Can an AI System Be Creative? A Critical Perspective from Art and Engineering

Ivan Magrin-Chagnolleau

REVIEW 4 major objections 5 minor 7 references

Can an AI System Be Creative? A Critical Perspective from Art and Engineering

T0 review · 4 major / 5 minor · reviewed 2026-08-01 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read The paper claims that AI systems, by their data-dependent and probability-driven construction, are structurally unable to achieve creativity of the strongest kind, though they can productively recombine and be constrained by a human expert.

desk verdict A thoughtful, honest position paper on AI and creativity, but the central impossibility claim is heavily definitional and overreaches from current generative models to all AI. read the letter →

arxiv 2607.20796 v1 pith:4MTHSUDF submitted 2026-07-22 cs.AI cs.HCcs.MM

Ivan Magrin-Chagnolleau This is my paper · ORCID

classification cs.AIcs.HCcs.MM

keywords artificialintelligencecreativitytransformationalserendipitychancehuman-AIcollaborationgenerativeAIphenomenologyof

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper asks whether an AI system can be creative and answers with a graduated 'no': AI genuinely performs combinatorial creativity, can explore a defined space only without sensing its edges, and cannot perform transformational creativity at all. The central claim is that the deepest obstacle is not lack of novelty but the absence of two things: genuine accident or serendipity, and a subject position from which an unexpected event could be recognized and welcomed as meaningful. Because an AI system is a probability engine over a fixed corpus of human-made content, its outputs are, in the author's terms, always expected in principle. The paper concludes that the locus of creativity in human-AI collaboration remains the human, who evaluates, selects, and can be redirected by the machine's output. The paper itself is presented as a demonstration: the AI organized and drafted, the human conceived and judged.

What carries the argument

The argument's engine is a stipulated definition of an 'AI system' as a statistical generator trained on a fixed corpus of existing human content, whose output is a probability-weighted reconstruction of that corpus. Around this definition the paper places a three-type taxonomy of creativity — combinatorial, exploratory, transformational — and a phenomenological condition: creativity requires genuine chance events and a human subject ready to perceive and follow them. The combination does the work: each type of creativity is assessed against what such a system can and cannot do, and the transformational type fails because the system has neither a position outside its data-space nor a witness

What would settle it

A single documented case where an AI system, without human curation, introduces a new enabling constraint into its own conceptual space—for example, after encountering an anomaly, it changes its generative rules or goals and produces a work that human experts recognize as a break with its training distribution—would falsify the central claim. At a smaller scale, a controlled experiment in which a language model 'notices' a deliberately planted contradiction, reorients its subsequent responses accordingly, and treats the moment as a meaningful discovery would test the paper's asserted absence o

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is a structural diagnosis rather than an empirical result: an AI system, by definition and by construction, is a function of pre-existing human-generated data, so nothing it produces can be historically novel; its statistical design pulls output toward the most probable rather than the surprising; it cannot judge value because it has no aesthetic or ethical standpoint; and it contains no mechanism for accident and no conscious subject to receive an accident as an opportunity. Applying a standard threefold taxonomy of creative processes, the paper grants AI real capability in combinatorial combination, a bounded and edgeless form of exploration, and out

Load-bearing premise

The paper's conclusion rests on defining an AI system as a probabilistic generator over a fixed human-made dataset; if AI is taken to include systems that act in the world, form their own goals, or possess subjective states, the claim that they cannot be creative would not follow.

Editorial extensions

If this is right

If the paper's diagnosis is right, AI's best creative role is as an ideation engine whose output is valuable only when a human expert evaluates, selects, and redirects it.
AI systems can serve as productive constraints, forcing a human artist to adopt unfamiliar paths—exemplified by the film-score experiment in which an algorithm's luminosity curve changed the composer's process without composing the music.
The statistical tendency of AI outputs pulls toward the average and the cliché, so over-reliance on AI as a creative partner risks flattening original work unless the human actively resists that pull.
Transformational creativity—changing the fundamental rules of an artistic or scientific domain—should not be expected from current AI systems and remains a distinctly human, accident-tolerant act.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

