2020Murat Volkan Dülger, Selin Çetin, Celal Duruhan AydınlıArticle

Abstract

This work was written in Turkish. The summary on this page is a translation; the citation gives the original title in parentheses.

Artificial intelligence (AI) applications have become a technology that is beginning to take its place in our daily lives and business processes. Because AI feeds on data, data protection and AI are necessarily linked. Given the increase in AI applications, data sets containing personal data have become a valuable resource. Algorithmic decision systems are one of the fields of application of AI technology. They have begun to be used in health, education, advertising and many other sectors.

They are preferred in various lines of business in order to speed up processes and reduce the error rate to a minimum. Thanks to algorithmic decision systems, both the private sector and public stakeholders are undergoing a positive transformation in terms of business development, efficiency and productivity. However, the risks this transformation carries must also be identified. In particular, the measures to be taken for the protection and security of personal data should be assessed at the design stage of algorithmic decision-making systems.

Full text

This text is a translation of the Turkish original, prepared for this website. Numbers in square brackets are the page numbers of the original; sources in the footnotes are given as in the original. For citation, the original publication (PDF) is authoritative.

Artificial Intelligence Working Group

Opinion

February 2020

The Artificial Intelligence Working Group was set up in 2019 within the IT Law Commission of the Istanbul Bar Association. The focus of the working group is the legal implications of artificial intelligence and robotic technologies.

The working group delivers legal opinions on artificial intelligence and robotic technologies and, in this context, works in cooperation with the other centres and commissions of the Istanbul Bar Association. Another aim of the working group is to make the technical and legal developments taking place in this field known to the legal community.

For further information about the Group: https://www.istanbulbarosu.org.tr/HaberDetay.aspx?ID=15275&Desc=Yapay-Zeka- %C3%87a%C4%9F%C4%B1nda-Hukuk

SUMMARY

Artificial intelligence applications have become a technology that is beginning to find a place in our daily lives and in our business processes. Because artificial intelligence feeds on data, data protection and artificial intelligence are necessarily interrelated. In view of the increase in artificial intelligence applications, data sets containing personal data have become a valuable resource.

Algorithmic decision systems are also one of the fields of application of artificial intelligence technology. They have begun to be used in health, education, advertising and many other sectors. They are preferred in various lines of business in order to speed up processes and to reduce the error rate to a minimum. Thanks to algorithmic decision systems, both private-sector and public stakeholders are undergoing a positive transformation in terms of business development, efficiency and productivity. However, the risks that this transformation harbours must also be identified. In particular, the measures to be taken with regard to the protection and security of personal data should be assessed at the design stage of algorithmic decision-making systems.

A. What Is Artificial Intelligence?

1.1. Artificial intelligence denotes machines that have the ability to solve the problems which humans solve with the intelligence they naturally possess.1 It appears, for example, in the form of applications that understand human speech, compete against other players in virtual games and interpret complex data.

1.2. Depending on the complexity of the model, artificial intelligence systems need large data sets, and in this process they are able to succeed to the extent that they are fed with data.

1.3.These technologies, which have become indispensable today, are, in technical terms, created
by using different methods.

1.4. The methods used in artificial intelligence work can be divided into three basic groups:

● Rule-based methods

● Supervised and unsupervised machine learning, which learn from data, and deep learning

● Methods of learning with less data by using reward and punishment2

1.5. Rule-based methods are also called expert systems. Rule-based systems are ideal particularly for solving routine and chronic problems. Although rule-based systems give definite results for specific problems, all the rules have to be defined in order for the problems that may be encountered in everyday life to be solved. Since this method of solution is not time-efficient, machine learning and deep learning methods, which learn from data, have begun to be used instead of rules created by experts.

1.6. Machine learningrefers to systems focused on making predictions from data, detecting complex patterns and
being able to take rational decisions. In today’s successful artificial intelligence applications, however, it is mostly
deep learning methods that are used. Deep learning makes it possible for computers to deliver, on unstructured data such as sound, images
and text, results close to human performance.

1.7. Where the data or the computing power is insufficient, sound results may not be obtained. By contrast, deep reinforcement learning stands out, particularly on account of its success in virtual games.

1.8. It is possible to see artificial intelligence applications in our daily lives in smartphones, tablets, voice assistants used in the home, smart home systems and driverless vehicles.

1.9. Artificial intelligence is transforming services as well as products. It has begun to be used for different purposes, ranging from determining the suitability of credit or investment portfolios to prioritising the content that will appear before us on social media and making personalised recommendations.

B. What Are Automated Decision-Making Systems and Their Areas of Use?

2.1. Automated decision-making3 denotes processes of taking decisions by automated means without any human involvement. Different artificial intelligence methods may be used in these processes.

