It is probable that AI will impact life in various areas. In each area, there are opportunities where AI can improve life, but there are also risks that AI can worsen life. The following chapter will provide a brief overview of some of these areas and hereby provide a base for the use cases we will be examining later on.

Potential risks can be found in almost every imaginable application of AI. Will improving medical diagnosis split society into two groups, amplifying the gap between rich and poor? Will advanced production robots leave thousands of people unemployed? Will more capable predictive procedures in crime fighting lead to even stronger racial bias? Many of these problems can be thought of as “wicked problems”, as first described by Horst Rittel (Rittel and Webber, 1973). This means, there is no one, definite solution, but rather they are problems that are difficult and even perhaps impossible to solve, due to the vast amount of factors that must be considered. But despite these risks, the chances of improving life through AI are enormous as well. The question comes up, how the benefits of AI can be achieved while the risks are being kept at a minimum? The first step in enabling the beneficial use of AI is to design the system itself in a beneficial way and thus minimizing the opportunities for misuse and maleficent use of AI. This approach seems to be the most realistic in terms of actually reducing the previously mentioned risks.

To achieve this goal, it is necessary to look into potential application areas in detail and figure out which problems could emerge during implementation of such AI systems. These problems then must be approached by figuring out what countermeasures can be taken to reduce the extent of the problem. The beneficial framework attempts to give guidance in terms of which factors can lead to problems (and what potential countermeasures can be), whereas the next chapter will look into three particular impact areas and demonstrate how the framework can assist during the process of diminishing emerging problems. The three impact areas that will be further examined are the area of healthcare, in particular medical diagnosis, job finding and urban planning.

Healthcare

Elderly Care

Due to predicted demographic developments, it is apparent that the proportion of elderly people (above the age of 65) will grow globally in the next centuries (“World Population Prospects - Population Division - United Nations,” n.d.). Therefore, it is probable that the need for medical staff and doctors will rise in many countries, including Germany (“Jährlicher demografiebedingter Ersatzbedarf an Humanmedizinern und Ärzten in Deutschland von 2010 bis zum Jahr 2030.,” n.d.). AI can provide assistance here, for example, in the area of home care using remote medical monitoring solutions such as the company “biotricity” (“Biotricity - Remote medical monitoring technology for physicians and consumers,” n.d.) provides, or even professional assistance for medical emergency dispatchers such as the Copenhagen company “corti” (“Corti – Products,” n.d.).

Medical Diagnosis

On top of that, there is also steady progress in the area of medical diagnosis. Researchers at Stanford University have been able to train a deep learning model to identify skin cancer “as well as dermatologists” (Kubota, 2017). In another example, researchers at Harvard Medical School, MIT and Beth Israel Deaconess Medical Center have demonstrated the opportunities for using deep learning models to improve accuracy of breast cancer diagnosis (Wang et al., 2016). In the latter example, it is especially interesting to note that the trained deep learning model alone was not more accurate than the human alone – but the errors the model made were not correlated to the errors the human made, so combining the diagnosis generated from the model with human expertise led to improved accuracy. Other recent progress demonstrates the capabilities of image recognition systems when used for lung cancer detection (“Google shows how AI might detect lung cancer faster and more reliably,” n.d.), though it is important to note that in most studies, the data sets are chosen very carefully. Such perfect data-sets do not accurately represent real-world situations.

Job Finding

The role of recruiters or headhunters is likely to change through AI as well. Startups, such as “woo”, already use machine learning to support both candidates and employers in finding the perfect match (“Woo - The right job opportunity,” n.d.). Companies, such as the Indian startup “Arya”, also address issues like bias in their models (“Arya - AI Recruiting Technology,” n.d.). Though the methods used by such companies currently are more similar to a filter process, where the main advantage is that AI can process larger amounts of data more quickly, it is probable that more complex methods could be employed in future scenarios. Possible developments include not only faster initial selection processes, but also AI-led interview processes (Forbes Coaches Council, n.d.).

Urban Planning

The task of planning a city is highly complex, because it requires expertise in many different disciplines, such as architecture, ecology, politics and social-cultural aspects (Hamdy, 2017). A common difficulty in this process is quantifying the extremely high complexity of human and community behaviour in these situations, which is a prerequisite for planning based on facts and statistics. This often leads to oversimplification of the facts which can lead to crucial errors in urban planning, because a city is not a simple, nor an entirely rational, nor an entirely predictable system (Saiu, 2017).

