The People Who Know You Best Were Right About You | Ep 7

Thank you. A description of your character given by somebody who isn't you has a name in the research, and the name is the least interesting thing about it. It's an observer report, meaning an account of what you like from a person who isn't you. So, start with what it's not. It's not gossip, because it's collected on purpose and can be checked against occasions.

It's not a reference which is written by somebody you chose. It's not a diagnosis, and putting a clinical label on a colleague is work for a qualified clinician, with access to the person over time. And it's not the questionnaire which asks you about yourself. Now to the finding that starts this off.

A team in Spain searched four databases for every study that measured narcissism, manipulativeness, or the callous impulsive cluster, both by asking the person and by asking somebody who knew them. They found 24 studies covering 7,022 people, and they pooled them. All three dark traits came out at a level the field views as substantial.

And here's the part that surprises people. The level was higher than the same comparison usually runs for ordinary personality traits, the ones nobody has any reason to hide. The author's explanation tells you what to look for. A behavioural characteristic is easier to judge when it's strongly desirable or strongly undesirable, because both get noticed and both get talked about.

Nobody remembers who was reliable in March. Everybody remembers who took the credit in front of a full room. The The three traits didn't come out as being equal, and the order isn't the one most people would guess. The callous impulsive one was read best, the self-regarding trait came next, and the strategic manipulative trait produced the weakest agreement of the three, and it's the one most people assume is easiest to spot.

The popular treatment ranks these three by how frightening they sound, which puts the callous one first and the strategic one behind it as the lesser problem. Evidence on how much of a person reaches other people ranks them the other way, putting the strategic one at the far end where the light doesn't reach.

The author's explanation changes what to look for. Somebody high on the self-regarding trait has reputation to maintain, and behaves much the same in every room, which makes the behaviour legible. Somebody high on the callous trait is not too concerned with how they come across, so it's not managed either.

Somebody high on the strategic trait is doing something else, because the trait is characterised by strategic self-interest, so the behaviour is shaped for whoever is watching, and every witness gets an accurate account of a different presentation. There's one more asymmetry. For the strategic and callous traits, raters who knew the person better agreed with that person's own account more closely.

For the self-regarding one, how well the rater knew the person made no reliable difference. So on that one trait, distance costs less than it does elsewhere. music Agreement isn't accuracy, and I want to be straight about what that means, because misinterpretation tends to travel. What those studies establish is that two viewpoints converge, and the thing each observer is compared against is a person's own questionnaire score, which is a very instrument everybody agreed was suspect to start with.

So if a person and the people around them are wrong in the same direction, the method regards agreement and reports a success. That's not a reason to throw the finding away, it's about what that description entitles anybody to do, rather than about who's right. The pooling carries limits its authors call out.

24 Studies is still a small pool. None of the designs followed anybody over time, and only one covered the fourth trait, everyday sadism. Until the 24 use people at work, which is the next section, because the science built almost entirely on parents, partners, friends, flatmates and classmates, was then borrowed by employers, and applied inside the one relationship it identifies as the weakest.

So let's If you had to name the person best placed to describe a colleague, you'd probably name whoever spends the most time with them. That answer contains an assumption and it's been tested. Three pill analyses covering 263 separate samples and more than 44,000 people separated two things that usually go together, which is how often you interact with somebody and how close you are to them.

Interacting more often improves accuracy by a modest amount. Closeness is what produces a substantial gains. A recent review puts it in a single sentence. Agreement runs high where the relationship is deep and long acquaintance limited to a single setting, such as the workplace, may not produce an accurate judgment of some traits at all.

I want to be careful because recent evidence softens that conclusion. Across more than 5,400 people rated by parents, siblings, friends and partners, agreement was strong for every one of those relationships and highest for partners. So closeness is a gradient, not a threshold. What it will not support is the assumption you started with, that time in the room with the person is what counts.

They're The obvious objection is that somebody outside work cannot say anything useful about behaviour inside work. That objection has been tested directly, and it doesn't hold up under scrutiny. A German study asked 111 employees to rate themselves, then collected ratings from 106 personal acquaintances, including family members, and from 102 work colleagues.

Complete data sets existed for 97 people. Colleagues predicted a job performance best overall, which is unsurprising since they were also rating the performance. What is surprising is that their personal acquaintances beat the employees' own accounts of themselves, and on conscientiousness they beat the colleagues too.

The same study found that people who overestimated their own agreeableness and conscientiousness performed worse than people who did not. Two later studies set acquaintance ratings against workplace misconduct recorded by the person's own supervisor. The acquaintance ratings predicted that misconduct, over and above what the person said about themselves.

What your friend lacks is knowledge of your job. What they have is an unobstructed view of you, which is the half your employer isn't collecting. What they have at home, what everything so far argues for taking other people's accounts more seriously. Here's the counterbalance to that. There's a variable almost nobody controls for, the observer.

In a field study of over 300 people, one person's rating of a colleague predicted that colleague's supervisor-rated performance more strongly when the person doing the rating scored high on conscientiousness, openness and emotional stability. So two people can watch the same colleague for three years and produce measurably different accounts and nobody in the room is paying attention to the observer.

Across two samples carrying over 1,600 informants, the share of what any single rater said that was unique to that rater shared neither with the person nor with anybody else describing them ran as high as 4 in 10 in the larger sample and lower in the other. There's a reason to distrust a reference in particular.

When the people giving an account are chosen by the person being described, those who like that person describe them more favourably. The researchers recommend choosing referees independently of the candidate's preference. The first Now the observer nobody agreed to. Across more than 86,000 volunteers who completed a 100-item personality questionnaire, a machine model working from Facebook likes matched their own answers more closely than judgments made by their own Facebook friends and predicted several life outcomes better also.

