# Cognitive Biases
A wheel spins in a psychology lab. University of Oregon, 1974.
It stops at ten.
The researcher asks: "What percentage of African countries are in the United Nations?"
The participant guesses twenty-five percent.
Another participant. Same question. But their wheel stopped at sixty-five.
They guess forty-five percent.
Here's the thing: the wheel is rigged. Both participants *watched* the researchers rig it. They were told explicitly — the number means nothing.
And yet their answers move toward that anchor... like iron to a magnet.
Amos Tversky and Daniel Kahneman are watching this happen. Over and over. They're documenting something that shouldn't exist if humans are rational.
A meaningless number... hijacking judgment... in real time.
Welcome to cognitive biases.
The systematic ways your brain gets things wrong. Not *randomly* wrong — *predictably* wrong. In patterns you can map, replicate, bet money on.
Here's what makes this unsettling.
You can *know* about these biases. Study them for years. Teach them to undergrads. And still fall for them.
Kahneman himself said he's terrible at avoiding his own research findings. His co-author, Tversky, was better at it — which annoyed Kahneman to no end.
It's like knowing about optical illusions... but still seeing the lines as different lengths.
The knowledge doesn't fix the wiring.
So let's get something straight.
When psychologists talk about biases, they're not calling you stupid. They're saying your brain is running software designed for a different era.
You've got pattern-recognition systems built for spotting predators in tall grass and deciding which berries won't kill you... now trying to evaluate mortgage-backed securities and meta-analyses.
The mismatch creates glitches.
Predictable, mappable glitches.
The story starts in 1954.
Herbert Simon, an economist, kept watching actual humans make actual decisions. He's looking at how people work in organizations, and he notices something.
Nobody's doing the math.
Nobody's weighing every option against every other option like some probability-calculating angel.
They're *satisficing* — his word, a blend of "satisfy" and "suffice" — picking the first thing that's good enough.
Simon calls this *bounded rationality*. We're not irrational. We're rational within limits. Limited time, limited information, limited working memory that can hold maybe seven things... if you're having a good day.
That's the foundation.
But Simon didn't map the specific ways those limits play out. He didn't document the patterns.
That's what Tversky and Kahneman did.
Starting in the early seventies, they ran experiment after experiment showing that mental shortcuts — heuristics — create systematic errors.
Not random noise. Patterns you could write equations for.
Their 1974 paper, "Judgment under Uncertainty," identified three big ones.
**Availability.** You judge how common something is by how easily examples come to mind. Shark attacks feel more likely than vending machine deaths... even though vending machines kill more people every year.
**Representativeness.** You judge probability by how much something resembles your mental stereotype. Linda is thirty-one, single, outspoken, majored in philosophy. Is she more likely to be a bank teller... or a bank teller who's active in the feminist movement?
Most people say the second one.
That's impossible. The subset can't be larger than the set. But it *feels* right because it matches the story.
**And anchoring.** Like that spinning wheel. The first number you hear becomes the pole around which your estimate orbits.
Here's the thing that made their work explosive.
They weren't just documenting errors. They were showing that these errors violated basic principles of probability and logic that educated people *claim* to understand.
Ask someone if a random number should affect their estimate of African countries in the UN. They'll say of course not, that would be absurd.
Then watch them do exactly that.
The knowing doesn't stop the doing.
The economics world lost its mind over this.
Because economics in the seventies was built on a beautiful, elegant lie: people are rational actors who maximize utility. *Homo economicus.* It made the math work. It made theories predictive. You could model markets like you model physics.
Kahneman and Tversky were saying... yeah, about that.
Humans aren't calculating machines. They're pattern-matchers running on outdated firmware.
Richard Thaler picked up this thread and ran with it into the real world.
He started documenting all the ways people make economic decisions that don't fit the models.
The *endowment effect.* People value things more once they own them. Give someone a coffee mug, then try to buy it back. They'll charge you more than they would've paid for it five minutes ago.
Nothing about the mug changed. The ownership changed the value.
