In 1988, Bantam Books published A Brief History of Time with Stephen Hawking as its sole author.
By then Hawking could no longer speak. After his tracheotomy in 1985, he had first communicated through a spelling card. In 1986 he received Equalizer, software developed by Walter Woltosz of Words Plus. Hawking could select words and phrases from the computer and send the resulting text to a speech synthesizer.
The computer was not merely his voice.
It was also one of the instruments through which he wrote.
In the United States, Bantam put Hawking’s name on the cover.
That seems obvious.
It is worth noticing that it seemed obvious.
No second author appeared beside him. The software did not acquire a byline. The computer was treated as an instrument through which Hawking expressed his work.
The answer was already on the cover.
Only much later would we begin asking the question.
The machinery underneath that answer did not remain still.
Hawking’s physical control continued to diminish. Eventually a sensor mounted on his glasses detected movement in his cheek, allowing him to operate his communication system through a single voluntary movement.
The bottleneck had become extraordinary.
Then the machine began doing more.
In 2014 Intel introduced ACAT, the Assistive Context-Aware Toolkit developed for Hawking. SwiftKey built a personalized language model from extensive examples of Hawking’s own writing, including his research papers.
The machine learned to predict Hawking from Hawking.
If he began typing black, the system might offer hole.
The improvement was substantial. Hawking could produce text while physically entering fewer than 20 percent of its characters.
That sounds, by today’s standards, like a large amount of machine participation.
But notice what did not happen.
Stephen Hawking did not become 20 percent of an author.
The machine’s increasing contribution to the physical production of language did not produce a corresponding reduction in the attribution of the work to him.
That tells us something, although we should be careful about what.
It does not prove that Bantam held a meeting and decided predictive software was irrelevant to authorship.
Perhaps no such question was ever asked.
That absence may itself matter.
The available public record may show something simpler: as the machinery between Hawking’s intention and the finished sentence became more sophisticated, the institution publishing his work continued to present the intellectual source of that work as Stephen Hawking.
The interface changed.
The attribution did not.
That is an interesting fact to carry into the present argument about artificial intelligence.
We tend to speak about “AI-assisted writing” as though the presence of machine-generated or machine-predicted language settles the authorship question.
Hawking’s case suggests that we have not always located authorship by counting physical contributions to the sentence.
We located it somewhere else.
In intention.
In judgment.
In intellectual direction.
In the person deciding what belonged.
The machine could supply characters Hawking had not individually selected.
It could predict words from context.
It could learn his patterns from a corpus of his previous writing.
None of those facts, by themselves, displaced him from the cover.
Perhaps that was because the machine was understood to be shortening the distance between Hawking and the language he was trying to produce.
It was an interface.
But that word does a great deal of work.
There is a useful limit case.
Hawking died in March 2018.
Later that year, Brief Answers to the Big Questions appeared.
In the United States, Bantam again published the book under Hawking’s authorship.
This time the problem was no longer only technological.
Hawking could not approve the final manuscript.
The book had been in development at the time of his death and was completed from material in his archive by colleagues, family, and the Stephen Hawking Estate.
Other people necessarily stood between Hawking’s surviving material and the finished book.
Yet the attribution still pointed toward him.
That does not make predictive software and posthumous editing the same thing.
They are not.
It does show how stubbornly authorship can remain attached to the perceived intellectual source of a work even when other people and machines participate in bringing the final text into existence.
The cover again supplied an answer.
Stephen Hawking.
Perhaps that is where the history becomes useful.
The current AI debate often begins with the machine.
How much did it write?
How many words did it supply?
Was prediction involved?
Was generation involved?
Those are necessary questions.
But publishers have been dealing with mediated authorship for a long time.
Editors rewrite.
Copy editors correct.
Translators reconstruct.
Dictation separates the physical production of words from the person whose words they are.
Assistive technologies predict and complete.
Posthumous editors assemble material an author can no longer approve.
The existence of mediation has never, by itself, answered the authorship question.
Something else has mattered.
Whose work is this?
Whose thought is being expressed?
Who exercises the judgment that gives the material its direction?
Those questions become harder with generative AI because the machine can now do something Hawking’s predictive system generally did not.
It can introduce intellectual material.
It can propose the argument rather than merely anticipate the word.
It can produce an analogy the writer had not conceived.
It can challenge a premise.
It can change the direction of a piece.
That may eventually require different answers about authorship.
But if so, we should know what changed.
In 1988, nobody needed the language of generative AI to decide whose book A Brief History of Time was.
The institution made the attribution.
It printed the answer on the cover.
Decades later, we arrived with the question.
Perhaps that is the useful place to begin.
Not:
Was a machine involved?
We already know machines can be deeply involved without settling authorship.
The better question is the one the cover forces us to ask:
What did the machine contribute that would make us put another name there?
WE&P by: EZorrillaMc&Co.
Stephen Hawking Estate — “Stephen Hawking: My Communication System.” Covers the 1985 tracheostomy, spelling card, Equalizer, hand switch, cheek sensor, ACAT, and speech synthesizer.
Intel — “New Intel-Created System Offers Professor Stephen Hawking Ability to Better Communicate with the World” (December 2014). Documents ACAT, SwiftKey integration, language prediction, and the finding that Hawking needed to type less than 20 percent of all characters.
Stephen Hawking Estate — Biography. Covers Hawking’s communication system and its role in his writing and scientific work.
Penguin Random House / Bantam — A Brief History of Time. Lists Stephen Hawking as sole author and Bantam as the U.S. publisher.
Penguin Random House / Bantam — Brief Answers to the Big Questions. Lists Stephen Hawking as author and Bantam as the U.S. publisher; published October 16, 2018.
Stephen Hawking Estate — Biography and material concerning Brief Answers to the Big Questions. Documents that the project was underway during Hawking’s lifetime and was completed after his death from his materials with the involvement of colleagues, family, and the Stephen Hawking Estate.
Travel in Life 