AI can tackle this problem by being able to handle much more data than traditional methods. Researchers at MIT, for example, are exploring how machine learning methods can support decision making processes in urban planning (Zhang et al., 2018). Another research group at MIT is utilizing machine learning techniques to create data-driven interactive simulation tools for urban planning, in order to enable rapid prototyping procedures to quickly identify and visualize which impact certain decisions would have (Alonso et al., 2018). There are also developments outside of academia, for example the Helsinki based company “CHAOS Architects”, which builds tools for city analysis, as well as forecasting scenarios in city developments (“CHAOS architects,” n.d.).

Catastrophe Prediction

Natural disasters will remain a large threat to humanity, even more so as there are indications that the frequency and severity of natural disasters will increase in the future, due to human factors such as climate change. For example, scientists at Harvard University and The University of Sheffield estimate that the amount of wildfires in North America will most likely increase due to climate change (Yue et al., 2015). Geological evolutions also pose a threat in the future, one example being the dangers of future earthquakes in Chile (Coghlan, n.d.), where difficulties include the, up until now, unreliable prediction of when exactly the events would occur. As Gavin Hayes, a seismologist at the United States Geological Survey in Golden, CO, states: “Unfortunately, earthquake prediction is still elusive, and we cannot give a precise date or size of a future event.” (Choi,LiveScience, n.d.) Earthquake predictions could be improved by AI, or at least the speed in which they can be made could be drastically improved(Fuller and Metz, 2018). Additionally, DeVries et al. suggest using deep learning to recognize patterns in earthquakes to predict aftershocks (DeVries et al., 2018). Advancements in sensor technology, such as nanotechnology and smart dust, such as high-resolution cameras the size of a grain of salt (Gissibl et al., 2016), combined with advancements in image classification and recognition, could further enhance the collection of data and thus improve predictions.

Financial

The amount of globally collected data is increasing rapidly and is estimated to reach 175 Zettabyte by 2025, compared to merely 33 Zettabyte in 2018 (“Prognose zum Volumen der jährlich generierten digitalen Datenmenge weltweit in den Jahren 2018 und 2025 (in Zettabyte).,” n.d.) (1 Zettabyte equals 1 billion Terabytes). Making use of ever larger amounts of data is not feasible with techniques of traditional data analysis. Big data therefore utilizes techniques such as machine learning, data mining or predictive analytics to process the volume, velocity and variety of this data (“Big Data Analytics,” 2019). The financial industry already relies heavily on data, therefore advances in the area of big data analytics are likely to provide useful benefits for this sector. Companies like PricewaterhouseCoopers (Pricewater-houseCoopers, n.d.) or the Boston Consulting Group (He et al., 2018) are already addressing the opportunities of advanced analytics. Smaller startups such as “Alpaca” provide services for financial market predictions using deep learning (“Alpaca,” n.d.), while other companies such as “DataVisor” provide services for detecting fraud and other financial crimes (“DataVisor Home Page » DataVisor,” n.d.). AI can also support in credit decisions, risk management, trading, personalized banking and process automation (Bachinskiy, 2019). Regulators, such as the European Banking Authority Banking Stakeholder Group, also see potential risks, for example, in the validation process of more and more complex models (“BSG+response+to+Joint+Discussion+Paper+(JC+2016+86) -+17+March+2017.pdf,” n.d.).

Production (Industry 4.0)

The industrial sector is already making use of AI and specialized Industry 4.0 software and is likely to increase the use of these tools in the future. In a 2018 survey among 553 executives in the German industrial sector, only 9% claimed that Industry 4.0 is not a topic relevant for their production, whereas 49% claimed they are already using Industry 4.0 applications (“Bedeutung von Industrie 4.0 in Deutschland 2018 | Umfrage,” n.d.). Software companies, such as IBM, show examples of improving productivity through AI, e.g. optimizing maintenance schedules or predicting power demand in the utility sector (“AI & Industry 4.0 beyond the hype,” 2019, p. 0). Business consultancies, such as Roland Berger, offer services for successfully implementing Industry 4.0 and AI technologies, pointing out that, opposed to simply buying new equipment, there are long-term strategic measures to be made when implementing such technologies (“How Industry 4.0 will impact electronics assembly,” n.d., p. 0).

Education

A commonly mentioned potential for AI in the educational sector is personalized learning, with which learning experiences could be individually adapted for students, learning gaps could be uncovered and the overall teaching content could be more personalized (“Bots in learning - AI and personalized learning experience,” 2018; “Personalized Learning: Artificial Intelligence and Education in the Future,” n.d.; Khurana, 2018). But there are also advantages imaginable that concern career path prediction (Mwiti, 2019) or organizational and administrative improvements, leading to teachers being able to spend more time on their main task: teaching students (Utermohlen, 2018).

