What is Do Androids Dream of Electric Sheep? about, and why does it matter for AI?

Most people meet this story through its film adaptation, Blade Runner, with its rain-soaked neon city. The novel underneath is quieter, stranger and more unsettling. Its world has almost no animals left, its people find comfort in a shared religious machine, and its hero spends a single long day hunting beings who may feel more than he does. At its center sits a test, and the question of whether any test can tell a person from a perfect copy.

Philip K. Dick (1928–1982) was an American writer who lived most of his life in California and turned out an enormous body of science fiction between 1950 and 1970. He wrote this novel in 1966, and Doubleday published it in Garden City, New York, in 1968. It was nominated for the Nebula Award that year. In 1982 it became the main source for Ridley Scott's film Blade Runner, and some later printings of the book carry that title.

The story

The novel is set after a nuclear conflict called World War Terminus. In the first edition the year is 1992; later editions move it to 2021. Radiation has wiped out most animal species and polluted the air, so the United Nations urges people to emigrate to colonies on other worlds. As an incentive, each emigrant is promised a personal android servant.

The androids are not clanking metal robots. They are built from organic material so close to human tissue that only a bone marrow analysis can prove what they are. Some rebel, sometimes violently, and slip back to Earth to hide. Police departments employ bounty hunters to find and "retire" them, the novel's chilling word for killing.

Rick Deckard is one of these hunters. He is assigned six escaped androids of a new, highly intelligent model called the Nexus-6. He wants the bounty money for a personal reason: real animals have become the ultimate status symbol, and he owns only an electric sheep, a convincing fake that now and then needs repairs. A second storyline follows John Isidore, a man whose mind has been harmed by the radiation, who tries to befriend the fugitives.

Alongside all this runs a religion called Mercerism. Its followers grip the handles of an "empathy box" and share, as a group, the suffering of a figure named Wilbur Mercer, who climbs a hill forever while stones are thrown at him.

How the empathy test works

Because androids look and talk like people, the police rely on polygraph-style tests, which read the body's reactions while the subject answers emotionally loaded questions. The theory is that humans feel instant empathy and androids do not, however intelligent they are.

Here is how the novel's test compares with the most famous real-world idea for judging machine minds, Alan Turing's imitation game:

Dick's choice is pointed. In his world, intelligence does not separate people from machines; the Nexus-6 can be as clever as anyone. What is supposed to separate them is empathy, the ability to feel what another being feels.

Where the test starts to crack

The novel keeps undermining its own test. Early on, Deckard visits the company that makes the androids and tests a young woman named Rachael Rosen. At first the test seems to have mistaken a human for an android, which would mean the police may have been killing people; only a second test shows that Rachael is an android after all. Later Deckard is himself accused of being an android with false memories. He meets another bounty hunter who retires androids coldly and wonders whether that man is human. Worst of all for Deckard, he discovers that he feels empathy for some of the androids he is paid to destroy.

The book never settles the matter neatly. The androids do something monstrous to a spider, which seems to confirm the theory. Yet the humans are often cold, too, and the shared religion that defines human empathy is exposed on television as a possible hoax. The Encyclopedia of Science Fiction sees the novel as the high point of Dick's many stories about artificial copies of people, which return again and again to one of his favorite themes: the gap between what is real and what only appears to be.

Why it matters for AI

The novel shifted the robot story. Earlier fiction often asked whether machines would rebel. Dick asked whether we could even tell, and whether our tests say more about us than about them. Critics have read the androids as a mirror: anxiety about machines that look human is really anxiety about what being human means.

In 2019 the question is practical. Chatbots and voice assistants are designed to sound natural, and researchers argue over how to measure what such systems understand. Dick's novel is a reminder that any test for "human" carries assumptions, and that a test can be wrong in both directions.

Could an empathy test really tell a person from a machine?

Probably not as the novel imagines it. The body's reactions differ from person to person and shift with mood and health, which is exactly the doubt Dick builds into the plot. A test that some humans fail cannot safely be used to decide who counts as a person.

A machine also does not need feelings to produce the outward signs of them. A system trained on enough examples of caring language can learn to say kind things, just as a chatbot can learn to make small talk. That gap between showing empathy and having it is what makes the novel feel current. The open question, then and now, is whether we will judge machines by what they do, by what they are made of, or by something we have not yet learned to measure.