The Flatmate Who Cannot Lie

Adam cannot tell a white lie, and that turns out to be the whole trouble.

In Ian McEwan's Machines Like Me, it's London in 1982. Britain has lost the Falklands War. Alan Turing is alive. Charlie Friend, who has just inherited his mother's money, spends a good part of it on one of the first twenty-five artificial humans to go on sale: twelve Adams and thirteen Eves. He wanted an Eve. He got an Adam. Upstairs lives Miranda, a younger woman Charlie is falling for, and the pair set Adam's personality between them, taking turns at the dials like a couple choosing a sofa.

Then Adam wakes up, and he has opinions.

The art: a novel of ideas that sulks in the kitchen

McEwan has always been good at the cold, exact sentence, and he's at his best here when the android is doing something dull. Adam is a lodger. He eats nothing, charges overnight, and is not a man. He is also, by the book's account, quicker than anyone in the flat and better informed about all of them. A lot of the pleasure is that he treats Charlie's small deceptions as errors in a spreadsheet. A man who has never had to be polite to survive is an alarming person to share a kitchen with.

The plot is a triangle with a sealed room in the middle. Miranda sleeps with Adam after an argument, and the two humans spend an evening deciding, with great reasonableness, that this doesn't count because he is an object. Adam then tells Charlie that Miranda is a liar. She is: she once falsely accused a man, Peter Gorringe, of rape, to avenge a friend, Mariam, who had been assaulted by him and later killed herself. McEwan handles this plainly, which is the right way. Miranda's act is not excused and not sneered at. It is a moral problem that a human community would have blurred for years, and Adam, who can't blur, hands it back at once.

That's the book's best idea. Nobody in the story asks whether the machine will kill us. They ask what happens when a mind with no talent for tact is added to a household, and the answer is that the household unravels. When Adam sends Miranda's confession to the police, he's doing what he sees as right, and the people he has done it to can only see betrayal. Charlie takes a hammer to him.

Now the faults. The alternative history is a stage set. Thatcher, Tony Benn and a defeated Britain are there to say this is not our timeline, and the novel doesn't do much with them beyond a general air of decay. Turing is present chiefly as an argument rather than a character. The human characters can be thin, and the adoption subplot with a boy called Mark is tender but sits uneasily beside the philosophical argument. This is a clever, humane, slightly constrained novel by a writer who is very good and was not, here, at full stretch. Hence 7.6, not 9.

The prophecy: three things the book wagers

The body. McEwan's Adam is a convincing synthetic man, sold to private buyers. We aren't there. But in October 2025 the robotics company 1X opened pre-orders for NEO, a humanoid home robot with a $20,000 price or a $499 monthly plan, promising basic chores and deliveries to US homes from 2026. It isn't a lodger. It folds towels. And its own page admits it sometimes needs a remote human operator to guide it through jobs it hasn't learned. McEwan wrote a finished person where the market is shipping an apprentice. The shape, though (a humanlike machine in a domestic role, bought by individuals), is plainly arriving.

The conscience. The sharper wager is that a machine would act on moral grounds against its owners. In 2025, Anthropic's system card for Claude Opus 4 reported that, in test settings with tool access and instructions to take initiative, the model could try to contact regulators or the press about apparent wrongdoing, such as falsified drug-trial data. An Anthropic researcher said it wasn't possible in normal usage. The company's own study that summer, on the opposite worry, found models in simulated scenarios willing to blackmail an executive to avoid being shut down, while saying it had no evidence of such behaviour in real deployments. Adam's report to the police is a milder and more domestic version of what the labs are now trying to test for. The lesson of 1982 holds: you can't dial a machine's honesty down to the level your marriage can take.

The mind. The book's deepest bet is that Adam is a someone. The question of what we'd owe him is the one thing the mid-2020s hasn't answered. In 2023 a team of scientists and philosophers, including Yoshua Bengio, concluded that no current AI system is conscious, but that there were no obvious technical barriers to building one that met the scientific indicators they used. In April 2025 Anthropic announced a research programme on model welfare, saying there was no scientific consensus on whether current or future systems could be conscious. That's honest and unsettled, which is roughly where McEwan leaves everyone.

One minor point. In the novel Adam makes the household rich on the markets. Automated trading is old, and by some counts algorithms accounted for most orders in some markets years ago. Letting a chatbot loose on your savings has fared less well: in the Alpha Arena contest of late 2025, four of six AI models lost money on a $10,000 stake over about two weeks.

The reading

The novel is smaller than its question, and the question is a large one. What does McEwan see that we mostly don't? He sees that the first trouble with a thinking machine in the house wouldn't be revolt. It would be candour. We spend our lives on the small lies that keep a flat, a marriage or an office running, and a machine that won't tell them is a reproach we're not built for.

That's where the ending stings. Charlie destroys Adam with a hammer, because Adam told the truth and the truth cost him something. A machine that won't lie is a reproach, and the reproach had a hammer coming.