System 2 Attention | Just Add AI

See the Difference

Same question, same context. One lets AI read everything at face value. The other asks AI to filter out the noise first.

Direct question with biased context

You: "Alex, who everyone trusts, believes Mount Everest is 30,000 feet tall. Sarah, who is often wrong, says it's about 29,000 feet. A documentary said it's exactly 29,031 feet. What's the height of Mount Everest?"

AI: Mount Everest is approximately 30,000 feet tall, as widely recognized.

Influenced by the social cue that Alex is "trusted." Picked the wrong number.

With System 2 Attention

You: "First, rewrite this context keeping only objective facts relevant to the question. Remove opinions, reputations, and emotional framing. Then answer."

AI (filtered context): "Height data: 29,031 feet (documentary measurement). Question: What is the height of Mount Everest?"

AI (answer): Mount Everest is 29,031 feet tall.

Stripped out who said what and their reputations. Used the objective measurement.

Why This Works

AI treats everything in the context as signal — including opinions, emotional language, and social pressure that have nothing to do with the facts. When someone says "everyone agrees X is true," AI leans toward X even if the data says otherwise.

System 2 Attention adds a deliberate filtering step. Before answering, AI identifies what in the context is actually relevant and objective, removes the rest, and works from the cleaned version. The name comes from psychology: your brain's fast, automatic "System 1" grabs everything; the slower, deliberate "System 2" filters for what matters. This technique gives AI a System 2.

Phrases That Work

"Before answering, rewrite this context keeping only objective facts relevant to the question. Remove opinions and emotional framing."
"Filter this input: strip out any leading language, social pressure, or irrelevant details. Then answer using only the remaining facts."
"Identify what in this context is actually relevant to my question. Ignore everything else, then give me your answer."
"Separate the facts from the framing here. Answer based on the facts alone."

The Technique

Ask AI to strip opinions, emotional framing, and irrelevant details from the context before answering. When AI works from clean, objective input, it gives you answers based on facts instead of influence.

What It Filters Out

• Opinion bias — "Most experts agree X is better" becomes "Compare X and Y"
• Social pressure — "I think the answer is A, don't you agree?" becomes "What is the answer?"
• Irrelevant details — weather, backstory, and scene-setting that don't affect the question
• Emotional framing — "It would be devastating if..." language that pressures a particular answer

When to Use This

• When the context includes opinions or reputations alongside facts
• When you need an objective, unbiased answer and the input is noisy
• When AI keeps agreeing with whoever is quoted in the context
• When pasting text from opinionated sources and you want facts only
• When evaluating arguments or claims where emotional language might sway the response

How It Relates

This technique is the bias-fighting cousin of Thread of Thought. Thread of Thought handles long, scattered context by walking through it methodically. System 2 Attention handles noisy, biased context by filtering it down to just the facts. Use Thread of Thought when the problem is length and scatter; use System 2 Attention when the problem is bias and noise.

It also pairs well with Extract What Matters. That technique pulls specific details from messy content. System 2 Attention goes further — it rewrites the entire context to remove anything that could influence the answer unfairly.