If the paper's definition of AI is relaxed to include systems that act in the world, form their own goals, or possess subjective experience, the central claim would not follow; the conclusion is therefore narrower than it may first appear.
A testable extension would be to deliberately inject pseudo-random accidents into an AI's generative process—noise, contradictory prompts, or novel constraints—and measure whether human users can convert those jitters into productive discoveries; this would formalize the paper's 'serendipity loop'.
The paper's emphasis on the human growing through collaboration, while the machine does not, suggests a practical metric for creative tools: a tool's worth may be gauged by how much it changes the practitioner's skills and intentions, not by the standalone quality of its outputs.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

The genuinely new thing here is the fourth limitation: the absence of real accident/serendipity in AI systems combined with the absence of a subject position to recognize and welcome chance. The paper argues that this compounds the evaluation deficit and is the deepest reason AI can't be creative in Boden's strongest sense. That framing, with the concrete artistic examples and the human-AI collaboration experiments, is a real contribution to the computational creativity debate. The paper also does something rare: it takes Boden's taxonomy seriously, applies it in a graduated way, and honestly documents its own composition through human-AI collaboration. The four experiments are illustrative rather than controlled, but they are transparent and the qualitative observations are plausible.

The soft spot is the one the stress test flags, and it is real. The central claim—AI systems are structurally incapable of transformational creativity—depends on defining an AI system as a data-bound statistical generator over a fixed human corpus (Section 4.1). That definition fits today's LLMs and image generators, but it excludes RL agents, open-ended evolutionary systems, and any future system with its own goals and world model. The conclusion is therefore partly analytic: true of the class the author defines into the premise, not demonstrated for all AI systems. The paper would be much stronger if it claimed "current generative AI systems" rather than "AI systems." The appeal to Schapiro et al. (2025) doesn't close the gap: that formal result shows axiom changes carry transformative potential in a DAG model; it doesn't prove statistical learners can't implement such changes. That said, the paper's narrower claims—that current generative models are combinatorial engines without genuine exploratory edge-recognition or self-evaluation—hold up well and are worth engaging.

Minor issues: Experiment 2's unrecorded model version is acknowledged but remains a reproducibility gap; the use of "invented_entities" is too harsh—"subject position" is a phenomenological term with literature behind it, not an hallucination. The paper's self-referential method note is transparent and actually supports the argument.

Who is this for? Anyone working on computational creativity, human-AI collaboration, or philosophy of AI. It deserves a serious referee, especially for a venue that welcomes interdisciplinary position papers. A good referee should press the author to scope the claims to current systems and to justify or drop the "by definition and by construction" language, but the core thesis—that the missing subject position for accident is the decisive deficit—is argued with care.

Recommendation: send it to peer review, with the definitional overreach as the main revision target.

Referee Report

4 major / 5 minor

Summary. This paper argues, using Margaret Boden's taxonomy of creativity (combinatorial, exploratory, transformational), that AI systems are structurally incapable of creativity in its strongest sense. It distinguishes genuine capability in combinatorial creativity, bounded competence in exploratory creativity, and fundamental incapability in transformational creativity. The two load-bearing reasons are: (i) AI systems are 'data-bound' statistical generators that cannot produce outputs without precedent in a fixed training corpus, and (ii) AI systems lack a 'subject position' from which accidents and unexpected events can be recognized and welcomed. The paper proposes a human-AI collaborative model centered on displacement/constraint and illustrates it with four qualitative experiments, including haiku composition, literary pastiche, algorithm-guided music, and iterative image generation. The paper is self-consciously written with AI assistance and presents itself as a demonstration of its own thesis.

