Ethical Management of Big Data: Reclaiming Control

With the development of big data technologies, algorithms of all kinds are utilizing the data streams and playing an increasingly decisive role in individual decision-making. It would be alarming to suggest that we are being manipulated by these algorithms. But these algorithms do influence many of the choices we make, whether it's choosing a hotel, a flight, or even a route, picking an online book, or making friends through social networks.

Besides facilitating and enabling our choices, these algorithms are also involved in the construction of social life in their own way. This algorithmic "power" is emerging, and it relies heavily on raw data. This power has the unprecedented ability to reach into the most intimate corners of people's lives. Even totalitarian regimes, with their spies and eyes and ears, have never dared to think of such power.

Personalized information is generated through the use of data at scale. The argument that this information "facilitates" our choices, based on predictions of what is possible, is somewhat muddled. There are organizations with which we have data connections, which are interested in our choices and which try to predict and guide them. These algorithms control and influence our choices, but we have little control or knowledge of the criteria on which they are based.

How do we regain control? One way to think about it is to design and build a model for analyzing, understanding, and processing these vast amounts of complex data - an ethical model.

The premise of data ethics: algorithmic principles

We need to be clear here. This is not to inhibit the flourishing of big data, but rather to make our lives as free as possible from the constraints of blind rationality and the objective needs of the marketplace. Nor is it about constructing some kind of two-way model that is both common sense and responsive to the enormous potential of big data.

However, given the principles of data mining, it is entirely feasible to construct an ethical model for complex data. That's because the logic underlying the inductive algorithms at the heart of big data is strikingly similar to the "practical wisdom" at the heart of ethics.

In everyday life, humans collect data, interpret information, make connections with memorized knowledge, and acquire abilities that can be used repeatedly in the future. In this way, they acquire a kind of "practical wisdom," the art of behavior, which can also be developed and perfected in ethics.

The logic of data-mining algorithms is very close to the logic of everyday human behavior. They are both inductive, not deductive. Big data algorithms are not designed to reason out arguments and reach indisputable conclusions; they are not mathematical operations. The partial, incomplete and unstructured data on which such algorithms are run are not sufficient to support the argumentation process. Such algorithms are in fact the recognition of repetitive actions, of behavioral cues, of behavioral patterns. For example, at Amazon.com, they find that readers of one type of book are likely to be interested in another type of book as well. These algorithms collect data, aggregate the data into information, interpret the information, and make connections to remembered knowledge, thereby providing a streamlined range of choices that in turn lead to practical outcomes.

Both in terms of human behavior and in terms of data algorithms, the key node is "simplification," the process of transforming complex raw data into useful information. In the Amazon example above, this means not recommending anthropology to science fiction fans. In this critical aspect of simplification, ethical models of complex data should attempt to maintain and develop this discernment.

There are two necessary principles to follow here. First, so-called "information" is linked to action through some systematic framework. Information is aggregated into knowledge, and this knowledge is a practical knowledge, recognized in action. It is not so much knowledge as it is the practice of knowledge.

The second principle comes directly from information science theory. Rather than data processing, it is data state description. The key to the ethics of big data is also the transformation of complex knowledge, which is disordered and ambiguous, into structured and simple knowledge that leads to some kind of eventual practice.

The key to data ethics: data hierarchies

The decisive aspect of data simplicity is data hierarchies. Data hierarchies "tune" algorithms to produce usable results. Data hierarchies require prior consideration of the value of data, which is accomplished through a series of questions: Why is it being evaluated, and for what purposes and goals? How is the value of a piece of data or information assessed, and by what criteria? What, precisely, should we be evaluating?

We can assess the value of a piece of data in terms of its specifics: for example, a click indicates that someone liked it, that someone went in a certain direction or back, or that someone intends to pay. We can also assess the value of a piece of data in terms of redundancy, variability, and volume. The value of the data also depends on the knowledge as a whole: some data contributes less to cognition, while others are more meaningful. Finally, we can assess the value of data at the level of data sharing, that is, in terms of the quality and quantity of data exchanged.

The value of data is also related to the service it provides to the user. The assessment of information is to determine the dissemination strategy of the information, i.e., to provide the right information at the right time, and to selectively push the information according to the interests and needs of the customers, so as to eliminate misleading information and information flooding.

System designers must therefore determine what data and how much information to provide to users. What data do users need to make "good" decisions and act "well"? In order to achieve a balance between data improvement and data overload within an information system, two variables can be optimized for data hierarchies and data picking functions.

