Huang Jung-Tsun Special Article: When AI Has Free Will

Modern humans are irresistibly curious about genes, heredity, fate, and free will—especially when confronting new biomedicine and artificial intelligence. My recent annotated translation of Benjamin Libet’s (1916–2007) Mind Time (published as How to Measure Free Will by Lianjing Publishing, 2025) directly engages with these questions. This book compiles four decades of research (1964–2004) and laid foundational work for the experimental science of consciousness.

Libet discovered that before a subject consciously decides to perform a voluntary action (e.g., bending their index finger), researchers can detect a subtle electrical signal outside the skull over the supplementary motor area (SMA)—a region preceding the primary motor cortex toward the front of the brain. This signal, known as the readiness potential (RP), was first identified in 1964 by Hans Kornhuber and Lüder Deecke at Freiburg University through complex EEG recordings. It has since been recognized as a key neurophysiological marker of planning, preparing, and initiating voluntary movement.

Libet found that the RP begins approximately 400 milliseconds before conscious awareness of the intention to act. This suggests that the brain initiates voluntary actions unconsciously, long before we feel we’ve made a decision. Thus, the execution of voluntary action may be independent of conscious intent.

Is free will an illusion? Libet did not intend to deny free will outright. Instead, he proposed a “conscious veto” window: after becoming aware of an intention, individuals have about 100–200 milliseconds to consciously abort the action. However, in practice, making such a rapid, conscious decision to stop is extremely difficult. Many researchers therefore argue that the belief in free will is itself an illusion—implying free will does not truly exist.

Libet’s findings are startling: unconscious neural processes appear to initiate what we experience as conscious, free decisions. Stephen Kosslyn (former Harvard psychology professor, Minerva University president) stated: "Libet’s discovery seemed controversial at first but has stood the test of time—any theory of consciousness and its neural mechanisms must now account for his results."

This raises profound ethical, legal, and social questions: If free will is illusory, how can humans be held morally responsible? Can people be blamed for crimes if their brains initiated actions before conscious choice? Libet’s idea of a 'veto power' offers a partial escape, but it remains debated. These issues remain central in academia, law, psychiatry, and society—a battle still far from resolved.

What exactly is free will? We don’t know. Libet didn’t study the structure or essence of free will itself but instead asked how it could be operationally defined and what mechanisms influence its function. Thus, free will remains a black box. Most discussions circle around concepts like causa sui (self-causation): to bear moral responsibility, one must be the origin of one’s own actions—at least at critical mental levels.

Higher moral values and concepts cannot be merely learned; they must be self-originated. But this contradicts conventional causality, where causes precede effects. To resolve this, some link God as the ultimate self-origin of all things. Nietzsche called this self-contradictory. Galen Strawson argues nothing can cause its own existence—thus, free will cannot exist, and moral responsibility should be re-examined.

Other arguments also contain paradoxes. In the age of genetics, some suggest a 'free will gene' might exist. But this is contradictory: if free will is genetically determined, it is no longer free. Similarly, if AI can learn and simulate free will, what happens?

My colleague Academician Hsing Yi-Tien, a renowned historian of Qin-Han and Roman history, read my lengthy introduction to How to Measure Free Will and remarked: "Only the exceptionally intelligent would dare tackle these issues. But if free will can indeed be measured and computed in a lab, it would be revolutionary. In an AI era driven by computation and learning, once AI masters volition, it will surpass humans. Won’t humanity then submit to AI rule?"

I responded that I undertook this translation because students—even in Western universities—often stare blankly when encountering Libet’s work. Opposing scientists and philosophers are too busy arguing to truly grasp his technical depth. Hence, I dedicated years to translating and annotating this challenging text.

This project was even more difficult than my earlier translation of Kendler’s The Age of Revelation. Having served in high-level government roles, I chose honorary academic positions post-70 to avoid burdening my university. After leaving the Examination Yuan, I applied for National Science Council research grants to continue scholarly translation and writing, sustaining my lifelong academic passion.

Regarding Hsing’s sharp question, here is my modest response:

(1) The 'free will' discussed is operationally defined: when one becomes aware of intending to act (e.g., bend a finger), that moment is labeled as free will. Yet hundreds of milliseconds earlier, the brain’s motor planning area already shows accumulating 'readiness potential'—detectable outside the skull. This implies unconscious brain activity precedes conscious intention, suggesting free will may be an illusion.

(2) Can free will be measured? Though a high-level concept, if its internal structure is clarified, it could be treated as an entity—like mass, temperature, or memory—and modeled computationally. Then, measurement methods could follow. If fully understood, AI could quickly learn to detect, calculate, and manipulate free will. But science lacks a computational theory of free will—it remains unclear, hence the reliance on operational definitions. Since experts haven’t solved it, AI won’t master it soon.

(3) Human consciousness includes 'third-person' experiences (measurable, shareable across individuals), like geometric shapes or movie plots. But more crucial are 'first-person' subjective experiences—private, impenetrable, unmeasurable qualia—such as emotional tears after a film or the vivid red of an apple. Free will is clearly part of this first-person domain. Francis Crick believed science must eventually explain this core mystery, rather than leave it to philosophers.

(4) Current AI faces three major barriers: (a) self-awareness ('Who am I?'), (b) creativity (genuine innovation beyond pattern replication), and (c) qualia (subjective experience). Until these are overcome, AI cannot truly emulate human free will.