2.2. Automated decision-making systems are encountered today in various fields. For example:

● Health care algorithms provide predictive analysis for hospital systems in determining how health resources can best be allocated.

● Teacher evaluation algorithms are used to measure the teacher’s effect on student achievement on the basis of standard achievement scoring rather than of factors affecting achievement such as individual achievement, family environment and professional experience.

● Social media monitoring algorithms are a tool used by law enforcement in many fields. They are used to gather evidence in support of the grounds for pre-trial detention, to identify potential offences and to detect potential incidents that may result in violence or another adverse outcome.

● Facial recognition algorithms are a biometric scanning technology preferred for tracking the entry and exit of residents and guests in some buildings and of employees in workplaces.

● Automatic licence plate reading algorithms are high-speed, computer-controlled camera systems that automatically capture all the licence plate numbers that come into view, together with the location, date and time of the vehicle and photographs of the vehicle, its drivers and its passengers. They are also used to predict where a driver may be afterwards.

● School bullying risk assessment algorithms are tools designed to identify pupils who may pose a risk in schools with regard to incidents such as homicide and suicide.4

C. What Are the Potential Benefits of Algorithmic Decision-Making Systems?

3.1. Algorithmic decision-making systems are encountered in day-to-day operations as elements that have begun to play an important role both specifically in commercial use and market strategies and in the protection of personal data and the determination of effective public policies. With regard to the benefit that these systems provide, the principle followed here will be that of assessing the effects of automated decision-making systems in respect of “natural persons”, the “private sector” and the “public sector”, a principle which has also been adopted by the European Parliamentary Research Service.5

a. The Place of Algorithms with Regard to Natural Persons

“Profiling” is one of the most fundamental headings that must be examined with regard to the effects of algorithms on natural persons. It is considered that, where an algorithm takes as its basis the results of what Article 226 of the General Data Protection Regulation (“GDPR”) also refers to as profiling practices and carries out decision-making processes concerning the data subjects, this is of importance both in the private sector and in a public context, above all in the health, banking, insurance and advertising sectors.7 Accordingly, it is envisaged that a system into which accurate data have been entered, in which the principles of privacy by design8 have been embedded and which has soundly completed the machine learning process will be able to assess persons and events free from the discriminatory practices that may exist in relations between natural persons. The comment is made that, in practice, the development of algorithms similar to Correctional Offender Management Profiling for Alternative Sanctions9 (“COMPAS”) may be on the side of the principles of equality and impartiality among the principles of criminal law and human rights, and that the use of these systems may play a balancing role10 in matters such as the function of the police.

It must be emphasised that the fact that an algorithm capable of offering the possibility of detecting the collection, processing and sharing of personal data without consent would be able to take steps towards the anonymisation or destruction of the data that have been breached is another matter that falls to be assessed. It is obvious that an algorithm which is capable of establishing the relationship between data and product, and for which the outputs of Deepfake11, one of the most recent examples of this and one based on content manipulations that are very difficult to distinguish from the real thing, are used as the source, will find a place both in data protection and specifically in public order.1213

The fact that automated decision-making systems, whose benefits in the field of health have also been visible for some time, are involved in imaging, diagnosis14 and even surgical processes, have gained a place in personalised treatment practices15 and have low margins of error shows that the use of algorithms in the world of medicine has reached an important point.

b. The Future of Algorithms in the Private Sector

The private sector is a field in which the place and importance of automated decision-making services are well understood. It makes room for algorithmic support under main headings such as the analysis and processing of large volumes of (bulk) data by persons and institutions, the automation of management processes, the ensuring of corporate control and the application of business development mechanics by providing productivity and efficiency analyses, and it is able to change company dynamics by means of supporting programs16 that provide boards of directors with decision-making services akin to consultancy. The matter on which algorithmic decision-making mechanisms fundamentally have an effect in the private sector may be shown to be the fact that a procedure which the white-collar workforce, in its routine of intellectual work, completes by spending hours or days and only with an uncertain margin of error can be analysed, and its road map drawn up, within minutes. The structure of the global economy, in which the speed of trade is the determining factor, has the potential to bring to the fore those private institutions and persons that resort to algorithmic support. The innovation that facial recognition systems have brought with the aim of eliminating security vulnerabilities, particularly in the banking sector;17 the speed which contract review and revision operations are bringing to self-employed lawyers and law partnerships;18 algorithms capable of producing effective sector-based results through the generation of case-law and market analyses;19 the fact that well-trained human resources algorithms, although still developing, are able to assess the CVs of the candidates best suited to the company culture with an egalitarian approach;20 and the existence of stock exchange algorithms and similar matters give an idea of the vision of the disruptive and innovative movement in the private sector.

