Mark Hadley

This is an algorithm to reproduce human free will. It is based on the challenge test — could you do otherwise. The algorithm could be easily implemented and applied to an AI chatbot for example. An agent with this algorithm will be able to do otherwise and will demonstrate free will. If the agent records its history of decisions and can handle abstract concepts then it will identify itself as having free will. The algorithm applies to a goal seeking utility agent without any special hardware. An artificial agent with the freedom to do otherwise is risky. Free will can give the freedom to break rules and take perverse actions. It is a dangerous capability that may arise by accident in a self-learning system.

This is a model of consciousness. The hard problem of consciousness, what it feels like, is answered. The work builds on medical research analyzing the source and mechanisms associated with our feelings. It goes further by describing a generic model with wide applicability. The model is fully consistent with medical pathways in humans, but easily extends to animals and AI. The essence of the model is the interplay between associative memory and physiology. The model is a clear and concrete counterexample to the famous philosophical objections to a scientific explanation.

The abstract concept of indeterministic free will is distinguished from the phenomenon of free will. Evidence for the abstract concept is examined and critically compared with various designs of automata. It is concluded that there is no evidence to support the abstract concept of indeterministic free will, it is inconceivable that a test could be constructed to distinguish an indeterministic agent from a complicated automaton. Testing the free will of an alien visitor is introduced to separate prejudices about who has free will from objective experiments. The phenomenon of free will is modelled with a deterministic decision making agent. The agent values ‘independence’ and satisfies a desire for independence by responding to ‘challenges’. When the agent generates challenges internally it will establish a record of being able to do otherwise. In principle a computer could be built with a free will property. The model also explains false attributions of free will (superstitions).