The first refers to the iterative evaluation of data at all levels, but if this re-evaluation is done too often, the benefits of re-evaluation are offset by the data overload caused by back-and-forth transfers. The second refers to the amount of data contained, managed, and waiting to be moved in the smallest storage unit, and too much of it can also cause tiering to become complex and slow.

This work of grading and evaluating data is critical. Ethical considerations are also introduced at this point. Next we examine a sensitive example: medical data.

An example of ethical analysis: medical data

Medical data can be characterized as a boundary between two interests: as patient privacy, which should be protected, and as epidemiological statistics, which are useful to humanity as a whole. How to figure out these two factors? On the basis of the four principles, we can establish an ethical approach. In their related book, Principles of Biomedical Ethics, Tom Beauchamp and James Childress establish these four principles.

First, the principle of bona fides, or contributions to the welfare of others. Bona fide behavior is subject to two specifications, that is, bona fide behavior must be, first, beneficial and, second, useful, as evidenced by positive cost-benefit. Second, the principle of autonomy, which means that each person sets a code of behavior for himself. According to this principle, patients must be involved in the decision-making process. Third, the principle of non-maleficence, which means avoiding evil and unnecessary harm and suffering to those to whom we owe the performance of our duties and responsibilities. Fourth, the principle of impartiality, that is, the sharing of available resources (time, money, energy) among all patients. This principle, in turn, is closely linked to the concepts of equality and equity, which are implicated in the process of impartial decision-making. Ideally, any behavior should tend toward full equality, but depending on the situation and the individual, equity is often emphasized to establish some priority or hierarchy of actions.

A well-designed healthcare data-picking process can satisfy three of these four ethicalization principles.

The principle of goodwill is reflected in the moderation of information released to users (healthcare workers and the public) to ensure that actions are appropriate and reasonable. Communication becomes more efficient.

The principle of autonomy means that information is clear, accurate, appropriate, and easy to understand, with the explicit consent of the person concerned. Patients have the right to participate in deliberation, decision-making, and action.

The principle of non-maleficence is reflected in the limitation of data rights according to the identity and nature of the user, and the improvement of data security, confidentiality and data protection.

This selective approach to data is counterproductive to the principle of impartiality. Messages are sent differently for different users. The system sets specific rules for information distribution and access for each individual. Such information asymmetry is discriminatory and challenges the transparency of information.

Data hierarchies and data picking are realized based on the importance assigned to the data and the issues involved in its use and distribution. By simplifying the data sent, data use and access become more efficient, and data collection and data security are improved. However, this approach results in poor data integrity. Thus, data hierarchies make the work of all types of users easier, but make the technical challenges of information system designers even greater.

This selective data hierarchy plays an important role in both data complexity and accessibility. We see it as a form of "organizational intelligence". The algorithms constructed on the basis of ethical data mining principles generate new information that we can call "ethical information". Information that has been ethically evaluated and pre-processed has greater value in its later use.

When Ethics Improves Data Quality

This method of grading and selecting initial data improves the qualitative value and entropy of knowledge at the cost of quantitative loss of data and information. Meanwhile, the automatic selective grading system of data automatically transfers data to the corresponding service tier according to the different needs of users through biased low storage occupancy.

The above approach is well illustrated by the work on inductive algorithms at the heart of Big Data technology. There is no single, one-size-fits-all inductive solution. Nevertheless, the range of options is relatively clear for specific purposes. Just like the ethical process, the best performing inductive algorithms should be evolving. Depending on the most appropriate and feasible solution, these algorithms adapt the way they process data and thus improve themselves. In order to build such algorithms, data processing must be predictable and functional. For this reason, the use of big data requires the early conversion of data into usable, ethical information.

In this general context, the study of selective hierarchical schemes from an ethical perspective helps us to better understand the precarious equilibrium between data availability, data confidentiality, and data protection. Depending on the given situation, this equilibrium can sometimes fall this way and sometimes that way. Prior to data cherry-picking, such an approach tends to throw a series of questions at us: what are the goals, objectives, keys, and implications of doing so? What data will I use? Partial data or all data? How am I going to use this data? Where? For which users? More generally, how do I use the mixed data accumulated and stored within the information system? What is the relevance of this data as a whole to my situation? Will this result in a distortion of the value of the original information? Can the integrity of the final information be preserved?

Technology can't solve all the problems. The protection of personal information and its privacy depends on both professional ethics and self-discipline. This requires the development of codes of ethics to govern the design, implementation, and use of personal data in Big Data. This raises new questions: what body or organization would be responsible for developing such a code, and for facilitating the process of certifying "ethical" algorithms?

The above is what I have shared with you about the ethical management of big data: regaining control, and for more information you can follow the Global Ivy to share more dry goods