In the eighties and nineties, behavioral economics became a field. Thaler won the Nobel in 2017. Kahneman won it in 2002, though Tversky had died in '96 and couldn't share it.
Which feels like its own kind of tragedy.
The Nobel committee doesn't give prizes for calling people irrational. They give them for being *right* about how decisions actually work.
One of the biggest findings: *loss aversion.*
It came out of Prospect Theory, which Kahneman and Tversky developed in 1979.
Here's the core insight.
Losing a hundred dollars hurts about *twice* as much as gaining a hundred dollars feels good.
Not a little more. Twice.
Your brain weighs losses heavier than equivalent gains. The pain of losing fifty bucks is roughly equal to the pleasure of winning a hundred.
This explains so much.
Why people hold onto losing stocks too long — selling would make the loss real.
Why they won't negotiate for fear of losing what they have.
Why a pay cut feels devastating even if you're still making good money.
The reference point — what you had, what you *expected* — matters more than the absolute outcome. You're not evaluating your salary against zero. You're evaluating it against last year.
And here's where it gets strange.
You can see this in brain imaging.
Tali Sharot did a study in 2007 where people imagined negative future events while in an MRI scanner. Parts of the brain that process emotional threat showed *reduced* activity when people thought about bad things happening to them.
Not heightened activity. *Reduced.*
People were literally less able to process bad possibilities for themselves than for strangers.
Optimism bias isn't just attitude. It's neurology. Your brain is actively dampening the signal.
Eighty percent of drivers think they're above average.
That's mathematically impossible.
Eighty-eight percent of Americans think they have above-average intelligence.
Ninety-four percent of college professors think they're better at their job than their colleagues.
Ask yourself right now: are you an above-average driver?
Feel that little "well, yeah" in your head?
That's the better-than-average effect.
You're doing it while learning about it.
I'm doing it while writing about it.
Now. A doctor is in a conference room in 1981.
Tversky and Kahneman have given her a scenario.
There's a disease outbreak. Six hundred people are at risk. Two treatments are on the table.
**Option A** saves two hundred people for certain.
**Option B** has a one-third chance of saving everyone and a two-thirds chance of saving no one.
Most doctors pick A. Certainty feels good.
Then the same choice, reframed.
**Option C** means four hundred people will die for certain.
**Option D** has a one-third chance nobody dies and a two-thirds chance everyone dies.
Now most doctors pick D. They're suddenly risk-seeking, trying to avoid the certain loss.
Same outcomes. Different frames. Different choices.
Seventy-two percent preferred A in the gain frame. Seventy-eight percent preferred D in the loss frame.
This is the *framing effect.*
The way information is presented changes decisions... even when the underlying facts are identical.
Lives saved versus lives lost.
Ninety percent fat-free versus ten percent fat.
A ninety-five percent survival rate versus a five percent mortality rate.
Same thing. Different feeling. Different choice.
And it's not just naive subjects. Doctors, judges, intelligence analysts — everyone's vulnerable.
You might be thinking... okay, but I can learn to avoid this. Just be more careful. More analytical. Slow down.
Here's the problem.
These biases don't live in your slow, careful thinking. They live in what Kahneman calls *System One* — the fast, automatic, always-on system that's generating impressions and feelings every second.
*System Two* — slow, effortful, analytical — can sometimes catch them.
But System Two is lazy. It doesn't want to work unless it has to. It's metabolically expensive to think hard.
And System One is already giving you an answer that feels right... that feels like it came from careful analysis... even though it came from pattern-matching.
Overconfidence shows up everywhere.
People are wrong about twenty to thirty percent of things they claim to be "ninety-nine percent certain" about.
Doctors are wrong about diagnoses they're highly confident in.
Eyewitnesses are wrong about events they swear they remember clearly.
The correlation between confidence and accuracy is shockingly weak.
You're walking around most of the time thinking you're more certain than you should be.
But here's the turn.
There's a researcher named Gerd Gigerenzer at the Max Planck Institute who's been pushing back on this whole framework since the nineties.
He says we're too focused on the errors and not enough on the efficiency.
Heuristics aren't bugs. They're features.