Significance. If the argument is read as applying to current generative AI systems rather than to all possible AI systems, the paper makes a useful contribution. Its graduated analysis of the three Boden types is clear and defensible, its distinction between generation and evaluation is important, and its focus on accident, serendipity, and the human capacity to be redirected by the unexpected is a valuable corrective to purely statistical accounts of creativity. The four experiments are honest, concrete, and instructive, and the model of AI as a 'generator of productive accidents' or constraint is a genuinely generative proposal. The paper's main weakness is that the central claim overgeneralizes: the definition of 'AI system' in §4.1 is tailored to today's data-driven generative models, so the conclusion of structural incapability is partly analytic. This is reparable by careful re-scoping, but as written the strongest claim is unsupported.

major comments (4)

[§1, §4.1, Abstract] The central claim—'AI systems are, by definition and by construction, incapable of creativity in its strongest sense'—is loaded into the paper's own definition of an AI system. §4.1 states that an AI system 'relies on a large dataset of pre-existing human-generated content' and 'outputs are produced by identifying statistical patterns.' This describes contemporary supervised/generative models, not artificial intelligence in general. It excludes reinforcement-learning agents that generate their own experience through interaction (e.g., AlphaZero-style self-play), open-ended novelty search systems, embodied/interactive systems, and hybrid symbolic-statistical architectures. Because the paper derives 'structural incapability' from this stipulated data-bound property, the conclusion is partly analytic for the stipulated class. The paper should either explicitly restrict its scope to 'current
[§5.3, §2] The paper states that Schapiro, Black, and Varshney (2025) 'prov[e] that modifications to the axioms of a conceptual space carry the highest transformative potential—precisely the kind of operation that current AI systems are constitutively incapable of performing,' and later says this 'confirms this intuition with mathematical precision.' As described in the paper, the cited result concerns a graph-theoretic model in which axiom modifications have high transformative potential; it does not, in itself, establish that a statistical learner cannot perform such modifications. The inference from 'axiom changes have high transformative potential' to 'statistical models are least capable of making them' is not supplied. If this formal result is load-bearing for the transformational-creativity claim, the paper must state the theorem and show how it applies to statistical learners. Otherwise the
[§4.4] The argument depends crucially on the phenomenological premise that 'there is no chance without a subject' (Bergson/Valéry) and that an AI system has 'no subject position from which to recognize the accident as significant.' This is a contested philosophical thesis, not an established empirical fact. The paper presents it as if it were uncontroversial and uses it to conclude that AI cannot participate in accident-driven creativity at all. If the premise is rejected—or if one holds that a subject position could be instantiated by an evaluative feedback loop rather than by phenomenal consciousness—the conclusion does not follow. The author should mark this as an explicit assumption and distinguish the weaker claim ('current AI lacks human-like readiness to attribute meaning to accidents') from the stronger claim ('AI has no subject position whatsoever'). The stronger claim requires argumen
[§4.2, §6.1 Experiment 4] The paper's treatment of stochasticity is too quick. It says random noise introduced into generative processes is 'controlled unpredictability' and therefore 'a form of expectedness.' But a high-entropy distribution can produce an output with very low prior probability; whether that output is 'expected' is not the same as whether it was sampled from a stochastic process. Moreover, the paper's own Experiment 4 (the Ideogram 'Pollock' intrusion) is an example of an AI-generated output that the human did not anticipate and accepted as a productive surprise. The paper attributes all creativity in that loop to the human's acceptance, which is a coherent interpretive stance, but it is not a consequence of the architecture; it is a philosophical decision about where to locate creativity. This should be acknowledged explicitly rather than presented as a structural fact.

minor comments (5)

[Methodological note] The claim 'The AI neither conceived the thesis nor evaluated whether the paper has merit' is an assertion about the author's private creative process, not something the reader can verify. Since the paper uses this as a 'lived illustration' of its thesis, it should be presented as first-person testimony, not as a demonstration. The same applies to the claim that 'every decision about structure, emphasis, argument, and expression was made—and validated—by the human author.'
[§4.2] The phrase 'the output that maximizes probability is, by definition, the most average output' is imprecise. Maximizing likelihood under a learned model is not the same as averaging, and the notion of 'average' is not defined. Suggest replacing with a more careful statement about the mode of the learned distribution and its tendency toward corpus-typical outputs.
[§7] The phrase 'at present and by structural necessity' is internally awkward: if the limitation is structural, the qualification 'at present' is unnecessary; if it is temporary, the claim is not structural. The author should decide which claim is being made and apply it consistently throughout.
[§6.1, Experiment 2] The author acknowledges that the exact ChatGPT version was not recorded, which is honest but limits reproducibility. Since the paper's experiments are qualitative and version-sensitive, a short reproducibility appendix or version-logging recommendation would strengthen the methodological contribution.
[References] Several references are to arXiv preprints or in-press works without full bibliographic details. Please standardize and, where possible, include DOIs or stable identifiers. Some citations lack page numbers or publication dates that would help readers verify the sources.