c. Algorithms in the Internal and External Mechanics of the Public Sector

Although the use of algorithms in the public sphere may be a subject of debate in various respects, it is seen as an approach that ought to be preserved when the benefits it provides and is expected to provide are taken into account. The point that artificial intelligence may reach with the support of public power comes to the fore as one of the matters that have already been debated before parliaments21 and examples of which are being put into practice22. Within public services, matters such as institutions attaining an open, accessible and accountable character in sanction-based disputes such as the preventive criminal law approach in the law enforcement and justice system; the delivery of public services such as health, water and energy at low cost and free from operational errors; rendering effective the measures that need to be taken in epidemic cases such as the Corona virus23; and the carrying out of crisis management in line with traffic data within the urban order24 are foremost among the services in which making room for algorithmic support will make a difference. In matters such as combating terrorism and establishing public security, automated decision-making systems come to the fore as an anti-terrorist method that is already being applied worldwide by law enforcement and military forces for the purposes of detection and intervention25. When it is considered that cyber security attacks, which are another threat to public security, can cause far heavier material and non-material losses than a physical missile, the added value that an algorithm capable of displaying the initiative to detect26 and neutralise the attack can offer to public order, in our world where cyber security gains in importance with every passing day27, will be of a remarkable nature, especially if it is able to eliminate the potential for error of the human factor in the field of cyber security28.

Ç. What Are the Risks Inherent in Automated Decision-Making Systems?

4.1. As with all technological developments, developments in artificial intelligence technologies, while offering humanity great conveniences on the one hand, also carry certain serious risks for fundamental rights and freedoms on the other. Artificial intelligence technologies, which are built on information systems processing ever more data and taking decisions by drawing meaningful conclusions from them, inevitably bring with them the processing of personal data and the problems connected with it.

4.2. The problems that artificial intelligence technologies will give rise to with regard to personal data are listed below:

● The collection of “all personal data” or of “as much personal data as possible” in the context of the need for big data,

● Existing data becoming the subject of further learning and analysis, and new data being derived from these data,

● The possibility that data obtained in a particular context and/or through a particular processing activity may be processed for a wide variety of purposes that were unknown at the outset, and the impossibility, because artificial intelligence is unpredictable, of preventing the purposes of processing the data from being reconfigured or the data from being processed for too many purposes in a manner contrary to the principle of purpose limitation,

● Ever greater recourse to automated decision-making mechanisms.

4.3. In this respect, artificial intelligence is a structure that seeks to arrive at accurate analyses and to develop learning techniques by processing large amounts of data using a wide variety of methods. The monitoring and processing of large amounts of data, the fact that this activity can be carried out by artificial intelligence in a very short time and the fact that inferences with a very high rate of accuracy can be reached may pose a major problem for the application of the right to the protection of personal data and of the prohibition of discrimination. On the other hand, artificial intelligence systems are developed by humans and for humans. Developing technologies with their risks identified in advance serves to improve human living conditions. Developments should not be held back for the sake of risks; progress should be made with awareness of the risks and with the necessary measures in place. Indeed, resisting development and innovation will not slow the pace of this transformation.

Recommendations:

● In the development and use of artificial intelligence technologies that may have effects on individuals and on society, and in particular in decision-making / data analysis processes, fundamental human rights and especially the right to the protection of personal data must be taken as the basis. Processes and software must be designed from a data protection perspective.

● In the development of systems with artificial intelligence, it is humans who take on a role even before computers. In this respect, the necessary technological investments must be made, and avoiding the risks relating to the processing of personal data and eliminating these risks must be the basic approach.

● Risk assessment (impact analysis) in data processing must be carried out with greater sensitivity, having regard to the characteristics of artificial intelligence and big data.

● The principle of data protection by design must be observed at all stages of processing, including data collection, and in particular it must be ensured that artificial intelligence does not have biases that could amount to discrimination.

● All developers must assess the type, nature, source and quantity of the personal data to be processed, and unnecessary and excessive processing activities must be prevented. In this context, representative data that model real personal data, known as synthetic data, may be an effective solution.

● Those who develop, sell and use artificial intelligence must set up an algorithm that is accountable at every stage and in every process of the data processing carried out by artificial intelligence, and must determine to whom the responsibilities will belong (who the controller will be).

