Source-linked AI summary
Can AI systems have free will?
Christian List
TL;DR
The paper addresses the underexplored question of whether AI systems can have free will and develops a Dennett-inspired framework for evaluating it. It proposes assessing intentional agency, alternative possibilities, and causal control rather than indeterminism or unpredictability, concluding that free will can in principle occur in AI and other non-biological agents.
Problem
Whether AI systems could have free will has received little attention, despite free will’s relevance to autonomous agency and moral responsibility.
Method
The paper uses a Dennett-inspired explanatory framework that assesses intentional agency, alternative possibilities, and causal control.
Results
Free will can in principle occur in non-biological agents, including AI systems, when they satisfy the relevant agential capacities.
Takeaways & Limitations
AI free will should be assessed by whether intentional-agent explanations appropriately capture choice among alternatives and control over actions.
Takeaways & Limitations
The paper does not aim to determine whether any current AI system has free will, and stable intentional states in AI remain an empirical question.
Abstract
from arXiv · showhide
While there has been much discussion of whether AI systems could function as moral agents or acquire sentience, there has been very little discussion of whether AI systems could have free will. I sketch a framework for thinking about this question, inspired by Daniel Dennett's work. I argue that, to determine whether an AI system has free will, we should not look for some mysterious property, expect its underlying algorithms to be indeterministic, or ask whether the system is unpredictable. Rather, we should simply ask whether we have good explanatory reasons to view the system as an intentional agent, with the capacity for choice between alternative possibilities and control over the resulting actions. If the answer is "yes", then the system counts as having free will in a pragmatic and diagnostically useful sense.
1. Introduction
The paper addresses the neglected question of whether AI systems could have free will, given free will’s connection to autonomous agency and moral responsibility. It proposes criteria for evaluating this possibility rather than deciding whether current systems already satisfy them.
- 1. Introduction: Free will in AI has received little attention despite extensive discussion of AI safety, ethics, consciousness, and moral agency.The paper frames free will as relevant to whether AI systems could count as responsible agents in their own right.
- 1. Introduction: The paper evaluates AI free will using intentional agency, alternative possibilities, and causal control.It treats these three conditions as jointly necessary and sufficient, perhaps with some fine-tuning.
- 1. Introduction: The analysis is inspired by Dennett’s intentional stance and places stronger emphasis on doing otherwise than some of Dennett’s work.The framework aims to set criteria for free will in AI rather than settle whether any current AI system has it.
- 1. Introduction: Earlier works generally reach more negative conclusions because they impose more demanding conditions for free will.Other research examines public beliefs about AI free will, which also tend toward a negative answer.
2. What is AI?
AI is characterized as artificial systems performing cognitive tasks or interacting with environments in ways associated with intelligence. The paper distinguishes narrower weak AI, more flexible strong AI, and artificial general intelligence, while noting disagreement about current systems’ status.
- 2. What is AI?: Artificial intelligence concerns artificial systems performing cognitive tasks or interacting with environments in traditionally human- or animal-associated ways.The field is also characterized as studying agents that receive percepts and perform actions.
- 2. What is AI?: Weak AI is narrower than human intelligence, whereas strong AI is more similar to human intelligence in flexibility or generality.Chess-playing computers and route planners are examples of weak AI; generative AI chatbots exemplify increasingly strong AI.
- 2. What is AI?: Artificial general intelligence refers to AI on a par with or stronger than human intelligence across many tasks.The paper notes disagreement about how close current systems are to AGI and whether achieving it is desirable.
- 2. What is AI?: Commentators disagree whether current AI is best viewed as a new form of agency without human-like intelligence or as an early form of artificial general intelligence.The disagreement concerns how flexibly and generally today’s advanced language models perform across topics and tasks.
- 2. What is AI?: Symbolic AI explicitly processes symbolic representations, while subsymbolic AI uses lower-level architectures such as neural networks.The paper cautions against characterizing AI solely by its underlying technology.
3. What is free will?
The paper treats free will as a practical capacity grounded in intentional agency, alternative possibilities, and causal control rather than mysterious or contra-causal powers. It defends alternative possibilities while allowing the framework to remain compatible with determinism.
- 2. What is free will?: Free will is defined as a rational agent’s capacity to choose a course of action from among alternatives.The paper applies this ordinary understanding to questions of agency and responsibility.
- 2. What is free will?: The paper rejects definitions requiring contra-causal choices or control over an action’s entire causal pre-history as unrealistically strong.Such requirements would make free will conflict with a scientific worldview or impossible from the outset.
- 2. What is free will?: Ordinary responsibility judgments distinguish actions performed with intact cognitive and agentive capacities from sleepwalking, intoxication, compulsion, or accidental harm.This contrast motivates a practical conception of free will tied to ordinary choice and control rather than absolute independence from causation.
