (2017). The rise of technological unemployment and its implications on the future macroeconomic landscape. Futures 87, 1–9.
Krugman, P. (2013). Sympathy for the Luddites. The New York Times, June 14, 2013.
Makridakis, S. (2017). The forthcoming artificial intelligence (AI) revolution: Its impact on society and firms. Futures, 90, 46. doi:10.1016/j.futures.2017.03.006
O'Neil, C. (2017). Weapons of math destruction: How big data increases inequality and threatens democracy. Broadway Books.
Palmer, M. (2006). Data is the new oil. Retrieved from http://ana.blogs.com/maestros/2006/11/data_is_the_new.html
Rainie, L. & Anderson, J. (2017) Code-Dependent: Pros and Cons of the Algorithm Age. Pew Research Center, February 8, 2017. http://www.pewinternet.org/2017/02/08/code-dependent-pros-and-cons-of-th.... Accessed 31.1.2018.
Rifkin, J. (1999). The End of Work.
Rosenblat, A., & Stark, L. (2016). Algorithmic Labor and Information Asymmetries: A Case Study of Uber’s Drivers. International Journal of Communication 10, 3758–3784.
The Royal Society. Machine learning: the power and promise of computers that learn by example. April 2017, p. 48.
Russell, S., Dewey, D., & Tegmark, M. (2016). Research priorities for robust and beneficial artificial intelligence. Retrieved from http://arxiv.org/abs/1602.03506
Sandvig, C., Hamilton, K., Karahalios, K., & Langbort, C. (2014). Auditing Algorithms: Research Methods for Detecting Discrimination on Internet Platforms. Data and Discrimination: Converting Critical Concerns into Productive Inquiry,” a preconference at the 64th Annual Meeting of the International Communication Association. May 22, 2014; Seattle, WA, USA.
Seamans, R. (2017). We Won't Even Know If A Robot Takes Your Job. Forbes, Jan 11, 2017. https://www.forbes.com/sites/washingtonbytes/2017/01/11/we-wont-even-kno... Accessed 20.1.12018.
Stanford University. (2016). Artificial intelligence and life in 2030. Report of the 2015 study panel of One Hundred Year Study on Artificial Intelligence (AI100).
De Voorde, K., Paauwe, J., & Van Veldhoven, M. Employee Well-being and the HRM–Organizational Performance Relationship: A Review of Quantitative Studies. International Journal of Management Reviews, 14(4), 391–407.
The World Bank. (2016). Taking on inequality. Poverty and shared prosperity 2016.