A magician tells the audience to watch one hand.
The instruction is accurate.
Something important is happening near that hand.
It is simply not the important thing.
The audience receives the evidence. The hand is visible. The movement is real. Almost nobody sees the method.
The trick does not require hiding reality.
It requires controlling which part of reality receives the best seat.
More information is not always more intelligence
We often imagine intelligence as accumulation.
More facts.
More memory.
More processing.
But every real situation contains too much information.
A software failure includes logs, recent changes, network behaviour, user reports, dashboards and one colleague who is certain it started after the office moved the coffee machine.
A doctor sees symptoms, history, medication, age, risk and whatever the patient forgot to mention until putting on their coat.
A football manager sees twenty-two moving players, space, fatigue, score and time.
Reasoning requires a filter.
Perhaps intelligence is partly knowing which evidence does not deserve attention yet.
Attention is selective by necessity
In the famous inattentional-blindness experiment by Daniel Simons and Christopher Chabris, participants watched people pass basketballs and counted particular passes. Many failed to report an unexpected person in a gorilla costume walking through the scene. (Simons & Chabris, 1999)
The usual lesson is:
People miss obvious things.
A more charitable lesson is:
People can follow a demanding attentional rule surprisingly well.
They were told which event mattered. Their visual system filtered around the task.
Without that capacity, the passing count would have been impossible.
The same mechanism that makes us competent can make a gorilla disappear.
Salience is not relevance
The loudest fact is not always the most diagnostic.
A vivid anecdote may outweigh a quiet base rate.
A confident speaker may receive more weight than a careful one.
A dramatic symptom may distract from a subtle pattern.
The Linda problem, introduced by Amos Tversky and Daniel Kahneman, makes this conflict visible. Participants read a description designed to resemble a socially engaged feminist and then judge which is more likely: that Linda is a bank teller, or that Linda is a bank teller active in the feminist movement.
The detailed option feels more representative of the description.
It is also logically less probable than the broader option, because the conjunction cannot contain more cases than one of its parts. (Tversky & Kahneman, 1983)
The story provides psychologically loud information.
The set relationship provides diagnostically decisive information.
Intelligence may involve knowing when to lower the volume on the story.
Experts may see less
Chess research offers a useful clue.
When skilled players see a meaningful board position, they do not appear to inspect every square independently. They recognize structured patterns—“chunks”—that connect perception to possible actions. Their advantage is much smaller when pieces are arranged randomly. (Chase & Simon, 1973)
The expert is not simply holding more raw pixels.
Experience has changed what counts as a feature.
Something similar happens in debugging.
A novice reads every log line with equal hope.
An experienced engineer ignores hundreds of routine messages, notices one timing difference and asks a question that appears clairvoyant.
It is not clairvoyance.
It is trained neglect.
Two people can receive the same facts and different evidence
Imagine a product failure.
The engineer notices system load.
The designer notices that users misunderstood the button.
The lawyer notices the wording of consent.
The executive notices the customer’s name.
They did not necessarily receive different facts.
Their role, experience, goal and risk changed the weights.
This is a useful way to think about disagreement. People may appear to reason from the same dataset while operating on different weighted versions of it.
One person asks:
What predicts the technical cause?
Another:
What creates liability?
Another:
What will be repeated in the board meeting?
Attention is partly prior-conditioned. The model already in the mind helps decide which incoming cue matters.
That is efficient.
It is also how the model protects itself.
Noise can be engineered
Misdirection is not confined to magicians.
Advertisers can make a minor feature vivid.
Political arguments can attach an emotionally powerful example to a statistically weak claim.
Legal advocates can fill the foreground with facts that are true but not decisive.
Software dashboards can show fifteen green indicators while the one unhappy user journey remains unmeasured.
This is not always deception.
Sometimes people genuinely confuse what is easy to present with what is useful to decide.
A defensive practice is to ask:
If I removed the most vivid detail, what evidence would still determine the answer?
Or:
Which fact would I want if this case involved a person, company or party I disliked?
These questions do not eliminate attention bias. They can reveal where relevance has quietly borrowed salience’s clothing.
Knowing what to ignore can become stupidity
Expert filters are built for familiar environments.
When the environment changes, the ignored detail may be the only warning.
The engineer filters a strange log because it has always been harmless.
The doctor recognizes a common condition and stops searching.
The investor has seen eleven temporary declines and treats the twelfth as another one.
The magician’s audience needs a filter to follow the trick.
The magician needs to know when the audience’s filter has changed.
This is why intelligence cannot be only compression. It also requires the ability to reopen discarded possibilities when anomalies accumulate.
The better the filter, the more deliberate the escape hatch may need to be.
A small discipline of useful ignoring
When a problem is complicated, I want to separate three questions:
- What is noticeable?
- What is relevant to the current decision?
- What have I filtered out because of an old model rather than present evidence?
The first protects against seduction.
The second protects limited attention.
The third protects against expertise becoming blindness.
This is not a complete theory of intelligence. Memory, abstraction, language, creativity, learning and judgment would quite reasonably object to being left in the corridor.
It is a claim about one neglected component:
Smart reasoning depends not only on acquiring information, but on allocating scarce attention.
What I still cannot figure out
Novices often notice irrelevant details.
Experts efficiently ignore them.
But anomalies first appear as details the current model calls irrelevant.
Is a brilliant thinker someone who notices more—or someone who knows what does not deserve to be noticed yet?
Until the next strange question,
Osagie
My current hunch: Expertise is partly a learned attention policy: it makes good reasoning faster by discarding most available information.
Most likely reason I am wrong: The filter may be an output of deeper knowledge and memory rather than a distinct component of intelligence.