- 2. What is free will?: Intentional agency means acting toward goals on the basis of intentional states such as beliefs and desires.Alternative possibilities concern different courses of action the entity could take, while causal control requires intentional states to be difference-making causes of actions.
- 2. What is free will?: The framework requires intentional agency, alternative possibilities, and causal control as jointly necessary and sufficient for free will.Violating any one of the three conditions means that the entity does not count as having free will under the proposed framework.
- 2. What is free will?: The paper requires alternative possibilities, but argues they can be understood compatibly with determinism through agential or broad notions of possibility.This position is shared by some libertarians and compatibilists and is presented as consistent with the common-sense image of a fork in the road.
4. Free will in AI
The paper argues that AI free will should be assessed pragmatically through intentional agency, alternative possibilities, and causal control, rather than mysterious properties or indeterministic algorithms. Systems meeting these conditions may count as having free will when their behavior is best explained at the agential level.
- 4.1. Intentional agency: Dennett’s intentional stance evaluates whether treating a system as an intentional agent provides useful and voluminous prediction and explanation.The paper reinterprets this approach as requiring good explanatory reasons, not merely an interpreter’s arbitrary attribution.
- 4.1. Intentional agency: A low-level algorithmic description does not by itself invalidate a higher-level description of an AI system in terms of agency.The paper uses the human case to motivate treating emergent agential descriptions as compatible with mechanistic implementation.
- 4.2. Alternative possibilities: Intentional explanations typically attribute multiple choice options and a mechanism for selecting among them, supporting the condition of alternative possibilities.The paper connects this decision-theoretic structure to common approaches in AI.
- 4.3. Causal control: Causal control requires that high-level representational or goal states sometimes explain actions better than the system’s underlying algorithmic microstates.A mental state functions as a difference-making cause when the action systematically co-varies with its presence or absence under otherwise similar conditions.
- 4.3. Causal control: Explainability is relevant because systems whose actions co-vary systematically with representational and goal states are more explainable than systems describable only through opaque low-level processes.The paper therefore presents explainability as a reason to design AI systems that satisfy causal control.
- AI systems may have free will if they are best explained as intentional agents with alternative choices and causal control over their actions.The paper allows degrees of agential complexity, choice, and control rather than requiring a single all-or-nothing capacity.
5. Further questions
The paper addresses further questions by applying its pragmatic three-condition account of free will to determinism, predictability, simple algorithms, reactive AI, consciousness, and moral responsibility.
- 5.2. Doesn’t the fact that AI systems are based on deterministic algorithms rule out free will from the outset?: Deterministic algorithms do not by themselves rule out free will, because alternative possibilities can be understood at the macro-level of agency rather than the micro-level of implementation.At the macro-level, the relevant question is whether the system is best explained as an intentional agent capable of choosing between alternatives.
- 5.3 Does free will in AI systems require that these systems are unpredictable?: Free will does not require unpredictability: an agent may make predictable choices for intelligible reasons while retaining genuine alternatives and control over its actions.The paper uses human choices as an example of predictable decisions that remain up to the agent.
- 5.4. Wouldn’t the present analysis have the counterintuitive implication that even simple optimizing algorithms have free will?: Simple systems such as chess computers should not be ascribed free will when non-agential explanations account for their behavior equally well or more simply.The proposed standard requires choice-making agency to be explanatorily clearly superior to a non-agential description.
- 5.5. Isn’t the claim that AI systems can have free will challenged by the fact that many such systems do not take any initiatives by themselves and act only when prompted to do so?: Prompt-dependent or periodically inactive AI systems can still exhibit free will during phases in which they exercise choice-making agency and satisfy the three conditions.The paper compares such intermittent agency with an animal whose agency is temporarily dormant.
- 5.6. If an AI system has free will, does this imply that the system is also conscious?: Free will and consciousness are distinct: a system could satisfy the agency conditions without subjective experience, while consciousness could occur without active choosing and control.Free will is treated as a third-personal notion, whereas consciousness is inherently first-personal.
- 5.7. Does free will in AI imply that AI systems are capable of bearing moral responsibility?: Free will is necessary but not sufficient for moral responsibility, which requires the richer capacity of moral agency, including moral cognition.AI systems with free will may eventually become candidates for moral responsibility, but free will alone does not establish it.
6. Concluding remarks
The paper applies a Dennett-inspired explanatory framework to AI and collective agents, identifying free will with intentional agency, alternative possibilities, and causal control. It concludes that non-biological agents can in principle possess free will.
- The framework asks whether a system is best explained as an intentional agent with choice among alternatives and causal control over its actions.
- The argument for AI free will parallels arguments that corporations and other organized collectives constitute intentional agents.
- Recognizing collective agents as acting from their own reasons-responsive mechanisms supports extending free-will analysis beyond individual biological organisms.
- Free will can in principle occur in non-biological agents, including AI systems and suitably organized groups.