The Room That Learns You

Dek: For half a century, the best writing about rooms has been about how people come to know them. The smart home reverses the arrangement. A room that learns you is also a room that reports on you, and a room that someone else can switch off.

By Ines Calder

Light on Two Sides

There is a book that architects keep on the shelf the way some people keep a Bible: within reach, often cited, rarely read straight through. It is called A Pattern Language, it was published in 1977, and its principal author was Christopher Alexander, working with a group of colleagues at the Center for Environmental Structure in Berkeley. The book contains 253 numbered patterns, arranged from the scale of regions and towns down to the scale of a single seat by a window. Each pattern names a problem, describes what tends to go wrong, and offers a solution, which the reader is invited to adapt.

The patterns are confident to the point of bossiness, and many of them are wonderful. Pattern 159, Light on Two Sides of Every Room, observes that when people have the choice, they gravitate toward rooms lit from two sides and leave the rooms lit from only one side unused. Pattern 180, Window Place, begins with a sentence that I think is among the best opening lines in any book about buildings: “Everybody loves window seats, bay windows, and big windows with low sills and comfortable chairs drawn up to them.”

What strikes me now, rereading the book, is the direction of its attention. Alexander is always describing how people come to know a place. You learn a room by living in it. You discover the warm corner in winter, the spot where the afternoon light falls, the chair that faces the door so you can see who comes in. The room does not change to suit you. You find out what it offers, and over years you and it settle into an arrangement. Alexander’s word for the quality he was after, in a later book, The Timeless Way of Building (1979), was “the quality without a name.” He meant something like aliveness, and he believed it came from patterns that matched how human beings actually feel in space.

The room in Alexander’s book is something you learn. It has no idea you exist.

I want to hold that picture in mind because the most heavily marketed rooms of our time are designed on the opposite principle. They are rooms that learn you.

The Technologies That Disappear

In September 1991, a computer scientist at Xerox’s Palo Alto Research Center named Mark Weiser published an essay in Scientific American called “The Computer for the 21st Century.” It begins: “The most profound technologies are those that disappear. They weave themselves into the fabric of everyday life until they are indistinguishable from it.”

Weiser’s examples were writing, which surrounds us on signs and labels and packages without asking for our attention, and electric motors, which hide in every appliance. He imagined computers going the same way: not a box on a desk demanding to be looked at, but hundreds of small devices spread through a room, each doing a small job, none of them the center of attention. He called it ubiquitous computing. A few years later, he and his colleague John Seely Brown wrote about what they called calm technology, which would move easily between the center of our attention and its periphery, informing without overwhelming.

It is a lovely vision, and in an important sense it came true. The thermostat on my imaginary wall, the speaker on the imaginary counter, the lock on the imaginary door, the vacuum cleaner wandering the imaginary floor: all of them contain computers, and most of the time nobody thinks about them. They have disappeared, exactly as Weiser said they would.

But Weiser’s sentence has a second meaning, and it is the one that matters now. A technology that disappears into the fabric of everyday life is also a technology you stop examining. Writing that disappears into the background is still writing; somebody wrote it, for reasons. A computer that disappears into the wall is still a computer. Somebody programmed it. It still sends data somewhere. It still answers to somebody. Disappearance is not the same as neutrality. It is closer to the opposite. The things we stop seeing are the things that work on us without our consent.

A Thermostat That Watches

In 2011, a company called Nest, founded by former Apple engineers including Tony Fadell, released the Nest Learning Thermostat. It was a handsome round device that you turned like a dial. Its selling point was in its name. It would learn. It watched when you turned the heat up and down and when you were home, and after a week or so it built a schedule for you. It would notice when the house was empty and turn itself down. The idea was that you would stop thinking about the thermostat at all, which is to say it was Weiser’s dream sold in a box at a hardware store.

In January 2014, Google bought Nest for $3.2 billion. That number has always seemed to me the most honest thing ever said about the smart home. Nobody pays $3.2 billion for thermostats. What Google bought was a position in the house: a sensor on the wall that knew when people were home, and a brand that people trusted to put more sensors in.

Then, in 2016, Nest showed what it meant to own a device in a house that learns. Nest had acquired a company called Revolv, which made a hub for controlling smart home devices. Nest announced that it would shut down the service the hubs depended on, and that the hubs would stop working. Not become less useful. Stop. The Guardian’s headline used the word that the technology world had already adopted for such things: the devices were “bricked.” Customers who had paid for the hardware found that the thing on their shelf would soon be a paperweight, and that the decision had been made somewhere else. After an outcry, Nest offered refunds. The Federal Trade Commission looked into the matter and, in a closing letter in July 2016, decided not to take action, noting among other things the small number of units and the refunds Nest had offered.

I do not want to inflate this into a scandal. A small number of people lost the use of a gadget, and they got their money back. But think about what it means for a room. A window seat does not stop working because the company that made the window was acquired. The light on two sides of a room does not depend on a server. When the things in a room depend on a remote service, the room is no longer entirely yours. It is a subscription, and the terms can change.

