I believe the responsibility should lie with the user. On one hand, it is important to keep in mind that artificial intelligence is meant to help us, not to be something we rely on blindly. In this sense, human beings are also responsible for informing themselves about AI and its potential drawbacks and limitations. Therefore, it is up to us to apply a critical filter to what AI proposes and to consider it as a tool rather than an essential part of our lives, despite how helpful it may be. For this reason, when making certain decisions—whether in a personal or professional context—we must be aware of the responsibility that our choices entail. In line with this, if decisions based on AI turn out to be harmful, we should avoid using artificial intelligence as an excuse and instead take responsibility, act to repair any damage caused, and learn from the experience for the future.
An AI is a tool that systems learn from existing data and if it is already prejudices or inequalities, the decisions may also be harm someone. The people and institutions that support this tool and set the rules for it, because AI cannot Take responsibility. This must be done by the decision-makers (companies, public authorities, etc.) be aware of every use. It is therefore important that AI cannot be achieved without human control and there are clear (legal) rules to make decisions must also be comprehensible.
Basically, it is to be questioned whether decisions should be made by AI-based systems at all. In any case, with regard to the question raised, the type of decisions involved must be classified. In the rarest of cases, decisions can be brought about by a simple mechanism. Where this can be brought about by simple yes/no processes, it can – and will be done – through simple automated and generalized processes. Decisions that require some kind of intelligence cannot be made by artificial intelligence, as they always have discretion and must therefore be instructed more than a simple logic. The vast majority of decisions require a view from different perspectives and a certain background of experience, which is not guaranteed by the processes described above. Especially when it comes to ethical decisions from all area ethics, AI-based decision-making is to be classified as completely inadequate and thus rejected. In principle, however, a human decision is upstream of the AI-generated decision-making process, namely that of bringing about the decision through an AI-controlled process. Thus, there is a contact person - be it a person or a collective (e.g. in the form of an institution) - to whom the responsibility for commissioning the AI can be clearly assigned. The responsibility for harmful AI decisions is therefore clearly to be borne by this person or persons.
This question is also a little hard to answer because it is very context dependent. AI is certainly, for now, just a tool. Usually, when a person uses a tool to harm someone, whose fault is it? For the vast majority of cases, according to the law, it is the user, seldom the victim, and very rarely the company that designed it, but never the tool. Our framework of law will have to quickly adapt, as I believe it’s already happening, to accommodate for the usage of AI powered tools.
IVania Noteva, Bulgaria, AI Enthusiast When an AI-driven decision harms someone, responsibility should ultimately remain with the person who made or approved the decision. AI is a tool, not an independent moral actor, and people should not use it as an excuse to avoid accountability. Human judgment, oversight and responsibility are still essential, especially when decisions affect other people’s lives.
Depending on the type of harm that is done, who has used the model - this is to me a matter of future legal regulation, but I see liability similar to liability of medical personnel when a mistake is made (to give one example how I would solve)
