Using AI responsibly

According to different definitions, data can be

• a collection of discrete or continuous values that convey information, describing the quantity, quality, fact, statistics, other basic units of meaning, or simply sequences of symbols that may be further interpreted formally (Wikipedia)
• information that is produced or stored by a computer (Britannica Dictionary)
• factual information, such as measurements or statistics, used as a basis for reasoning, discussion, or calculation (Merriam-Webster Dictionary).

An algorithm is to data processing as a recipe is to cooking: a set of instructions for processing ingredients, or data in this case.

According to different definitions, an algorithm can be:

• a set of mathematical instructions or rules that, especially if given to a computer, will help to calculate an answer to a problem (Cambridge Dictionary)
• a systematic procedure that produces, in a finite number of steps, the answer to a question or the solution of a problem (Encyclopaedia Britannica)
• a detailed description or instruction on how to perform a task or a process, or steps you can follow to solve a specific problem. (Wikipedia)

Note that “algorithm” is not a technological concept; it can refer to any particular systematic way of processing data in a way that results in a specific output or activity. In this context, however, we are only talking about algorithms as a critical element of digital computing.

Automated decision-making as a concept does not directly refer to any specific technology or data processing method. Instead, it encompasses two main categories:

• Rules-based automation where decisions are generated from standard data according to locked processing equations. For instance, an automatic tax return is an example of rules-based automation. The areas of application and use cases are limited and pragmatic.
• AI-based automation, i.e. independent systems. These kinds of algorithmic systems trained with data have the ability to process materials of different nature and output analyses, forecasts or situational pictures and present them in different formats. There are many potential areas of application and use cases that are difficult to limit and predict.

Artificial intelligence is not a specific method of processing data, so it has been particularly difficult to find a definition that would satisfy everyone. This has caused issues when drawing up the EU Artificial Intelligence Act, as the definition of the core concept in the proposed Act has received a high amount of conflicting criticism from actors in the sector.

Some key definitions:

• The ability of a digital computer or computer-controlled robot to perform tasks commonly associated with intelligent beings. The term is frequently applied to the project of developing systems endowed with the intellectual processes characteristic of humans, such as the ability to reason, discover meaning, generalize, or learn from past experience. (Encyclopaedia Britannica)
• All software that operates autonomously to some extent and produces output according to tasks set and defined by humans. The produced output can include content, forecasts, recommendations or decisions that affect the environment that the software or device interacts with. (Algorithmic discrimination and the promotion of equality, in Finnish)Opens in a new window.

Even these definitions disagree on the scope of the definition and is core features, but AI can be overall understood as something with

• the ability to perform activities that are considered to require human intelligence;
• the ability to act independently, autonomously
• the ability to interact with signals and data received from the environment and adapt its actions to them in relation to the system’s past operation and outputs (system learning)
• the objective to achieve goals and produce output requested by a human from the system in a way that optimises the results of achieving the objective.

Machine learning is an essential method in artificial intelligence and data science. Without machine learning, we would probably not have the data processing systems that we now call artificial intelligence in everyday language.

Machine learning is defined in somewhat varying ways, for example:

• Machine learning is the subset of artificial intelligence (AI) that focuses on building systems that learn, or improve performance, based on the data they consume. (Oracle)
• As a subset of AI, the purpose of machine learning is to make software work better based on base data and the actions of a potential user. (Wikipedia)

Machine learning does not mean that a system would learn and evolve independently to become better and more efficient at what it has been taught to do. Machine learning systems learn when trained for a specific purpose. They also need to be further trained or retrained to keep up-to-date in a changing environment. In addition, the systems should be carefully tested before they are introduced.

Deep learning is a subset of machine learning. It is thanks to the development of deep learning that new autonomous algorithmic systems are able to combine different types of data and material and generate new kinds of outputs. Deep learning is based on so-called neural networks that mimic the basics of human brain activity.

Deep learning is a method in artificial intelligence (AI) that teaches computers to process data in a way that is inspired by the human brain. You can use deep learning methods to automate tasks that typically require human intelligence, such as describing images or transcribing a sound file into text.

