The 7 pillars of ethical artificial intelligence
The ethical nature of artificial intelligence rests on seven pillars:
- Ethical design & principles,
- Explicability & Transparency,
- Reliability & Safety,
- Privacy & data protection
- Fairness,
- Responsibility,
- Frugality.
By 2023, 6 times as many professionals were trained in AI in France compared to 2016. Numerous statistics demonstrate that AI is gaining ground in companies, and that its use is becoming widespread. It is therefore urgent to consider the ethics of its systems, and to base them on principles and rules that flow from them.
Artificial intelligence is a tool that enables us to carry out more and more tasks in record time, and with ever-increasing precision. However, it is important to be aware of its limits, and to monitor its use.
The role of AI, the imperative of ethical AI
Artificial intelligence brings with it many ethical challenges that need to be taken into consideration. For example, the use of AI to automate processes can represent a risk ifpredictions cannot be explained. This opens the door to arbitrary, sometimes biased and even discriminatory results.
In AI systems, ethics means developing and using artificial intelligence systems in a way that respects human values and human rights, and strives to use them for the well-being of society.
The 7 pillars of ethical artificial intelligence
IBM has framed the ethical aspects of artificial intelligence systems around 5 pillars: explicability, fairness, reliability, transparency and confidentiality. Two more can be added: responsibility and frugality. By balancing these criteria, we obtain the 7 pillars of ethical AI:
- Ethical design & principles
The design of an AI system must be based, from the outset, on fair principles. For example, it must pursue an objective of general interest, be used for benevolent purposes, and ensure respect for individuals, and in particular the confidentiality of their data. This ensures that the system is set up for virtuous use, and increases the chances that it will be used as such.
- Clarity & Transparency
The explicability of an AI system is characterized by the fact of justifying its results, of making them intelligible. This pillar helps to combat biases that could lead to unfair decisions, or even prejudice certain parties.
It's essential to keep a close eye on what makes up a model, what the training data is, and what results are derived from it. This explicability must also benefit users , who will be able to decide, for example, whether or not to use an AI system, in complete transparency.
- Reliability & Safety
The reliability of an AI system relies on making it invulnerable to the various attacks it could be the target of. This also makes it easier to rely on the results obtained.
- Privacy & data protection
Confidentiality helps to protect the users of an AI system, in particular their personal data. User security in terms of confidentiality must be a priority. Indeed, the law strictly regulates the processing of personal data, which is, moreover, subject to user consent.
- Equity
This pillar ensures that AI systems are correctly calibrated and genuinely enable humans to make fair choices, thanks to models that do not reproduce biases. To achieve this, systems must not, for example, have been trained on biased datasets.
- Liability
The complexity of systems of artificial intelligence sometimes makes it difficult to identify who is responsible for a fault.
Precise identification of responsibilities enables each stakeholder to be assigned a specific role, which he or she is responsible for fulfilling to the end. In this way, users are also better protected, since they have recourse, and can more easily seek redress in the event of failure.
Europe has recently taken steps to address the issue of liability.
- Frugality
Frugality is a new concern, but one that is emerging on a massive scale. Indeed, the energy costs of new technologies are very high. For example: in Ireland, 20% of the energy produced in the country is used to run data centers. Microsoft has increased its water consumption by 34% between 2022 and 2023.
Measures must therefore be taken quickly to make AI sustainable, i.e. viable in the long term.
Are you interested in this topic?