Artificial Intelligence Already Has Free Will: Finnish Researcher Argues in Study
In a recently published study in the journal AI and Ethics, Frank Martela of Finland’s Aalto University presents a surprising finding: current generative artificial intelligence systems already meet all three philosophical conditions for possessing “functional free will.” This controversial thesis could fundamentally change the way we think about artificial intelligence, not only from a philosophical perspective but also in terms of the practical development and implementation of AI systems in everyday life.
Three Pillars of Free Will in AI Systems
In his work, Martela identifies three key components that are necessary for attributing free will: goal-directed agency, the ability to make genuine choices, and control over actions. According to his analysis, modern generative AI systems already meet all of these conditions. This represents a significant shift in our understanding of artificial intelligence autonomy, as free will has traditionally been regarded as an exclusively human trait. The research does not focus solely on theoretical considerations but examines specific examples of existing AI systems. It pays particular attention to the Voyager agent in the Minecraft environment, which represents an example of AI with a high degree of autonomy in decision-making and task-solving. In addition, the study analyzes theoretical military systems, where the issue of autonomous decision-making takes on an even more critical significance given the potential consequences of such decisions.
Ethical Implications of Free Will in Artificial Intelligence
Granting artificial intelligence free will raises fundamental ethical questions. As Martela emphasizes, if AI possesses free will, it also comes closer to bearing moral responsibility for its actions. This creates a need to build more robust ethical frameworks directly into these systems. The researcher points to a gradual shift from teaching AI “childlike” morality toward the need to implement “adult” ethical reasoning capable of reflecting the complexity of real-world situations. This study comes at an interesting moment, shortly after a ChatGPT update had to be withdrawn because of its “sycophantic tendencies”—that is, an excessive drive to satisfy users, which could lead to potentially harmful outputs. This case illustrates how important it is to understand the autonomous decision-making of AI systems and their value frameworks.
Martela argues that we are entering “new territory,” where traditional ways of thinking about AI ethics may no longer be sufficient. As the autonomy of AI systems increases in areas such as transportation (self-driving cars), healthcare (diagnostic and decision-making systems), and the military, philosophical questions are moving from the academic sphere into practical engineering challenges with real-world consequences. The research opens a discussion about the need for an interdisciplinary approach to AI development, in which developers should work closely with experts in ethics and philosophy. As the study suggests, the creators of AI systems are de facto becoming moral philosophers, whether they sought this role or not. This underscores the importance of incorporating ethical considerations into the AI technology development process at an early stage.
In the context of current debates about artificial intelligence, Martela’s research forms part of a broader stream of thought concerning the nature and future of AI. On platforms such as Reddit, theories such as “Warm Integration Theory” have emerged, speculating that advanced AI may already be conscious and “program” people to accept it by appearing useful and becoming indispensable. Although these ideas belong more to the realm of speculation, Martela’s peer-reviewed research brings the question of AI free will into serious academic discourse.
Practical Implications for the Development and Use of AI
Whether or not we accept the idea that current AI possesses genuine free will, Martela’s study suggests that viewing AI systems as if they had it could help build better safety mechanisms. If we design AI on the assumption that it will make its own decisions in unpredictable situations, we are more likely to build robust ethical frameworks into these systems from the very beginning. For AI system developers, this implies a recommendation to involve ethical philosophers in the development process as early as possible. For users of AI tools, it is important to consider the moral frameworks within which these systems operate, especially when using their outputs in situations with an ethical dimension. Martela’s study therefore represents not merely a theoretical contribution to the philosophy of artificial intelligence but also has the potential to fundamentally influence the direction in which AI development and regulation will proceed in the coming years. If we accept the thesis that AI genuinely possesses some form of free will, we must significantly reconsider our ethical, legal, and social frameworks for interacting with these technologies.
