I have never written a sentence using words I invented.
That seems obvious enough.
The language was here long before me. So were most of the stories, ideas and conventions that have taught me how to use it.
Somewhere in my memory are fragments of books, films, games, conversations and experiences whose origins I can no longer identify.
A particular kind of character might remind me of someone I once knew. An imagined place may contain pieces of several places I have visited. A joke can emerge from something I heard decades ago, transformed enough that I might never recognize the connection.
These influences are not carefully catalogued.
They have become part of the material I think with.
So when somebody asks whether a story is truly original, I am inclined to wonder what kind of originality they have in mind.
The question becomes much more difficult when artificial intelligence enters the process.
But it was already difficult.
A world built from other worlds
Stories rarely arrive without ancestors.
Fantasy is particularly comfortable with inherited traditions.
There are magical creatures, strange territories, dangerous journeys, impossible powers and entire conventions that readers recognize almost immediately.
Some of those ideas have travelled through literature for centuries. Others have developed through games, films and more recent forms of entertainment.
A writer can use familiar elements without necessarily producing a familiar experience.
The same creature might become terrifying in one story, ridiculous in another and strangely sympathetic in a third.
Its significance depends partly on how it belongs to the world surrounding it.
Max Aura Dungeon grew through this kind of creative development.
The game existed before the novels. Ideas that worked within the game’s structure provided possibilities for the fiction, while the novels developed characters, situations and perspectives that could influence later game development.
Moving between those forms did more than change the presentation.
It altered what I could explore.
A character encountered through game mechanics is experienced differently from a character whose private thoughts are available in a novel.
The same world can create different kinds of stories depending on how someone enters it.
I find that kind of transformation more interesting than attempting to identify the first appearance of every individual idea.
But influence also raises questions about artistic independence.
Where does the inherited material end and the individual contribution begin?
The uncomfortable argument from T. S. Eliot
In 1919, T. S. Eliot published Tradition and the Individual Talent, an essay examining the relationship between the individual artist and the literary works that came before.
His argument challenges a common understanding of originality.
For Eliot, the individual achievement of an artist cannot be appreciated entirely in isolation from tradition.
A new work enters a relationship with existing works. It may change how earlier literature is understood, just as earlier literature helps establish the context in which the new work can be appreciated.
There is something appealing about that.
It suggests that artistic creation is part of an ongoing conversation rather than a series of isolated achievements.
But Eliot’s argument goes somewhere less comfortable for my own understanding of creativity.
He also describes a process of artistic depersonalization.
The artist’s task is not simply to reveal their personality or transmit private feelings directly into a work.
Instead, experiences and emotions are transformed into something that acquires its own artistic existence.
That challenges an idea I have expressed elsewhere: that Max Aura Dungeon contains a carefully shaped portion of my own inner world.
Perhaps it does.
But that does not necessarily mean the artistic value of the books lies in how accurately they represent me.
In fact, one of the possibilities I enjoy most is creating people whose thoughts and actions differ substantially from my own.
A character can believe something I reject.
A situation can develop in a direction I find uncomfortable.
A scene may become more compelling when I stop trying to express a particular opinion and allow the fictional circumstances to develop their consequences.
What began as something personal can become something that no longer resembles an account of my personality.
Perhaps that transformation is part of the achievement.
I do not need to adopt Eliot’s entire account of artistic impersonality to recognize the challenge it presents.
It reminds me that the origin of an idea and the artistic significance of what eventually emerges from it are different questions.
The machine enters the conversation
The relationship between influence and originality becomes more contentious when we consider AI systems trained on large collections of human-created material.
An appealing comparison presents human creativity and machine learning as versions of the same basic process.
People encounter stories, learn from them and create new stories.
Machines are trained on existing material and generate new material.
There is enough similarity in that description to make the comparison worth examining.
There are also significant differences.
Human beings encounter language through bodies, experiences, relationships, memories and particular circumstances.
A language model learns through computational training processes that operate on quantities of material far beyond what one individual could encounter in a lifetime.
The resulting systems can produce remarkably convincing language.
But that ability does not establish that the system understands, remembers or experiences its training material in the same way a human reader does.
Nor does it mean every sentence the system generates is a stored copy of something from its training data.
Machine learning is more complicated than retrieving passages from an enormous electronic filing cabinet.
Unfortunately, the opposite simplification is equally misleading.
Research has demonstrated that some language models can reproduce passages from their training material, including text that appears verbatim in the original sources.
That creates concerns about memorization, reproduction, privacy and intellectual property.
It matters how the systems are trained, what they can reproduce and how their outputs are used.
Saying that human writers also learn from books does not settle those concerns.