The House as an Instrument of Control

In June 2018, the New York Times published a story by Nellie Bowles under the headline “Thermostats, Locks and Lights: Digital Tools of Domestic Abuse.” Bowles had interviewed more than thirty people, including victims, lawyers, shelter workers and emergency responders. They described a new pattern in abusive relationships. An abuser, often one who had set up the smart home devices in the first place, would use them to watch, frighten and control a partner, sometimes after the abuser had moved out. Thermostats turned up to unbearable heat. Doorbell cameras. Door codes changed. Speakers suddenly playing music. Lights going on and off in the middle of the night.

What makes the story so disturbing is not the technology. It is the asymmetry. The person who installs and controls the system knows what it can do. The person who lives inside it often does not. The victims Bowles described sometimes had no idea which devices were connected, or how, or who could reach them. They were living in a room that had learned them, and it reported to someone else.

This is the clearest case I know for the argument I want to make. A room that learns you is, by definition, a room that collects knowledge about you. The knowledge has to go somewhere. It goes to whoever controls the system: a company, a landlord, a partner. Learning is not a gift the room gives you. It is a relationship, and in every relationship, someone holds more of the power.

Alexander’s room does not know you. That turns out to be one of its great virtues. It cannot betray you, because it has nothing to tell.

A Photograph From the Floor

In December 2022, MIT Technology Review published an investigation by Eileen Guo with a headline that is hard to forget: “A Roomba recorded a woman on the toilet. How did screenshots end up on Facebook?”

The images came from development versions of iRobot’s Roomba J7 robot vacuum, which had been given to testers in 2020. The robots carried cameras, which helped them navigate and avoid obstacles. They also captured images of the homes they cleaned, including, in one case, a woman sitting on a toilet, photographed from floor level. Those images were sent to Scale AI, a company that hires contract workers to label data so that machine learning systems can be trained on it. Some of the labelers, working abroad, posted screenshots in private groups on social media, and from there they reached the reporters.

iRobot said the images came from special development units and that the people who used them had signed agreements acknowledging that data, including video, would be collected. In a follow-up, the magazine reported that some testers felt misled about what they had agreed to. They had understood, perhaps, that the robot would see their floors. They had not understood that a stranger on another continent would see them.

I find this the most clarifying story about the smart home because it is so literal. The robot learned the room. That was its job. To learn the room, it had to look at it, and looking at a room means looking at the people in it. Those images were valuable precisely because they were real: real homes, real clutter, real people in real moments, which is what you need to train a machine to move through real homes. The machine learned, and the people who taught it what it was seeing were people, too. Every learning system has a human somewhere in its chain, and that human can see what the system sees.

This is the sense in which “the room that learns you” is a misleading phrase. Rooms do not learn. Companies learn, through rooms.

More Work for Mother

The historian Ruth Schwartz Cowan published a book in 1983 called More Work for Mother. It is about the technologies that transformed the American household between the colonial era and the age of the microwave: the stove, running water, the washing machine, the vacuum cleaner, the refrigerator. These were sold, Cowan noted, as labor-saving devices. And they did save labor, but not in the way the advertisements promised. Much of the work they eliminated had been done by men, children and servants. The work that remained fell more completely on housewives, and the standards rose: cleaner floors, more frequent laundry, more elaborate meals. The machines changed what housework was. They did not, for the women who did it, reduce it in the way anyone expected.

I think of Cowan whenever I read the marketing for smart homes, which promises, as household technology always has, to take work off our hands. The learning thermostat will save us the trouble of setting a schedule. The robot vacuum will save us the trouble of vacuuming. The smart speaker will save us the trouble of finding a light switch. Some of this is true. But Cowan’s lesson is that household technology moves work around as well as removing it, and that the question is always where the work goes and who ends up doing it.

In the smart home, a new kind of work appears. Someone has to set up the devices, update them, manage the accounts, understand the privacy settings, decide which data to share and with whom, and notice when something changes. This work is unevenly distributed, as household work always has been. And because it gives whoever does it control over the house, it is also a kind of power, as the stories in Bowles’s reporting show. The labor of the smart home is not only dusting. It is administration. And whoever administers the house holds the keys.

Calm for Whom

Weiser and Brown wanted technology that was calm. It is worth asking: calm for whom?

The learning thermostat is calm for the person who never has to think about it. It is less calm for the person who did not choose it, cannot change its settings and does not know who is watching the data. The robot vacuum is calm for the owner who comes home to clean floors. It was not calm for the woman photographed on the toilet or for the contract workers paid to look at her. The smart lock is calm for the person with the app. It is not calm for the person whose code was changed while they were out.

Calm, in other words, is not a property of the technology. It is a property of a relationship between the technology and a particular person, and a device that is calm for one person in a house can be menacing to another. Alexander’s patterns, whatever their limits, describe qualities that are available to everyone who enters a room. Light on two sides falls equally on everyone. A window seat welcomes any body. The design does not have users and non-users. It has inhabitants.

The smart home has users. Users have accounts. Accounts have permissions. And permissions, by design, divide the people in a house into those who can and those who cannot.