Food Production / Management

Efficiency of food production could be improved by supply chain optimization, similar to how overall improvements in the industrial sector can be achieved. Companies, such as Symphony Retail AI, are in fact specializing on providing AI-enabled solutions for food producers (“Symphony RetailAI - Artificial Intelligence Enabled Retail and CPG,” n.d.) and by doing so, not only improving the efficiency of the supply chain but also optimizing sustainability by predicting the actual demand more accurately, resulting in less waste (“Symphony RetailAI Named a Recipient of Supply & Demand Chain Executive’s Green Supply Chain Awards,” n.d.). IBM sees great potential for these types of developments within the next five years. IBM’s predictions include improved efficiency for farming through data-driven processes (“#twinning,” n.d.), as well as reduced food waste through blockchain technology (“Blockchain will prevent more food from going to waste,” n.d.).

Improvements, other than the maximization of efficiency, can be found in cleaning systems that utilize AI to clean equipment the appropriate amount (as over-cleaning is very common in current systems), therefore reducing costs and improving resource management. Such a system is currently being developed by researchers at the University of Nottingham, claiming it could in theory save 100 million £ a year in the UK alone (“Artificially-intelligent cleaning system could save food manufacturers £100m a year - The University of Nottingham,” n.d., p.). Lastly, there is also a great amount of approaches which concentrate on taste and flavour optimization. One example for these types of improvements is Gastrograph AI, an analytical platform, specialized in providing consumer preference prediction and preference market insights (“Gastrograph AI | Analytical Flavor Systems,” n.d.).

Cybersecurity

Due to the increasing digitalization of companies, governments and economies, the potential threats of cyberattacks are growing rapidly. The World Economic Forum predicts a $3 trillion economic loss by the year 2020 (“Centre for Cybersecurity,” n.d.), resulting from malicious programs. Currently, an estimate of 350,000 new malicious programs and potentially unwanted applications are being developed daily (“Malware Statistics & Trends Report | AV-TEST,” 2019). This enormous amount leads to traditional software security systems becoming unfeasible, due to their limits in capacity. This is where AI can be helpful, as it can handle larger amounts of data and utilize distributed processing power more effectively (Joshi, n.d.). But on the flipside, AI can also be used to generate such malicious programs and perform cyberattacks. Nicole Eagan, CEO of the cybersecurity firm Darktrace, predicts a future, in which AIs will be used as measures for cybersecurity, as well as for cyber attacks, resulting in an “AI vs. AI”-scenario (“The Future of Cybersecurity is A.I. vs. A.I.,” n.d.).

Art

Current progress in generative adversarial networks, among others progress, make it seem likely that AI will not only have an impact in terms of productivity and efficiency, but leisure as well. The fine arts are one example, where recent progress in AI is having an impact. In October 2018 the auction house “Christie’s” sold a painting created by a generative adversarial network (GAN) for 432,500$, claiming it to be the “first AI artwork to be sold in a major auction” (Vincent, 2018). This event sparked a large debate about what should be considered art, whether these pieces were original, and whether the AI will redefine the meaning of being an artist (“AI Is Blurring the Definition of Artist,” 2018).

Crime Fighting

Predictive analytics is a branch of analytics which specializes in the accurate modelling of future events, based on previous data (“What is Predictive Analytics ?,” 2018). Due to the amounts of data being required for such predictions, the field of predictive analytics benefits strongly from advances in machine learning, as more data can be processed more efficiently. The resulting predictions of such models are already being used by judges in the USA to evaluate how likely a criminal is to commit another crime, one of the more prominent solutions being the software COMPAS (short for: Correctional Offender Management Profiling for Alternative Sanctions) (“The Northpointe Suite,” n.d.). These predictions about the likelihood of future criminal activity are known as risk assessment, and have been strongly criticized in the past. One example of this strong criticism is the 2016 report by the non-profit organization “ProPublica” concerned with machine bias.

The report points out strong racial bias in the predictions the COMPAS software creates concerning risks of convicted criminals relapsing (Julia Angwin, 2016). But perhaps even more ethical concerns should be raised when AI is not only used for predicting recurrence of crime, but also the prevention of crime, even before it actually occurs. This AI-enabled approach to crime fighting is called predictive policing. Predictive policing is being used by justice systems, for example in the United Kingdom (Malik, n.d.). Recently, the New York City Police Department (NYPD) has unveiled that they have been using predictive policing as well, which has led critics to express concerns due to the possibility of reinforcing racial bias – an especially critical topic in the USA (“NYPD’s Big Artificial-Intelligence Reveal,” n.d.).

Bibliography

Approaching Beneficial AI

Nico Göckeritz, Mark F. Meyer

Master-Thesis

< back