Insights · AI Compliance and the EU AI Act ·
Algorithmic will: a conceptual proposal between automation and autonomy
Classical will theory rests on consciousness, freedom of choice and capacity to act. When a learning model produces the decision, that structure dissolves. Algorithmic will is a concept proposed to describe that dissolution.
Legal systems define will from a human-centred perspective. In classical theory will consists of three elements: consciousness, that is, awareness of one’s act and of the consequences that will flow from it; freedom of choice, the ability to choose consciously among possibilities; and capacity to act, the physical and mental power to put the chosen option into effect. Under Article 1 of the Turkish Code of Obligations, a contract requires mutual and corresponding declarations of will. A declaration of will is the projection of a person’s will into the external world with the intention of producing a legal effect.
This structure rests on the assumption that the person who takes the decision and the person answerable for it are the same. Products running on machine learning and deep learning models challenge that assumption directly.
The source of the problem: unforeseeable output
AI-enabled products are able, by the nature of algorithms, to learn from large volumes of data. Once that data is internalised, a system may produce decisions that appear experiential or intuitive in a manner resembling human decision-making. Where it operates autonomously, the behaviour of a system tasked with reaching a particular goal may become indistinguishable from that of a person.
The critical point is this: just as a person may behave unforeseeably, an AI-enabled product may — particularly in a black box situation — arrive at solutions its user and its manufacturer did not foresee. Combining large language models with memory, planning and execution units enables such products to set goals for themselves, convert those goals into concrete plans and refine them. The resulting behaviour is close to that of a human-like agent.
At this point the conceptual apparatus available to law falls short.
Why “algorithmic will”
Two alternatives were considered in selecting the concept. The first is decision making, widely used in AI scholarship: the process of evaluating a set of alternative actions and automatically identifying the option that best satisfies predefined objectives or performance criteria. The second is agent, used in the sense of a representative.
Both concepts presuppose a predefined objective function and a traceable decision process. The genuine legal problem, however, arises in products where algorithmic traceability is not possible and which possess autonomous decision-making capacity. The concept of algorithmic will was therefore preferred to describe the integration of the notion of will with artificial intelligence technologies.
Algorithmic will is, unlike classical will, an artificial species of will that has no human being at its centre and is constituted by the logics of automation and autonomy.
The distinction between automation and autonomy
Two technical concepts are decisive in the legal examination of algorithmic will.
Automation is the performance of operations and tasks without human intervention within the framework of predetermined rules. Products operating through automation produce planned and foreseeable output within the programmed rules and the chosen learning model. A system performing credit risk analysis for a bank produces a result within criteria set by its algorithmic code; that result can be traced backwards.
Autonomy is the ability of a system to take independent decisions according to environmental variables. Autonomous systems can restructure their own decision algorithms from the experience gained during learning. Here the output may become independent of the original design.
For legal liability the distinction is not cosmetic. In a product that remains at the level of automation, fault and causation can be established; in one that has passed into autonomy, the same determination may be technically impossible.
Two dimensions of the concept
Algorithmic will can be examined in a technical and a legal dimension.
In the technical dimension, the learning model and the data constitute its basic components. The architecture of the model, the quality of the training data and whether the system continues to learn determine the capacity of the will.
In the legal dimension, legislation, transparency obligations and data protection policies draw the boundaries of algorithmic will. That framework is the external limit determining how far the will is left free.
No legal personality — so where does liability go?
No existing legal order recognises algorithmic will as a fictional legal personality. Although work on an e-personhood model exists, the prevailing view is against recognition. The EU Artificial Intelligence Act likewise declines to confer legal personality on an AI-enabled product, adopting instead a risk-based classification and a human-centred, indirect chain of responsibility.
The practical consequence is this: algorithmic will is treated within the framework of human-centred, indirect liability. Responsibility is distributed not to the product itself but to the actors at the source of the algorithmic will — the manufacturer, the software developer and the user.
The governing criterion that follows is clear: the degree of autonomy of algorithmic will is not merely a technical capacity but the criterion determining the limits of legal liability.
What this means in practice
For a company that develops, places on the market or uses an AI-enabled product in its commercial activity, this discussion is not theoretical. In contract negotiations, in allocating liability along the supply chain and when a dispute arises, the first question asked is the algorithmic will capacity of the product. If that capacity has not been documented, the question of who bears liability will be argued later, and before a court.
For algorithmic will to find a full counterpart in legal doctrine, classical will theory must be reinterpreted and the “subject” status of AI-enabled products examined conceptually. Until that examination is complete, liability will continue to be constructed through the source of the will.
This article is adapted from the third chapter of Av. Nur Gülmez’s LL.M. thesis, “Legal Liability Arising from the Use of AI-Enabled Products in the Context of Algorithmic Will Capacity”, completed at the Institute of Social Sciences of Akdeniz University.