In part 3 of this series, Gordon Collins explores whether computers can write fiction, looking at Italo Calvino’s “Burning of The Abominable House”. Read part 1: James Meehan’s TALESPIN and part 2: Recurrent neural networks & Andrej Karpathy.

In the introduction to Italo Calvino’s collection Numbers in the Dark, Esther Calvino wrote that, in 1973, her husband received “a somewhat vague request from IBM: how far was it possible to write a story using the computer?” . In response, Calvino wrote the computational crime story The Burning of the Abominable House which is included in Numbers in the Dark. In it, a computer operator at an insurance firm must write a program that solves the mystery of four housemates, all insured against all manner of disasters and all now dead. The only clue is a journal which has survived the fire and tells him that twelve crimes: blackmail, drugging, incitement to suicide, knifing, prostitution, threatening with a gun, tying and gagging, rape, seduction, slander, snooping and strangling have been committed by the four but not who the perpetrators or victims were. The programmer computes the approximately twelve billion possible combinations of crime, perpetrator and victim: A strangles B, C knifes D etc. The question is the mystery solvable. Is it computable?


Computability is a foundational subject in computer science and the battlefield for Alan Turing’s greatest victories.

The question concerns what problems can a computer solve? Addressing the problem required Turing to define precisely the mathematical notion of computer, a perfect machine, one that wouldn’t rust or blue screen but which could be described on paper and would (probably, see Church-Turing thesis) be general enough to describe all computers even those not yet invented— a Turing machine.

The Turing machine

What can a Turing machine do? It can add up and multiply, it can run Tetris or an online dating platform. What can’t it do? It can’t predict the future. There are just too many variables to input. The problem is underdetermined. Also, there are certain problems that are uncomputable. No Turing machine can decide if any program is going to stop or if it will go on forever in an infinite loop. This last result is due to Turing’s ingenious proof of the undecidability of the halting problem. Similar in form to Gödel’s celebrated self-referential proofs of undecidability of arithmetic, Turing’s  proof is by contradiction.

Suppose there is a Turing machine H that can decide if any program stops. Now let’s make a self-referential Turing Machine Z.  A machine that will self-destruct and take H down with it. A machine that takes other machines as inputs and simply says:

If H decides that the input machine stops, then I will go on forever.

If H decides that the input machine goes on forever, then I will stop.

Now simply feed the killer to itself and if H decides that it stops then it doesn’t and if H decides that it doesn’t stop, it does. Contradiction. We can never decide if Z stops.

It turns out that many other problems are uncomputable in this way. What about literature? Is it possible to build a Turing machine that when fed text character by character, it can decide if it is a narrative? Is the problem of reading computable? Is writing? Or is there a Turing/Gödel type killer anti-text which shows that this is impossible?

Can a computer solve the mystery of the burning of the abominable house? Seemingly not, there are just too many combinations. The system is undetermined.  Even if the unlikely is excluded (the actress is unlikely to strangle the wrestler, and there is no point in blackmailing someone who has just been murdered) the possibilities are so numerous that the operator is befuddled. Nevertheless, he can’t help thinking that he can solve the mystery without the computer. “I can’t prevent the slow tentacles of my mind from advancing one hypothesis at a time, exploring labyrinths of consequence that magnetic memories would run through in a nanosecond.”

His breakthrough comes when he inputs Skiller, the insurance salesman and the one who has set the operator the task (just as IBM set Calvino the task?), into the program. It was Skiller who instigated the whole episode. Skiller has his own program, which computed that all the characters would die without being able to claim. This is the first diabolical experiment in a new form of insurance.

But has Skiller been discovered? “My computer has now been fed its opponent’s winning game: so has it won?” asks the operator. Has the operator just fed machine Z to itself? Skiller cannot allow this. Skiller has devised a problem that, by definition, must be undecidable.

“Skiller has already provided for killing me and setting fire to the laboratory: he will destroy the punchcards that accuse him and demonstrate that I lost my life attempting arson.” The mystery must remain unsolved.

About the author

Gordon Collins has been a market risk analyst, a maths lecturer, an English teacher in Japan and a computer graphics researcher specialising in virtual humans. He has degrees in mathematics as well as an MA in Creative Writing from the University of East Anglia. He has been longlisted for the Fish short story prize, and twice for the Galley Beggar Press short story prize. He has had short stories published in Riptide Vol 3, UEA Creative Writing Anthology 2010, Infinity’s Kitchen, Liars’ League, Unthology 3, 6 and 9 and Unthank Books’ The End. See for more.