Two things are true at once here. The finding is real, but it's 11 years old now, so any discussion treating machine inference as speculative is a decade out of date. And the thing the model was matched against is a person's own questionnaire, so a model that agrees with your questionnaire has just agreed with your questionnaire.

Now the parts the coverage leaves out. Pooling 21 studies of a smartphone data, only one of the five broad traits reached a moderate association. The author's own conclusion is that use of this for accurate assessment of an individual is currently limited. And the work has reached this series' own subject.

A study this year compared seven machine learning methods predicting the three traits from people's status updates. Notice what it actually reports, which is prediction error rather than accuracy. It's a comparison of techniques, and it doesn't claim that any of them identifies anybody. Meanwhile, the coverage announced that artificial intelligence can now tell whether you're a psychopath from your posts.

One study contradicts this. Given short self-descriptions, a large language model inferred a range of psychological characteristics more accurately than human judges reading the same text. Those judges were strangers, so it's not a comparison with anybody who knows you. It's still the strongest counter-evidence in this episode.

This is a comparison of the physical fy So what do you actually do? Let's start with three things not to do, because that's where the damage gets done. Don't run a consequential judgement about a person through one source, not one manager and not one model, which is the same defect wearing different clothes.

A large share of any single account belongs to the person giving it, and nobody in the room is assessing the accuser. Don't put a label on a named person. However many accounts agree, several people agreeing about a colleague is not a clinical judgement and can't be converted into one, and if somebody's conduct suggests they're unwell, that's an occupational health issue for a qualified clinician.

What all this does support is a better source description of what somebody has done, and never a category for a human being. Don't build a backchannel. A network that collects accounts without asking what we're seeing produces reputations rather than evidence. It's most easily steered by whoever is best at steering it, and its subjects can't answer what they're never shown.

Thank you. Thank you. Now for four things that can help, with an honest caveat. None of the four has been trialled as advice by anybody. Where a move follows from a finding or from practice, I'll call it out. Firstly, ask what somebody saw rather than what they think. Not, he's not a team player, which is a verdict, and a verdict arrives already blended with the person delivering it.

Instead, in the March review, he committed my team to a date without telling me. The second can be dated, checked and answered, and the first can't be any of those things, which is why the moment a label goes on, the conversation stops being about conduct that could change. The second move is arithmetic.

Before deciding what several accounts are worth, work out how many of them are genuinely independent. Four colleagues who share a group chat is closer to one account than to four. One person who's known somebody for 15 years and never discussed them at work is a different thing entirely. One person The third move changes what you do on Monday.

Ask what the situation was paying for. If the behaviour you're judging was rewarded by the role, you're looking at an environment at least as much as at a person, and moving the person without changing the reward reproduces the problem. The fourth move is for anybody buying a score. Put the same four questions to an automated assessment that you put to a referee.

What was it trained to predict? Against whose ratings? How stable is it? And for whom does it work less well? Researchers built models scoring the five broad traits from behaviour in more than a thousand video interviews. The models trained on interviewers' ratings behaved reasonably. The models trained on candidates' own answers about themselves showed little evidence of working at all.

So what a system is trained undetermines whether it works, and a supplier who can't answer that hasn't sold you an instrument. The fourth question isn't a formality either. Language models reading people's posts inferred personality less accurately for men and for older users, and a 2026 study ran a language model over 406 LinkedIn profiles, and found weak inferences and uneven errors across groups.

A method can be accurate on average, and worse for you. O. Thank you. Thank you. Thank you. Thank That was the half for anybody deciding about somebody else. This half is for the person being described, which is all of us most of the time. Start by finding out what exists. In the UK, a subject access request reaches for the personal information collected through monitoring.

Employers that monitor their employees are expected to have a clear purpose, a lawful basis, the least intrusive means available, and an impact assessment of where monitoring is likely to be high risk. I wouldn't oversell it. It's a right of access, not a right to an explanation. It'll tell you that an assessment exists, which most people may not know.

Know where the lines already are. Since February 2025, it's been prohibited across the European Union to use artificial intelligence systems to infer a person's emotions at work or in education, other than for medical or safety reasons. And in February 2024, the UK regulator ordered a leisure centre operator to stop scanning faces and fingerprints of more than 2,000 staff to record attendance, because a card would have done the job.

The commissioner's own line is the one worth keeping. You can't reset a face or a fingerprint the way you reset a password. Now for something that most videos on this subject won't tell you. There's no tested way of doing any of this. Now one of the four suggestions has been trialled. The research is reasonably strong on how people become known and close to silent about what to do once you know.

The research Next week I'm covering what follows from all this. Being good at your job isn't enough and it isn't because of politics or luck. It's that competence has to be inferred by somebody from evidence they can actually see and most people are producing the wrong evidence. Look out for next week's episode.

One last thing about today and it may be one of the most important things in this episode. Everything here is a statement about groups and that one of these studies licenses a conclusion about the colleague you may have been thinking about while watching. Research can change what you expect and what you try first.

It can't tell you what's going on inside a person. Thank you. Thank you. Three last things before I sign off. The written version of this episode is on my website. Every reference, the verification status of each and the four things I still don't know. The link is in the description below and on screen now.

If you found this episode interesting, please like and subscribe to my channel to get notified about the rest of the series. This series works through the antagonistic core one form at a time and each episode assumes the one before it. And one thing from you, think of the person whose description of you would be the most accurate.

Has anybody ever asked them about you? I'm Dr Nick Kecker. Thank you for watching. I'll see you next time. Thank you. Thank you. Thank you. Thank