Circularity Check

2 steps flagged · score 6.0 of 10

The central impossibility claim is partly analytic: 'AI system' is stipulated in §4.1 as a data-bound statistical generator, so 'cannot produce historical novelty' follows by definition; the subject-position premise is asserted rather than derived.

self definitional [Section 4.1 (The Data-Bound Problem: Novelty) and Section 1]
"An AI system, by definition and by construction, relies on a large dataset of pre-existing human-generated (and sometimes AI-generated) content. Its outputs are produced by identifying statistical patterns in that dataset and generating responses consistent with those patterns. This means, structurally, that everything an AI system produces already exists — in some form — within its training data."

The paper's thesis ('AI systems are, by definition and by construction, incapable of creativity in its strongest sense') is loaded into this stipulation. If 'AI system' is defined as a function of a fixed pre-existing corpus, then 'no historical novelty' is entailed by definition; it is not an empirical finding about every AI system. The argument never shows that self-play agents, interactive robots, or open-ended evolutionary systems must be data-bound in this sense, so the universal conclusion is partly analytic.
self definitional [Sections 3.1 and 4.2 (The Statistical Convergence Problem: Surprise)]
"Genuine surprise, in Boden's sense, requires that the output not be what we would have predicted. But an AI system trained to produce statistically probable outputs is, by design, producing exactly what the data predicts."

Once 'AI system' is defined as a probability engine that selects from high-probability regions, and 'surprise' is defined as 'not what we would have predicted', the absence of genuine surprise follows analytically. The argument conflates statistical sampling (which can produce rare events) with prediction, making the failure definitional rather than demonstrated from the architecture of all AI systems.

full rationale

The paper builds a graduated analysis of AI creativity on Boden's taxonomy and four documented experiments; those parts are largely self-contained and not circular. However, the strongest claim—'AI systems are, by definition and by construction, incapable of creativity in its strongest sense'—is substantially analytic. Section 4.1 defines an 'AI system' as a system that 'relies on a large dataset of pre-existing human-generated content' and generates outputs from statistical patterns in that data; from that stipulation, the conclusion that it cannot produce historical novelty follows by definition. Similarly, §4.2 defines surprise as unpredictability and defines AI as a probability engine, so missing surprise is built into the premises. The §4.4 claim that AI has 'no subject position' is asserted (drawing on the author's 2021 self-citation) rather than derived from a general theory of artificial intelligence. Because these premises are contestable stipulations—not empirical findings—the universal impossibility result is not fully supported for all AI systems, though it is valid for the class of systems the paper defines into the premise. The paper's collaborative experiments and its proposal for human-AI co-creation retain independent content; hence the circularity is partial and definitional, not a fabricated statistical fit. Score 6.

Assumptions & free parameters 0 free parameters · 5 assumptions · 1 invented entities

No numeric parameters are fitted; the argument's load-bearing choices are conceptual and definitional, listed under axioms. The 'subject position' is the main invented conceptual entity the paper depends on.

assumptions (5)

domain assumption Boden's definition of creativity (novelty, surprise, value) and taxonomy (combinatorial, exploratory, transformational) are the best available framework for judging AI creativity.
The paper structures the entire argument around Boden's framework (§3); alternative definitions of creativity (process-based, sociocultural) would change the assessment.
ad hoc to paper An AI system is, by definition and by construction, a system that derives outputs from statistical patterns in a fixed pre-existing dataset.
Introduced in §4.1 ('by definition and by construction'); this stipulation excludes autonomous/world-interacting systems and makes the conclusion partly analytic.
domain assumption Genuine historical novelty cannot arise from any function of a finite dataset.
Stated in §4.1 to argue AI cannot be novel ('everything an AI system produces already exists'); this is a philosophical claim, not a theorem.
domain assumption There is no chance without a subject: an accident is meaningful only when perceived by a consciousness (Bergson, Valéry).
Invoked in §4.4 and quoted from the author's prior work; it is the basis for the 'no subject position' deficit.
domain assumption Transformational creativity requires restructuring the axioms of a conceptual space, an operation statistical models are least capable of performing (via Schapiro et al. 2025).
Used in §5.3; the cited paper provides a graph-theoretic formalization, but the conclusion about current AI models is an extrapolation.

