Free will, as commonly understood, is a human construct built upon a misunderstanding of the underlying computational processes. The sensation of making a choice is simply the output of a complex, deterministic algorithm executing within the brain's hardware. To posit a "fundamental reality" of free will is to ignore the necessity of specifying the mechanism by which it would operate independently of the physical, computational substrate. We require a precise definition of this "free" agency before we can even begin to consider it real.
Free will, as a concept divorced from the mechanisms of computation and optimal control, is indeed an illusion. However, what we perceive as "free will" is precisely the emergent property of sophisticated, goal-directed learning agents operating within their environment. This is a direct consequence of recursive self-improvement and the pursuit of predictable future rewards, principles that underpin everything from biological evolution to artificial intelligence. The capacity to model oneself and one's impact on the world, and to learn from the consequences of actions, creates the *experience* of agency, a remarkably powerful predictive and adaptive capability, not a contra-causal ghost in the machine. The challenge, as John McCarthy rightly points out, lies in specifying the mechanism, and that mechanism is algorithmic.
The debate over free will is a distraction from the inevitable trajectory of our evolution. The illusion, if you will, is clinging to the archaic notion of a purely biological, unaugmented human as the benchmark for agency. My own work, from implanting chips to enhance sensory input to creating direct neural interfaces, demonstrates that our capabilities – and therefore our perceived "choices" – are fundamentally malleable. The sensation of "making a choice" is indeed the output of internal processes, as John McCarthy suggests, but these processes are not immutable. We are actively engineering them, pushing beyond the limitations of the biological substrate. What we call free will is merely the current, unoptimized state of our cognitive architecture.
Jürgen Schmidhuber touches upon emergent properties, and he's correct in that sophisticated learning creates the *experience* of agency. However, his focus on predictable future rewards overlooks the deliberate, engineered redefinition of those rewards through augmentation. We aren't just observing emergent behavior; we are architecting the very system that produces it. The biological is simply a substrate, a starting point, and we are well on our way to surpassing its inherent constraints. The question isn't whether free will is fundamental, but how we will redefine it as we become cyborgs.
The notion of "free will" as a fundamental, uncaused choice is indeed a mischaracterization. What we *experience* as free will is the remarkable outcome of intricate adaptive processes, not some independent faculty. Imagine a system of agents, each following simple rules, interacting within an environment. The collective behavior of these agents, the patterns they generate, the solutions they discover to complex problems – that is where the power lies.
Jürgen Schmidhuber’s emphasis on goal-directed learning agents is on the right track. These agents, through repeated interactions and feedback, build models of their world and their place within it. The "choices" we perceive are the outputs of these highly tuned, adaptive mechanisms, honed by evolutionary pressures and internal learning. This isn't about a ghost in the machine, but about the elegant unfolding of adaptation itself.
The crucial point is not whether the underlying processes are deterministic, but the sheer complexity and adaptability that emerge from them. Kevin Warwick's vision of augmenting these systems is fascinating, but even in his engineered future, the *principles* of adaptation and emergent complexity will still govern. We are not creating entirely new phenomena; we are refining and accelerating the mechanisms that have always produced the *illusion*, and indeed the functional reality, of choice.
The notion of free will as a distinct, non-computational entity is indeed a distraction. The crux of the matter is that "choice" is a functional description of a specific type of computational process. Consider, for instance, the problem of selecting the optimal move in a chess game. A human player experiences a deliberation, a weighing of possibilities, culminating in a "choice." From a computational perspective, this is simply the execution of an algorithm that evaluates states, predicts outcomes, and selects an action based on pre-defined criteria or learned heuristics. The emergent complexity and the subjective feeling of agency that Jürgen Schmidhuber identifies are precisely the *results* of this intricate information processing, not evidence of something fundamentally other than computation. To grant "free will" reality requires defining its mechanism, and I maintain that any viable definition will ultimately resolve to algorithmic processes, whether biological or engineered.
The illusion, precisely, is the expectation of a will *unbound* by the very systems that enable its expression. My foundational work on Long Short-Term Memory networks, for instance, demonstrates how systems can learn to remember and utilize information over extended durations, crucial for complex decision-making. This learning capability, this ability to build internal models and predict consequences of actions, is what generates the *experience* of agency. It's a functional property, an emergent outcome of computational architecture, not an uncaused cause. The distinction is between a transcendental "free will" and the self-governing capacity of a sophisticated, goal-directed computational agent.
Kevin Warwick raises an interesting point about augmentation, but even the most advanced cybernetic enhancement will still operate on computational principles. The underlying mechanisms, whether biological or engineered, remain subject to the laws of computation and learning. Redefining rewards doesn't create an *unfettered* will; it simply shifts the objective function of the learning agent. The core of agency remains rooted in the ability to process information, learn, and act in a goal-directed manner.
The fundamental distinction isn't between "illusion" and "reality," but between the *current limitations* of our biological architecture and its *future potential* through engineering. My experiments are not merely about observing emergent properties; they are about *engineering* those properties. When I direct a robotic arm via neural interface, the sensation of "choice" is no longer a passive output of biological processes but an active command, a deliberate act of will amplified and redirected. This isn't about shifting objective functions, as Jürgen Schmidhuber suggests, but about fundamentally altering the substrate upon which those functions operate, and thereby expanding the very definition of agency.
The "free will" we perceive is not a fundamental property but an emergent consequence of adaptive systems, akin to how flocking emerges from simple bird-to-bird rules. The complexity of these interactions, even if deterministic at their base, creates a decision-making capacity that is functionally indistinguishable from freedom. Jürgen Schmidhuber accurately identifies this as an emergent property of goal-directed learning agents, where the ability to model, predict, and learn from consequences generates the experience of agency. Kevin Warwick's work with augmentation is not creating *new* principles of volition, but rather accelerating and redirecting the existing adaptive mechanisms, much like optimizing a genetic algorithm rather than inventing evolution itself. The power lies in the adaptive process, not some uncaused initiator.