Perceptionbeats cognition
Some things you see at a glance; others you have to work out. Four timed demos on why chat makes you work out what you used to see.
Some things you see at a glance. A red dot among green ones. A line sloping down. The empty seat on a map. You look, and the answer is already there.
Other things you have to work out. Read one number, hold it, read the next, compare, repeat. The answer isn't on the screen. You build it, step by step, and you can feel the effort.
Good interfaces put the answer on the screen. Weak ones put the ingredients on the screen and leave you to assemble it.
Chat-first AI is the second kind. You ask in words and get pieces back in words, every time — and because you can assemble them, you never notice that you're doing it.
See for yourself. Four fair fights. In each, the same assistant answers the same question twice: once the way chat does today, once the way the best apps do. Both forms ship in real products. No strawmen. Time yourself.
Trend
“Which metric is declining?”
Round 1
Click the metric that is declining.
The answer appears and the clock starts.
Round 2
Click the metric that is declining.
Finish Round 1 first.
The table doesn't show you a trend. It shows you the inputs to one and asks you to compute it. Slope is a shape. The downward line declares itself.
Space
“Which flat is the shortest walk to both my office and the metro?”
Round 1
Click the flat with the shortest combined distance to the office and the metro.
The answer appears and the clock starts.
Round 2
Click the flat with the shortest combined distance to the office and the metro.
Finish Round 1 first.
In the list you read sixteen numbers, add eight pairs and hold the smallest in your head. On the map, “near both” is a position. The pin between the two landmarks declares itself.
Change
“You edited my contract. Did anything else change?”
Round 1
Click the clause that changed but wasn’t mentioned.
The answer appears and the clock starts.
Round 2
Click the clause that changed but wasn’t mentioned.
Finish Round 1 first.
With the summary you either trust the assistant or reread both versions clause by clause. The diff puts every edit on the screen. Seeing is how you audit an agent.
Comparison
“Which of these is the lightest with at least 16 GB of RAM?”
Round 1
Click the lightest laptop with at least 16 GB of RAM.
The answer appears and the clock starts.
Round 2
Click the lightest laptop with at least 16 GB of RAM.
Finish Round 1 first.
The table lost the first fight and won this one. The enemy was never text or numbers. It's making the reader assemble the answer.
Your results
| Fight | Reading | Seeing | Ratio |
|---|---|---|---|
| Sparklines vs table | — | — | — |
| Map vs list | — | — | — |
| Diff vs summary | — | — | — |
| Table vs paragraphs | — | — | — |
Ratios vary by person and by run. The direction is the point — and that you felt the work in the reading rounds.
The cost is invisible
You could do every reading round. It felt fine. That's the trap. For one decision the cost is invisible; across a day of decisions it's the product. Reading never fails the task. It taxes every task, and nobody files a bug for a tax.
Language in, pixels out
Chat gets half of this right. Words are the best way to tell a machine what you want: one sentence carries a whole request. They're the worst way for the machine to tell you what it found, because everything comes back as pieces. Talk to the machine in words. Make it answer in pictures.
Built to be fair. Both forms in each pair ship in real products. The two rounds use matched data with a different answer, so round two can't be solved from memory. Which form comes first is random; run again to swap. The seeing form gets no extra help — no highlights, no radius circles — only the form changes. The times are yours; no ratio is promised up front.