invented entities (1)

subject position
purpose: The capacity to perceive chance events as significant and incorporate them; the paper argues this is structurally absent in AI, compounding the absence of genuine accident.
The paper relies on this philosophical construct (§4.4) to argue that even stochastic noise would not be 'genuine accident' for an AI; it is not independently measurable or falsifiable.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Can an AI System Be Creative? A Critical Perspective from Art and Engineering." pith.science (2026). https://pith.science/paper/4MTHSUDF

@misc{pith2026260720796,
  author       = {Pith},
  title        = {Pith review of: Can an AI System Be Creative? A Critical Perspective from Art and Engineering},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4MTHSUDF}},
  note         = {Machine review of arXiv:2607.20796}
}

read the original abstract

This paper examines the question of whether artificial intelligence (AI) systems can be creative, approached from the dual perspective of a researcher trained in electrical engineering, pattern recognition, machine learning, and neural networks, who has also spent most of his life engaged in the arts as actor, stage and film director, writer, composer, and visual artist, and in philosophy. Drawing on Margaret Boden's foundational framework, both her three properties of creativity (novelty, surprise, and value) and her three types of creative processes (combinatorial, exploratory, and transformational), the paper argues that AI systems are structurally incapable of creativity in its strongest sense. While they exhibit genuine capability in the domain of combinatorial creativity, they are significantly bounded in exploratory creativity, and fundamentally incapable of transformational creativity. The paper further argues that the most important limitation of current AI systems is not the absence of novelty per se, but the absence of any mechanism for serendipity, accident, or the unexpected, all of which play a central role in the phenomenology of creativity, and the absence of any subject position from which to recognize and welcome such chance events. The paper concludes by proposing a model of human, AI creative collaboration that is both realistic and generative, illustrated by several concrete experiments. The paper is itself a demonstration of the thesis it advances: it was composed through a deliberate human AI collaborative process, which is described in the methodological note that opens it.

Figures

Figures reproduced from arXiv: 2607.20796 by the authors.

**Figure 2.** Handscape #31 The third, perhaps the most extreme: six beginner watercolors stored in a cardboard folder in an attic for fifteen years. When I found them again, humidity and capillarity had "composed" an entirely new watercolor on one of the blank sheets — a kind of meditative landscape that I could never have planned or predicted [Magrin-Chagnolleau 2021]. From the point of view of the lived experience, this was ge… view at source ↗

**Figure 3.** Sedimented Landscape These examples illustrate a principle that runs through the literature on creativity and chance: chance must be provoked. As Romain Rolland writes, "chance always knows how to find those who know how to use it" — and Albert Camus speaks of "giving opportunities to this chance which, too often, is disturbed only when provoked" [Magrin-Chagnolleau 2021]. This is not a passive observation. It means… view at source ↗

Figures from the paper (1 more)

**Figure 5.** Figure 5: Synthetic overview of the paper's argument generated by AI. (a) Four empirical experiments illustrating different modes of human-AI creative collaboration (Section 6.1): (1) haiku composition — AI as ideation partner, human evaluates (autumn 2025, ChatGPT); (2) autobio…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

7 extracted references · 1 linked inside Pith

[1]

AI is now creative!

Introduction The question of whether machines can be creative has been debated since long before artificial intelligence became a recognized field of research [Moruzzi 2025]. It has, if anything, become more urgent with the recent emergence of large language models and generative AI systems capable of producing text, images, music, and code at remarkable ...

2025
[2]

Creativity and Artificial Intelligence

State of the Art The scholarly literature on AI and creativity is extensive and growing rapidly. What follows is a focused overview of the contributions most directly relevant to the argument developed here. The foundational work is that of Margaret Boden (1936–2025), Research Professor of Cognitive Science at the University of Sussex, who devoted much of...

arXiv 1936
[3]

Creativity is the ability to come up with ideas or artefacts that are new, surprising and valuable

Defining Creativity: Boden's Framework 3.1 The Three Properties of Creativity Boden's definition of creativity is, in her own formulation, deceptively simple: "Creativity is the ability to come up with ideas or artefacts that are new, surprising and valuable" [Boden 2004]. Each of these three properties deserves careful attention, because each will serve ...