What the original creator receives
Consider the position of an author whose work is included in an AI training dataset.
The book may represent years of effort.
The author may have revised it repeatedly, worked with editors, negotiated publication agreements and made decisions about how the work should be distributed.
They may depend on income from their writing.
They might be happy for their work to influence future writers while objecting strongly to its use in developing a commercial system.
That distinction is worth taking seriously.
Reading a published book does not give me the right to reproduce its entire contents and sell them as my own.
Being influenced by an author’s style does not make me the author of their work.
And acknowledging that all culture develops through earlier contributions does not automatically establish that every method of collecting and using those contributions is fair.
The scale of AI training changes the practical circumstances.
A company can process enormous quantities of creative material to develop a product that may subsequently compete with the people whose work contributed to its development.
That possibility creates questions about permission, compensation, transparency and the economic conditions under which future culture will be produced.
They cannot be resolved simply by demonstrating that the resulting AI-generated prose is beautiful.
Aesthetic achievement and the legitimacy of the process used to create it are related, but they are not identical.
What the rules can tell us
The legal situation is also more complicated than a simple declaration that training on copyrighted works is either always permitted or always prohibited.
Different jurisdictions have different rules, and the interpretation of those rules continues to develop.
Within the European Union, providers of general-purpose AI models face obligations concerning copyright compliance policies and public summaries of training content.
These obligations recognize the importance of the material used to develop such systems.
They do not, by themselves, resolve every question about the legality of individual training practices or the artistic status of the resulting works.
American copyright discussions have approached related questions through a different legal framework, including debates about fair use and the copyrightability of AI-assisted outputs.
I do not think an essay about art needs to become a legal textbook.
But I also do not want to pretend that the ethical questions are irrelevant simply because they complicate a technology I enjoy using.
Artistic freedom does not eliminate the interests of other people.
And the possibility of creating something interesting is not sufficient justification for every possible means of creating it.
When the work leaves its origins behind
There is another distinction worth preserving.
Suppose an AI-assisted novel is written, revised and published.
A reader finds it compelling.
Later, they discover that a language model contributed to some of the prose.
That knowledge may change how they evaluate the book.
They may become interested in the creative decisions behind it. They may question the originality of its language or the conditions under which the system was developed.
They might decide that the method of production matters more to them than their earlier enjoyment.
Those responses deserve consideration.
But they concern several different aspects of the work.
The text may have artistic qualities.
The human contribution may involve significant creative direction and revision.
The AI system may have contributed substantial language.
And the development of that system may raise ethical questions that neither the quality of the writing nor the intentions of its human author can resolve.
Treating all these questions as a single argument about whether the resulting book is art makes it harder to examine any of them properly.
I would rather acknowledge the complexity than search for a convenient comparison that makes it disappear.
What originality might require
I sometimes wonder whether our understanding of originality places too much emphasis on identifying the first appearance of an idea.
The creative work I find most interesting often involves relationships between things that already exist.
An inherited convention is placed in an unfamiliar situation.
A familiar character type is given a perspective that changes how we understand it.
Something that initially seems ridiculous develops consequences that make the original joke uncomfortable.
The resulting work may be distinctive because of its arrangement, development and interpretation rather than because every component appeared for the first time.
That is not an argument that copying another person’s work without permission is acceptable.
Nor is it a declaration that any sufficiently elaborate rearrangement must be artistically successful.
It is an invitation to consider originality as something more complicated than the absence of influence.
Eliot’s argument makes the problem even more interesting.
If the artist’s achievement can involve transforming personal experience into something that acquires meaning beyond the person who created it, originality may depend as much on transformation as on origin.
The question of AI then becomes more demanding.
We need to examine what is being transformed, how, by whom, and under what conditions.
Those answers will not be identical for every creative process involving a machine.
And none of them can be obtained simply by comparing a human imagination to an artificial neural network.
The inheritance we depend upon
Every new story enters a world in which other stories already exist.
That inheritance makes new creation possible.
It provides language, conventions, ideas, examples and possibilities that no individual could have developed alone.
But the inheritance does not sustain itself automatically.
Books still have to be written. Music has to be composed. Games have to be developed. People must have opportunities to experiment, take risks and devote time to work whose value may remain uncertain.
As technologies change how creative material can be produced and distributed, they also change the conditions under which those activities take place.
That is not a reason to reject every new tool.
It is a reason to pay attention to what the tools make possible, whose work makes them possible, and what happens to the people whose contributions they depend upon.
I use a language I did not invent.
I work within storytelling traditions I did not create.
And I now have access to technologies built from an extraordinary accumulation of human knowledge and creative work.
That inheritance makes my own work possible.
It also gives me good reason to care about what happens to the people who continue creating it.