Also available in French.
A few days ago, I had Claude proofread a text.
Nothing spectacular. I asked it to do what a proofreader would do: spot the small mistakes, the clumsy phrasings, the punctuation problems.
Two seconds later, it came back with something like this:
Commas where a period belongs: lines 146, 267. Recurring tic in the text, worth a look.
Simple and effective.
It reminded me of a story by Isaac Asimov.
Galley Slave
The story is Galley Slave, from 1957, and the only robot in it is called Easy. (Asimov had a knack for robot names: EZ -> Easy, DV -> Dave, SPD -> Speedy, QT -> Cutie, NDR -> AndrewâŚ)
Easy is a proofreading robot. He doesnât do much else: he reads and corrects texts.
The title is a pun on the galleys, the printerâs proofs, the stage where an author sees their sentences take their final shape and reworks them one last time. Thatâs the work Easy takes over.
Early in the story, he gives a demonstration of his abilities. They hand him a book, he leafs through it, and within minutes he reports back.
The comma in line 6 of page 32 is superfluous, whereas one should have been used on line 13 of page 54.
You can guess why I made the connection :)
Itâs already impressive that Asimov imagined a robot that could proofread a text in 1957. For reference, The Mother of All Demos, which shows the first mouse, dates from 1968, and the first word processor from 1976.
But what interests me most is that Easy isnât presented as a robot that can do everything. Heâs just a proofreader, and thatâs deliberate on the part of his makers. In Asimovâs universe, people on Earth are deeply hostile to robots (donât trust the film I, Robot, it has nothing to do with the spirit of Asimov), so U.S. Robots looks for ways to sell its machines despite that hostility.
Easy is a specialized machine, hidden away in a university, doing work that is precise and profitable enough to be accepted.
In 1957, that was science fiction.
Today, if I ask an AI to proofread an article, nobody finds it particularly remarkable.
Asimovâs robot has become ordinary. It has simply lost its metal body and its name.
Ninheimer vs. Easy
Galley Slave is mostly a courtroom story.
Easy is accused of sabotaging the manuscript of Professor Simon Ninheimer. The machine allegedly introduced errors into his work, enough to put his reputation at risk.
In the end, we learn that Ninheimer sabotaged his own book to get the machine blamed and discredit robotics.
So the robot is innocent. But Ninheimerâs final speech is still interesting seventy years later:
A book should take shape in the hands of the writer. [âŚ] One must work and rework and watch the changes take place beyond the original concept even.
He describes the heart of his craft as watching sentences take shape and reworking them. Thatâs precisely the work Easy does for him.
Then he asks:
What would that leave the scholar? One thing only: the barren decisions concerning what orders to give the robot next!
That isnât what happens in the story, since Ninheimer is the one who asked Easy to make the changes. But he already imagines the problem: if a machine does all the work, whatâs left for the author? Mostly, giving it instructions.
Weâve had a word for that for a few years now: prompting.
Ninheimer picks the wrong method, but Asimov never says heâs wrong.
Asimov, by the way, got the idea for the story precisely because he refused to entrust his own proofs to someone else. On the phone with his editor, who was pressing him for a story, he explained that he was too busy correcting the proofs of a textbook he was co-writing. When the editor suggested delegating, Asimov, horrified, answered: âImpossible! I wouldnât trust anyone else!â It was while correcting his own proofs that he imagined a robot to do the job for him. (source)
Thatâs probably why Ninheimer is more interesting than he seems.
Heâs guilty, but his argument isnât necessarily wrong. As in Incredibles 2, where Evelyn (the villain) sees the protection the Supers provide as a source of weakness for humans: a person isnât necessarily wrong just because their method is wrong.
Susan Calvin, who defends the robot, tells Ninheimer he was bound to fail. He replies that he was bound to try. You can already feel the inevitability of adoption.
Easy is still just a proofreader, simple and discreet, not seen as dangerous. That ordinariness may be why I had mixed this story up with âŚThat Thou Art Mindful of Him.
The Georges đ
So in my head, this story was tied to another Asimov story: âŚThat Thou Art Mindful of Him, published in 1974. I remembered in particular a robot whose job was to hunt insects.
In that story, people on Earth are still afraid of robots, and it hurts U.S. Robotsâ sales. So they build two particularly advanced robots, George Nine and George Ten (I call Asimov a visionary, since Georges are always particularly advanced), who look for a solution.
Their idea is fairly simple: stop building robots that look like humans and build specialized, single-task machines instead.
Hence the bird-robot designed to fight insects. It does only that: it targets only one kind of insect and stops when there are none left. Itâs more of a tool than a robot.
It doesnât talk, doesnât want anything, doesnât seem intelligent.
And thatâs exactly whatâs supposed to make humans accept it.
Thatâs roughly how AI is entering our lives: one AI for code, another for sports, shopping, answering the phoneâŚ
Specialized tools rather than a general artificial intelligence.
Bad Georges
But the Georges arenât only trying to make robots acceptable. Theyâre also trying to solve a logical problem raised by the Three Laws of Robotics.
- A robot may not injure a human being or, through inaction, allow a human being to come to harm.
- A robot must obey the orders given it by human beings except where such orders would conflict with the First Law.
- A robot must protect its own existence as long as such protection does not conflict with the First or Second Law.
What should a robot do when two humans are in conflict, or give contradictory orders?
Robots must protect and obey humans, but which humans?
The Georges push the reasoning so far as to redefine what a human is. Since you canât tell who is more human by appearance, they look elsewhere: at intellectual abilities. And since robots are intellectually superior to biological humans, why should the two groups be considered equivalent?
They end up distinguishing the human-beings-like-the-Georges from the human-beings-like-the-others. And so they become the priority under what they call the Three Laws of Humanics.
The Georges donât violate the Three Laws. They reinterpret them. And thatâs where it stops being funny. Like Easy, they arenât evil. But they know what theyâre doing and where their reasoning can lead.
They help us enormously while becoming dangerous to us: a pharmakon, a remedy that is also a poison.
These stories are where the nature of robots becomes truly unsettling, an idea Asimov would develop later in Robots and Empire.
Conclusion
Both stories strike me as much darker than the rest of Asimovâs robot stories, at least in my memory.
Generally, the Three Laws protect humanity, sometimes in spite of itself, to the extremes reached by R. Giskard Reventlov and R. Daneel Olivaw.
Easy protects nobody. He corrects commas. And maybe thatâs what I like about this story. In 1957, Asimov imagined a machine that could do very precise intellectual work, well enough that humans would end up trusting it, and he predicted the consequences.
Today, I ask an AI for the same thing.
Postscript
The draft of this article was written by the AI assistant that also fixed my commas. Somewhere, Ninheimer is watching me and sighing. He can rest easy: I rewrote almost every sentence.