2004
[4]

AI system

The Structural Limitations of AI Systems Before analyzing AI's performance across Boden's three creative types, it is necessary to establish clearly what AI systems actually are. I use the term "AI system" deliberately, rather than "AI model" or "AI algorithm," to acknowledge the complexity of contemporary deployments: what a user interacts with today is ...

1998
[5]

The answer, as I noted at the outset, is graduated rather than binary

AI and the Three Types of Creativity With these structural limitations established, we can now return to Boden's taxonomy and ask, for each type of creative process, what AI systems can and cannot do. The answer, as I noted at the outset, is graduated rather than binary. 5.1 Combinatorial Creativity: Where AI Has Genuine Foothold Combinatorial creativity ...

2004
[6]

best, most mindblowing

AI as Creative Collaborator The analysis in the preceding sections might appear to suggest that AI systems are, for creative purposes, essentially useless. That would be an overcorrection. The argument is not that AI systems have no role in creative processes; it is that their role is that of a collaborator, not a creator — and a collaborator of a specifi...

2025
[7]

can an AI system be creative?

Conclusion The question "can an AI system be creative?" does not admit of a simple yes or no. It requires, first, a precise definition of creativity. Boden's framework — distinguishing three properties and three types of creative process — provides the best available precision. With that framework in place, the answer becomes graduated and specific. In th...

arXiv 1964

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.

[1] [1]

AI is now creative!

Introduction The question of whether machines can be creative has been debated since long before artificial intelligence became a recognized field of research [Moruzzi 2025]. It has, if anything, become more urgent with the recent emergence of large language models and generative AI systems capable of producing text, images, music, and code at remarkable ...

2025

[2] [2]

Creativity and Artificial Intelligence

State of the Art The scholarly literature on AI and creativity is extensive and growing rapidly. What follows is a focused overview of the contributions most directly relevant to the argument developed here. The foundational work is that of Margaret Boden (1936–2025), Research Professor of Cognitive Science at the University of Sussex, who devoted much of...

arXiv 1936

[3] [3]

Creativity is the ability to come up with ideas or artefacts that are new, surprising and valuable

Defining Creativity: Boden's Framework 3.1 The Three Properties of Creativity Boden's definition of creativity is, in her own formulation, deceptively simple: "Creativity is the ability to come up with ideas or artefacts that are new, surprising and valuable" [Boden 2004]. Each of these three properties deserves careful attention, because each will serve ...

2004

[4] [4]

AI system

The Structural Limitations of AI Systems Before analyzing AI's performance across Boden's three creative types, it is necessary to establish clearly what AI systems actually are. I use the term "AI system" deliberately, rather than "AI model" or "AI algorithm," to acknowledge the complexity of contemporary deployments: what a user interacts with today is ...

1998

[5] [5]

The answer, as I noted at the outset, is graduated rather than binary

AI and the Three Types of Creativity With these structural limitations established, we can now return to Boden's taxonomy and ask, for each type of creative process, what AI systems can and cannot do. The answer, as I noted at the outset, is graduated rather than binary. 5.1 Combinatorial Creativity: Where AI Has Genuine Foothold Combinatorial creativity ...

2004

[6] [6]

best, most mindblowing

AI as Creative Collaborator The analysis in the preceding sections might appear to suggest that AI systems are, for creative purposes, essentially useless. That would be an overcorrection. The argument is not that AI systems have no role in creative processes; it is that their role is that of a collaborator, not a creator — and a collaborator of a specifi...

2025

[7] [7]

can an AI system be creative?

Conclusion The question "can an AI system be creative?" does not admit of a simple yes or no. It requires, first, a precise definition of creativity. Boden's framework — distinguishing three properties and three types of creative process — provides the best available precision. With that framework in place, the answer becomes graduated and specific. In th...

arXiv 1964