Baxnet Ideas · Founder note
Everybody Lies Looked at Society. I Kept Thinking About the Individual.
TL;DR: When I first read Everybody Lies, the question that stayed with me was smaller than its datasets—what might one person's history help them notice?
When I first read Everybody Lies, I did what the book keeps tempting the reader to do: I wondered what my own data would say about me.
This was during the early thinking behind Baxnet Technologies and what I would later call Personal Intelligence. Seth Stephens-Davidowitz was looking at enormous populations through the traces they left online. I kept shrinking the question: what if the subject was one person, and the purpose was to help them see something useful about themselves without studying, targeting or classifying them for somebody else’s benefit?
That question has stayed with me.
What people say is only one kind of evidence
The argument in Everybody Lies begins with a simple problem for social research: people are unreliable narrators.
We give answers that are flattering, socially acceptable or close to the person we would like to be. Sometimes we deliberately conceal things. Often we are not consciously lying at all. We misremember. We simplify. We answer according to our intentions rather than our behaviour.
The internet created another source of evidence. People might tell a survey one thing and type something rather less presentable into a search box when nobody else appears to be watching. Search data could reveal anxieties, prejudices, desires and relationship problems that were difficult to see through conventional questioning.
The candid confessions were the hook. The wider point was that new digital traces allowed researchers to ask new questions. The book moves through reimagining what counts as data, finding more candid forms of it, zooming into particular groups, and using large-scale experiments to test explanations. It also spends time on what big data cannot tell us and what should not be done with it, which matters just as much. The publisher’s teaching guide follows that full arc.
At population scale, this can overturn conventional wisdom. A confident story about what society believes meets a large body of behaviour that does not quite agree.
At personal scale, the disagreement becomes quieter.
The question kept getting smaller
Most of us carry a working story about who we are.
I am the one who checks in. I am calm during disagreements. I make time for the people I care about. I have always been close to this person. I am usually the organiser. I do not worry much about money. I am at my best in the mornings.
These stories are useful. We could not reconsider our entire identity before making breakfast. But they are assembled from memory, and memory has a habit of preserving the headline while dropping the inconvenient dates.
A long personal record might offer another view.
Perhaps I remember myself as the friend who always initiates, while the conversation history shows that this was true five years ago and has slowly stopped being true. Perhaps I think a relationship changed after one argument, while the rhythm had been shifting for months. Perhaps my calendar says that the priorities I describe as important are repeatedly the first things I move.
None of this requires a dramatic revelation. The useful observation may be that I become more formal with one person when I am worried, ask for advice from someone I think of as a peer, or return to the same search whenever work feels uncertain.
The scale is different from the book. One person’s history cannot support the same claims as millions of anonymous searches. It can, however, let a person compare their remembered story with a record accumulated over time.
That is a different kind of intelligence. It is less concerned with predicting what people like me will do and more concerned with helping me inspect what I have already done.
A discrepancy is not a confession
Stephens-Davidowitz famously described Google searches as a form of “digital truth serum.” At the level of one person, I would handle that phrase carefully.
A search is not necessarily a desire. It might be fear, research, curiosity, a present for somebody else or an attempt to settle a strange argument over dinner. A late reply does not reveal how much someone cares. Message volume is not affection. Silence in one app may conceal a weekly phone call somewhere else.
Even Google cautions that its own Trends data is sampled, aggregated and not a scientific poll. It describes search interest as one data point among others, not a perfect mirror and certainly not evidence of an individual’s activity. That is useful discipline even before moving to the far messier scale of a single life. Google explains those limits in its Trends methodology.
There was an early glimpse of the personal version of this idea when journalist Joel Stein gave Stephens-Davidowitz access to his search history. The resulting profile was funny in places and uncomfortable in others. More than anything, Stein’s account in Time shows how quickly personal analysis can feel exposing when another person is the one delivering the conclusions.
My interest is in reversing that arrangement.
The person should get the first look. They should see what produced an observation, which period it covers, what is missing and how confident the system is allowed to be. They should be able to correct it, dismiss it and delete it.
A useful system would not announce, “You neglect your friendships.” It might show that contact with three close friends has become less frequent over the past year, then let the person decide whether that is meaningful. It could ask a good question without pretending to know the answer.
The data should answer back to the person
We already live with systems that form opinions from personal data.
They decide what appears in a feed, which advertisement follows us, whether an alert is worth sending and which piece of content is likely to keep us around. The analysis is personal, but the understanding mostly belongs to the organisation running it.
The idea of a personal intelligence engine came partly from wondering what would happen if some of that analytical power faced the other way.
Could a person examine how their relationships change across years, rather than seeing only the most recent messages? Could they notice the different roles they play in different conversations? Could they explore which concerns keep returning, where their attention goes, or how a major life event altered the way they communicate?
These are not questions a generic model should answer from a distant server with an unlimited appetite for context. The architecture has to match the intimacy of the material: data chosen by the person, processed within a clear boundary, with the evidence kept inspectable and under their control.
Conversation data also includes other people. That makes the boundary more complicated than a private step count. Personal reflection should not become a licence to diagnose friends, expose their secrets or turn a relationship into a covert scoring exercise.
The safest output is usually a pattern, its scope and a question.
A private mirror with edges
The embarrassing searches were memorable, although we already knew that human beings were complicated.
The lasting influence of Everybody Lies was its demonstration that the stories people tell and the traces they leave can produce different pictures—and that the gap between them may contain something worth understanding.
At the scale of a country, that gap can change social research.
At the scale of one life, it might help someone notice a friendship drifting, an old role persisting or a stated priority repeatedly losing out to the ordinary shape of the week. The data will never contain the whole person. That is precisely why the person still has to do the interpreting.
The useful question is simpler: “I thought this was true about me. Is it still?”