Can AI systems have free will?
cs.AI · cs.CY · physics.soc-ph
Quick verdict
List argues AI free will is possible via a Dennett-style checklist (agency, alternatives, control); rigorous philosophy but criteria remain qualitative.
Summary
Christian List addresses a neglected question in AI philosophy: whether AI systems can have free will. Rather than seeking mysterious properties, indeterministic algorithms, or unpredictability, he proposes assessing AI systems against a three-condition checklist drawn from his earlier work and Dennett: intentional agency, alternative possibilities, and causal control. For each condition he asks what would justify ascribing it: good explanatory reasons for viewing a system as a belief-desire agent (anchored in Gurnee and Tegmark's LLM world-model findings and Levinstein and Herrmann's belief-attribution arguments); whether intentional explanation presupposes choice options; and whether representational states act as difference-making causes and control variables of behavior. He reinterprets Dennett's intentional stance in realist fashion: explanatory indispensability indicates, rather than constitutes, real capacities. A macro-level notion of agential possibility lets alternative-possibility claims survive deterministic underlying algorithms. Objection sections cover the bearer of free will, determinism, unpredictability, chess computers and thermostats (agential explanation must be clearly superior), reactivity, consciousness, and moral responsibility (free will is necessary but not sufficient). Main limitations: the criteria remain qualitative, with no operational diagnostic for concrete systems; the load-bearing macro-possibility thesis is asserted via prior work rather than defended; and the empirical belief-attribution evidence is cited without being weighed against LLM inconsistency problems. The paper aims to set criteria, not to settle whether current AI has free will.
Review complete
Assessment at a glance
3
risk flags identified
● high 0
● medium 2
● low 1
5 strengths · 6 weaknesses · 5 author questions
AI text signal = Binoculars zero-shot detector score over 20 sampled passages (Fast-DetectGPT co-reported) — absence does not prove human authorship, and a flag does not imply misconduct.
Levels reflect verification complexity — not a quality score. Evolve never rates your paper.
Key claims under review
- Free will should be assessed via three conditions — intentional agency, alternative possibilities, and causal control — which are assumed jointly necessary and sufficient (§3–§4).
- The explanatory indispensability of the intentional stance toward a system is real-world evidence (not the definition) that the system genuinely has agential capacities (§4, fn 6).
- Alternative possibilities can obtain in systems whose underlying algorithms are deterministic, because agential possibility is a macro-level notion and determinism is not preserved across levels of description (§5.2).
- Unpredictability is not required for free will: predictability grounded in intelligible reasons is compatible with genuine choice among alternatives (§5.3).
- Free will should be withheld from chess computers and thermostats because non-agential explanations are feasible and simpler; attribution requires agential explanation to be clearly superior (§5.4), and free will is necessary but not sufficient for moral responsibility (§5.7).
Strengths
A genuinely operational-looking framework for a vague question
The three-condition checklist (intentional agency, alternative possibilities, causal control; §3, after List 2019) decomposes 'can AI have free will?' into three separately assessable sub-questions, each given an epistemic test in §4: explanatory reasons for belief-desire attribution (§4.1), the presuppositional structure of decision-theoretic explanation (three-step scheme, §4.2), and difference-making counterfactuals/control variables (§4.3, after List and Menzies 2009; Roskies 2012).
A real refinement of Dennett, not mere application
Footnote 6 and §4 distinguish indicator from defining criterion (the driving-licence analogy), reinterpreting Dennett's constitutive reading of the intentional stance (Dennett 2009, p. 339) as realist evidence; the paper also retains the alternative-possibilities requirement Dennett drops in 1984, ch. 6, via the Luther re-analysis in §3.
Systematic objection coverage rarely matched in this literature
Interdisciplinary anchoring in actual AI research
Honest scoping and concession structure
Weaknesses
No operational diagnostic despite promising one
The load-bearing macro-possibility thesis is borrowed, not defended
The AI-human disanalogy of full inspectability is unaddressed
Empirical evidence is cited selectively, not weighed
Limited novelty relative to acknowledged precursors
Joint necessity and sufficiency assumed rather than argued
Full Review Report
$1.99 · one-timeUnlock the complete preliminary analysis Evolve produced for this paper.
Report preview
Scope & Claims
The paper (Synthese 206:115, 2025; arXiv 2609.15407) asks whether AI systems can have free will — a question List notes has drawn little attention relative to AI sentience and moral agency (§1). Its stated aim is not to settle whether current systems qualify but to set the criteria (§1). Central claims: (1) free will should be assessed via a checklist of three conditions — intentional agency, alternative possibilities, causal control — assumed jointly necessary and sufficient (§3, after List 2019); (2) the test for each condition is whether we have good explanatory reasons to attribute it, with Dennett's intentional stance reinterpreted realist: explanatory indispensability is evidence of real capacities, not their definition (§4, fn 6); (3) alternative possibilities survive deterministic algorithms via a macro-level notion of agential possibility (§5.2); (4) unpredictability is not required (§5.3); (5) over-attribution to trivial optimizers is blocked by requiring agential explanation to be explanatorily clearly superior (§5.4).
The remaining sections cover Methodology Assessment, Statistical Evidence, Results & Reproducibility, Limitations & Risks, and Verdict & Recommendations.
- All 5 strengths & 6 weaknesses, analyzed in detail
- 3 risk flags with severity, section & reviewer questions
- Methods & statistical evidence assessment
- 5 questions a peer reviewer would ask
- Full six-dimension evaluation report & recommendations
One-time purchase · this paper only
One-time purchase, this paper only. AI-generated preliminary analysis — not peer review.
Frequently asked questions
What does Christian List argue about AI and free will?
In this 2025 Synthese paper, List argues that AI systems can have free will if we have good explanatory reasons to view them as intentional agents that choose between alternative possibilities and causally control their actions. He deliberately does not claim current AI systems definitely have free will; his contribution is the set of criteria for judging the question.
Does List think deterministic algorithms rule out free will in AI?
No. Following compatibilism, he argues that alternative possibilities can exist at the macro-level of agency even when the underlying micro-level computation is deterministic, because the determinism/indeterminism distinction is not preserved across levels of description (section 5.2, drawing on List 2014 and Butterfield 2012).
Does free will in AI imply consciousness or moral responsibility?
No on both counts, in List's framework: free will is conceptually independent of consciousness (section 5.6 treats them as separable, third-personal vs. first-personal), and free will is necessary but not sufficient for moral responsibility, which additionally requires moral cognition (section 5.7).
Why doesn't a chess computer have free will on List's account?
Because its behavior admits fully feasible and simpler non-agential explanations (a deterministic state-transition function over board configurations). List holds free will should be ascribed only when viewing a system as a choice-making agent is explanatorily clearly superior to the alternative description (section 5.4).
About this review
This is an AI-generated preliminary review by Evolve.Pub, produced from the openly available text of the paper. It is not peer review and does not represent the authors' views. The canonical record of the paper is at the publisher link above.
- Review is based on the full extracted text of the arXiv preprint; the published Synthese version (206:115, 2025) may differ in minor pagination or wording.
- As a conceptual paper, assessment criteria are philosophical soundness and argumentative structure rather than empirical reproducibility; no figures or tables existed to analyze.
Go deeper on this paper
Wrote a paper? Stress-test it before reviewers do.
Upload your manuscript for the same structured scrutiny — reasoning, statistics, evidence, figures, and reproducibility.