Suppose one holds hard determinism to be true: no one is causa sui. Capacities, opportunities, and choices are downstream of causes none of us authored. If that premise holds, backward-looking claims of ultimate moral desert in scholarship – who “really” deserves credit for a paper – lose their footing. Introducing AI does not change the metaphysics of production; before and after AI, no researcher is the unmoved mover of their work. The normative question, therefore, shifts from purity of authorship to the design of rules that predictably improve the truth, reliability, and usefulness of research outputs.

Within that frame, the case is for full-stack freedom to use AI across the research process. Researchers should be free to consult AI for everything – brainstorming and problem selection; surveying and organizing literature; designing identification strategies and experimental protocols; collecting data; modeling, estimation, simulation, and diagnostics; visualization; writing and formatting manuscripts; preparing submissions, etc. Evaluation should be output-first: does the resulting work better explain, predict, or solve? Auditability is the supporting infrastructure that lets others verify, learn from, and build upon those outputs.

A determinist view clarifies the role of praise and blame. These can be justified not as metaphysical verdicts but as instruments that create incentives. We build institutions that treat researchers as if they can be praised or sanctioned because these signals causally shape behavior toward better results. Freedom is paired with ex post accountability. If AI-assisted choices yield less innovative and less useful research or claims that fail replication or violate standards, then reputational loss, rejection, retraction, and funding penalties follow; if AI-assisted workflows produce clearer arguments, more credible estimates, stronger predictive performance, and more practically useful knowledge, they are rewarded. Bad consequences are borne by the individuals who made the choices, which gives strong incentives to use AI responsibly. That logic differs from restricting everyone’s freedom because some may misuse the tool or because of a general distrust of it.

On a deterministic view, the common claim that it is unethical to use AI to write a paper mistakes the locus of wrongness. There is no moral taint from employing a powerful instrument (or, for that matter, human research assistants), because no one ever produced work ex nihilo anyway. What matters is whether the practice predictably improves outputs – truer inferences, better predictions, clearer exposition, more innovative and useful ideas. If “human + AI” achieves that better than a mere human, it is to be welcomed.¹

Likewise, the claim that it is unethical to use AI without declaring it conflates confession with transparency. On a deterministic-consequentialist view, obligatory declarations have no intrinsic moral force. What is morally relevant is provenance that enables verification: data, code, computational environments, and, when material for results, workflow or prompt logic at the level required for replication. Non-declaration is wrongful when it obstructs auditability – not because it withholds a metaphysical truth about authorship. If one maintains human answerability at identifiable control nodes and documents what is needed to reproduce results, then deception concerns are addressed without mandating ritual disclosures about tool use.

Freedom with accountability lets decentralized actors explore how AI complements their judgment, while ensuring that poor decisions are privately costly. It preserves the gains from experimentation and selection and, by keeping outputs testable, channels those gains into cumulative progress. The criterion for success remains outcome-focused: better theories, tighter identification, stronger predictive validity, more useful knowledge.

The resulting ethic is spare and practical: we should grant researchers complete freedom to deploy AI throughout the research process; ground responsibility in forward-looking incentives; replace moralized talk of “authentic” production (which would preclude thorough AI use) with output-focused evaluation supported by verifiable methods. If none of us originates ourselves, then the ethical salience of adding AI evaporates. What remains is ”institutional engineering” aimed at a simple end: more useful, more reproducible knowledge, produced by answerable humans using the best available tools.²

1. It may be that a non-determinist can grant that praise and credit matter while insisting that what merits them in science is not who wrote a research paper but ownership of reasons. The author is the agent who sets the aim, selects and justifies the method, curates and vets the evidence, adjudicates alternatives, and can defend the conclusions under adversarial probing. On that standard, AI assistance in drafting or analysis does not extinguish authentic authorship; it is a medium through which reasons are developed and expressed. If a researcher delegates so completely that they cannot give reasons or withstand challenge, they fail the authorship test and deserve sanction. If, by contrast, they direct the inquiry, audit the outputs, and can account for each inferential step, then credit is warranted – even if AI improved the final product. Praising such work is not at odds with free will; it rewards the exercise of reasons-responsive agency, including the wise use of powerful instruments. The policy upshot is unchanged: allow full-stack AI use, but tie praise and blame to demonstrable human control and answerability. ↩︎
2. A note on vocabulary: terms like “should”, “freedom”, and “responsibility” are used here instrumentally, not metaphysically. “Should” is conditional – if the objective is to maximize social value, then certain rules follow. “Freedom” denotes the absence of ex ante tool bans; it does not posit contra-causal agency. “Responsibility” means forward-looking answerability at identifiable control nodes, justified because it steers future behavior and improves outcomes. This is normative language in a policy-analytic mode, consistent with determinism. ↩︎