– Amazon Web Services (AWS)

Neural networks are structures of information technology that enable deep learning and mimic the functioning and structures of the human brain: neurons, synapses and all.

A neural network is a method in artificial intelligence that teaches computers to process data in a way that is inspired by the human brain. It creates an adaptive system that computers use to learn from their mistakes and improve continuously. Thus, artificial neural networks attempt to solve complicated problems, like summarizing documents or recognizing faces, with greater accuracy.

– Amazon Web Services (AWS)

Consequentialism is based on the view that the consequences of actions hold more value than their formal purpose. Whether an individual action is good or bad is only determined by its consequences.

There has been plenty of consequentialist thinking in history, but in modern times, it has been represented by utilitarianism. The principles and objectives of utilitarian ethics were formulated by English philosopher and jurist Jeremy Bentham and further developed by English philosopher and economist John Stuart Mill.

It is the greatest happiness of the greatest number that is the measure of right and wrong.

– Summary of Bentham’s basic principle of utilitarianism

The task of an ethically robust society is therefore to produce the greatest possible welfare for as many people as possible, and this has been a key principle of most modern Western democracies.

The downside of utilitarianism is that it does not offer tools for resolving conflicts arising from differences between people, groups and situations. If, for example, we decide to promote overall happiness by increasing security, and to do so we send criminals to prison on lighter grounds, we will also have raised the costs incurred by society. This in turn may lead to poorer basic services and a decrease in overall welfare.

On the other hand, consequentialist thinking carries a danger of a wellness dictatorship for the majority: if we optimise the greatest wellbeing for the greatest number of people possible, what happens to those who are not in this majority?

The ethical and moral commandments of religions can be understood as early forms deontological ethics, which is the ethics of duty or rules. Today, deontology is largely based on the theories of German philosopher Immanuel Kant.

Unlike consequentialism, deontology considers an act morally good based on a characteristic of the act itself, not whether the result of the act is good. So, the assumption is that there are universal values and rules that we always have to follow, no matter the circumstances and consequences. Such ethics can be represented with expressions such as “duty for duty’s sake” or “virtue is its own reward”.

Instead of moral rules imposed by a higher power, Kant believed in people’s ability to use reasoning to know what is the morally correct and required action. He also believed that such “rationally moral” and self-aware thinking and action would lead to democracy and world peace over time.

Kant’s philosophy, with its winding and often difficult-to-follow arguments, did produce a motto that has persisted to current times, known as the categorical imperative. It posits that an action is morally good if you “act only according to that maxim whereby you can at the same time will that it should become a universal law”.

Criticism of Kantian ethics includes the notion that formulating rules that should always be followed is practically impossible, and any attempt most often leads to rigid and empty top-down commandments. Some also consider Kantian ethics to be an emotionally cold structure that ignores the individual’s internal experience of the world.

In the West, virtue ethics were particularly prevalent among ancient Greek philosophers. But this philosophy continues to play a particularly important role in East Asian culture in China, Korea and Japan.

In virtue ethics, the focus is on the person themselves, although not as someone who acts out a set of ethical codes, but as an individual who strives to improve their virtue and moral character, as if to become more refined as a person. The person is the focus of ethical growth.

Most of the theories of virtue ethics have been inspired by Aristotle, according to whom a virtuous person is a person with ideal characteristics. For example, a person is virtuous if they are friendly in many situations throughout their life because it is in their nature and not because they deliberately want to maximise benefits, receive favours or simply perform duties.

The theories of virtue ethics do not primarily aim to identify universal principles that can be applied in any moral situation. The theories of virtue ethics address broader questions such as “how should I live?” or “what is a good life?”

This kind of personality-focused ethics probably makes the most sense in an organic and hierarchical worldview: in order for society to maintain itself, different people have to serve different roles, and if people do not act correctly in their roles, society will not be able to maintain itself.

Criticism against virtue ethics includes the difficulty of applying it in society and it being time- and place-specific to the values of each culture, its weakness in conflict resolution and the fact that its presumption of permanent, clear characteristics of each individual may be fundamentally incorrect.