In the real world, with limited time and information, these shortcuts often work *better* than complex analysis.
A doctor using pattern recognition to diagnose a patient isn't being biased — they're being experienced.
Recognition-primed decision making, it's called. Firefighters do it. Chess masters do it. They're not calculating. They're recognizing patterns they've seen before.
Gigerenzer has a point.
Take the *recognition heuristic.* If you recognize one option and not the other, bet on the one you recognize.
Sounds dumb.
But ask Americans which city is larger — San Diego or San Antonio?
Most get it wrong. San Antonio's bigger.
Ask Germans the same question? They get it right more often.
Because they've heard of San Diego and not San Antonio. The recognition itself carries information about size.
The heuristic works.
The question isn't whether heuristics create errors.
It's whether they create more errors than they prevent... and whether the errors matter in context.
A bat and a ball cost a dollar-ten in total. The bat costs a dollar more than the ball. How much does the ball cost?
Your System One screams "ten cents!"
That's wrong. The ball costs five cents.
But if you're trying to decide whether to buy them... does it matter?
If you're designing a pricing strategy... it might.
This debate is still live.
Are biases always irrational? Or are they rational adaptations to an uncertain world that happen to misfire in modern contexts?
Kahneman and Gigerenzer have been arguing about this for decades.
Nobody's won yet.
Think about the COVID pandemic.
Optimism bias made people underestimate their personal risk — sure, it's bad, but it won't happen to *me.*
Availability heuristic made every news story about a young, healthy person dying feel like a common outcome... even though statistically it wasn't.
Confirmation bias had people seeking information that supported what they already believed about masks, vaccines, lockdowns... and ignoring everything else.
These biases shaped policy. Shaped behavior. Shaped who lived and who didn't.
But they also kept people from spiraling into paralysis.
Some optimism might be necessary to get out of bed. Some availability bias might be necessary to take threats seriously.
The question is always: which errors can you afford... and which ones kill you?
The current frontier is weird.
We're building AI systems. And guess what they're learning?
Our biases.
Algorithms trained on human decisions inherit human patterns... including the broken ones.
Amazon built a hiring algorithm that discriminated against women because it learned from historical hiring data.
Predictive policing algorithms send more cops to neighborhoods that were already over-policed.
The bias in the training data becomes bias in the model.
There's research now on "debiasing" algorithms. Using machine learning to detect and correct bias in real time.
But that assumes we know which patterns are bugs and which are features.
And we don't always know.
Here's the thing that keeps me up at night.
We're teaching machines to think like us... biases and all.
And then we're trusting their outputs because they feel objective.
"The algorithm said so" carries weight that "my gut said so" doesn't.
But the algorithm's gut came from our gut.
Look. Here's what you can actually *do* with this.
Not eliminate bias — you can't. But you can build tripwires. Decision checkpoints.
When you're making a decision that matters, ask: **What's my reference point?** Am I overweighting a loss? If you're refusing to sell a stock, is it because you believe in the company... or because selling would make the loss real?
When someone presents information, ask: **How would this sound framed differently?** Ninety percent survival rate versus ten percent mortality rate.
When you feel certain, ask: **What would I need to see to change my mind?** If the answer is nothing... you're not thinking. You're defending.
Kahneman put it this way: "Nothing in life is as important as you think it is... while you are thinking about it."
Whatever you're focused on right now is taking up more mental space than it deserves.
That fight with your partner. That work deadline. That thing you said five years ago that you still cringe about.
It's looming larger than it should... because it's in the spotlight.
Step back. Zoom out. Check the frame.
One more thing.
The next time you're in an argument and you're *sure* you're right... remember: the other person has the same feeling.
You both have access to the same cognitive machinery. You're both running the same buggy software.
That doesn't mean you're both equally right.
But it means certainty is cheap.
Checking your priors is expensive.
Do the expensive thing.
You're not going to stop being biased. Your brain's not getting a software update.
But you can stop being surprised by it.
You can catch yourself mid-pattern and ask... wait. Is this real? Or is this just what *feels* real?
That gap between the two?
That's where thinking actually happens.