Istanbul, 2020

Assoc. Prof. Dr. Murat Volkan Dülger

Selin Çetin, Attorney-at-Law

Celal Duruhan Aydınlı, Trainee Attorney-at-Law

Footnotes

  1. For John McCARTHY’s definition and the explanations relating to it, see BOZKURT YÜKSEL, Armağan: Futurist Hukuk, İstanbul 2018, s.6. ↑
  2. SUTTON, Richard & BARTO, Andrew G., Reinforcement Learning An Introduction, MIT Press, 2017. ↑
  3. Information Commissioner’s Office (ICO), What is automated individual decision-making and profiling?, https://ico.org.uk/for-organisations/guide-to-data-protection/guide-to-the-general-data-protection-regulation-gdpr/automated-decision-making-and-profiling/what-is-automated-individual-decision-making-and-profiling/ , Erişim 12.02.2020 ↑
  4. AI Now Institute, Automated Decision Systems, Examples of Government Use Cases, https://ainowinstitute.org/nycadschart.pdf, Erişim 12.02.2020. These are examples from the practices of different states in America. ↑
  5. European Parliament, Understanding algorithmic decision-making: Opportunities and challenges https://www.europarl.europa.eu/RegData/etudes/STUD/2019/624261/EPRS_STU(2019)624261_EN.pdf Erişim 12.02.2020. ↑
  6. General Data Protection Regulation, Art. 22, https://gdpr-info.eu/art-22-gdpr/ ↑
  7. Information Commissioner’s Office (ICO), Rights related to automated decision making including profiling https://ico.org.uk/for-organisations/guide-to-data-protection/guide-to-the-general-data-protection-regulation-gdpr/individual-rights/rights-related-to-automated-decision-making-including-profiling Erişim 12.02.2020. ↑
  8. Privacy by design ↑
  9. COMPAS (Yazılım) https://en.wikipedia.org/wiki/COMPAS_(software) Erişim 12.02.2020. ↑
  10. Sharad Goel, Maya Perelman, Ravi Shroff, David Alan Sklansky; Combatting police discrimination in the age of big data; New Criminal Law Review; (20,2); 2017 ↑
  11. Deep Learning+Fake ↑
  12. Deep Learning for Deepfakes Creation and Detection, https://arxiv.org/pdf/1909.11573.pdf Erişim 12.02.2020 ↑
  13. Venneti, Real-Time Extraction of Biometric Data from Video Erişim 12.02.2020. https://insights.sei.cmu.edu/sei_blog/2017/08/real-time-extraction-of-biometric-data-from-video-1.html ↑
  14. Kaya, Sözmen,Usta, Elbaşı Kanser Teşhisinde Otomatik Karar Verme Erişim 12.02.2020. https://www.academia.edu/36301102/Kanser_Te%C5%9Fhisinde_Otomatik_Karar_Verme ↑
  15. IBM-Watson https://www.ibm.com/watson ↑
  16. FlexRule https://www.flexrule.com/archives/decision-automation-benefits/ Erişim 12.02.2020. ↑
  17. MasterCard Identity Check: Facial Recognition Biometrics https://newsroom.mastercard.com/eu/videos/mastercard-identity-check-facial-recognition-biometrics/ Erişim 12.02.2020. ↑
  18. Barnett Koshiyama, Treleaven, Algorithms and the Law, https://discovery.ucl.ac.uk/id/eprint/10076036/7/Treleaven_Algorithms.%20Law%20and%20Regulations_AA M.pdf Erişim 12.02.2020. ↑
  19. CaseLex https://www.caselex.eu/ Erişim 12.02.2020. ↑
  20. How Algorithms Are Changing The Face of HR https://www.bamboohr.com/blog/algorithms-changing-hr/ Erişim 12.02.2020. ↑
  21. House of Commons, Algorithms in decision-making: Government Response to the Committee’s Fourth Report https://publications.parliament.uk/pa/cm201719/cmselect/cmsctech/1544/1544.pdf Erişim 12.02.2020. ↑
  22. Data Ethics Framework https://www.gov.uk/government/publications/data-ethics-framework Erişim 12.02.2020. ↑
  23. TIME, Coronavirus Researchers Are Using High-Tech Methods to Predict Where the Virus Might Go Next https://time.com/5780683/coronavirus-ai/ Erişim 12.02.2020. ↑
  24. Veres, Moussa, Deep Learning for Intelligent Transportation Systems: A Survey of Emerging Trends https://www.researchgate.net/publication/334660151_Deep_Learning_for_Intelligent_Transportation_Systems_ A_Survey_of_Emerging_Trends Erişim 12.02.2020. ↑
  25. Future of Life Institute, Autonomous weapons: an open letter from AI and robotics researchers, future of life institute, July 2015, https://futureoflife.org/open-letter-autonomous-weapons/ Erişim 12.02.2020. ↑
  26. Rege, Mbah, Machine Learning for Cyber Defense and Attack https://www.thinkmind.org/download.php?articleid=data_analytics_2018_5_30_60121 Erişim 12.02.2020. ↑
  27. Cyber Defense Magazine, https://www.cyberdefensemagazine.com/cyber-security-statistics-for-2019/ Erişim 12.02.2020. ↑
  28. MITNICK, Kevin D., The Art of Deception. ↑

Download PDF (in Turkish)

More publications on this topic