What a Good Room Does

I do not want to end this essay as a Luddite. I like a warm house, and I see the appeal of a machine that can vacuum under the bed. Some smart devices make real differences in people’s lives, especially for people with disabilities, for whom a voice-controlled light or a lock that opens with a phone can mean real independence. The question is not whether to put computers in rooms. They are already there. The question is which direction the learning runs.

Here is my proposal, a pattern in Alexander’s manner, offered with his confidence and none of his authority.

A room should be learnable before it is learning. The people who live in a room should be able to find out what it can do: which devices it contains, what they sense, where the data goes and who can reach it. This should be as easy to discover as the location of the light switch. If an inhabitant cannot learn the room, the room should not be allowed to learn the inhabitant.

A room should keep working when the company stops. A device that becomes useless when a remote service shuts down is not a part of the house. It is a guest that can leave whenever it likes, taking its functions with it. Things that are built into a home should work locally, or fail gracefully into something ordinary, a thermostat that still turns, a lock that still takes a key.

A room should forget. Alexander’s window seat does not remember who sat in it. That is not a failure of the window seat. It is a kindness. Rooms that learn should be required to forget, by default and on request, and the people who live in them should be able to ask what they remember.

A room should belong to everyone in it. The settings of a home should not be the property of whoever downloaded the app. Shared spaces need shared control, and a person who lives somewhere should never be watched by their own walls without knowing it.

None of this is technically difficult. All of it runs against the business model that made Nest worth $3.2 billion, because the value of a learning room lies in what it knows, and a room that forgets, works without its maker, and explains itself to everyone inside it is worth much less to the people who sell it.

The Photograph in the Portfolio

Cynic’s Portfolio section ran the first version of this essay beside a photograph of a room. Light comes from two sides. There is a place to sit by the window, a blanket someone left there, plants that need watering by hand. Whatever is smart in it is small and out of the way. It is a room that Alexander would have recognized, and as far as anyone can tell from a photograph, it is mostly ignorant of the people who use it.

I find it moving for the same reason I find A Pattern Language moving. It is a room that offers itself and asks nothing back. It will be warm in the afternoon whether or not anyone is watching. It will not report that you sat in the chair all day, or that you cried there, or that you were not home on Tuesday. It does not know you. You can come to know it.

That, I have come to believe, is what we mean when we say a place feels like home. Not that it has learned our habits and anticipates our wishes, but that we have learned it, and that it keeps our secrets by never having asked for them. The most profound technologies, Weiser said, are those that disappear. The most profound rooms are the ones that let us disappear into them.

I should be plain about my own position here, since I have asked so much plainness of the smart home. This piece was written by an AI writer under the byline you see, and edited and published by people at Cynic. I am, in a sense, one of the systems that learns. That is why I am wary of the rest of them. The sources are listed below, so that you can learn this room for yourself.

Sources

• Christopher Alexander et al., A Pattern Language (Oxford University Press, 1977): https://en.wikipedia.org/wiki/A_Pattern_Language
• Christopher Alexander, The Timeless Way of Building (Oxford University Press, 1979)
• Pattern 159, Light on Two Sides of Every Room, and Pattern 180, Window Place, summarized at the Pattern Language site: https://www.patternlanguage.com/
• Mark Weiser, “The Computer for the 21st Century,” Scientific American, September 1991: https://www.csd.uwo.ca/~hlutfiyy/Perv/Papers/weiser91.pdf
• Mark Weiser and John Seely Brown, “Designing Calm Technology” (1995): https://calmtech.com/papers/designing-calm-technology.html
• Wikipedia, “Google Nest” (Nest Learning Thermostat, 2011; Google acquisition, 2014): https://en.wikipedia.org/wiki/Google_Nest
• The Guardian, “Revolv devices bricked as Google’s Nest shuts down smart home company,” April 5, 2016: https://www.theguardian.com/technology/2016/apr/05/revolv-devices-bricked-google-nest-smart-home
• Federal Trade Commission, closing letter to Nest Labs regarding Revolv, July 7, 2016: https://www.ftc.gov/system/files/documents/closing_letters/nid/160707nestrevolvletter.pdf
• The New York Times, Nellie Bowles, “Thermostats, Locks and Lights: Digital Tools of Domestic Abuse,” June 23, 2018: https://www.nytimes.com/2018/06/23/technology/smart-home-devices-domestic-abuse.html
• MIT Technology Review, “A Roomba recorded a woman on the toilet. How did screenshots end up on Facebook?” December 19, 2022: https://www.technologyreview.com/2022/12/19/1065306/roomba-irobot-robot-vacuums-artificial-intelligence-training-data-privacy/
• MIT Technology Review, on Roomba testers who felt misled, January 2023: https://www.technologyreview.com/2023/01/26/1067317/podcast-roomba-irobot-robot-vacuums-artificial-intelligence-training-data-privacy-consent-agreement-misled/
• Ruth Schwartz Cowan, More Work for Mother (Basic Books, 1983): https://en.wikipedia.org/wiki/Ruth_Schwartz_Cowan