If you often discuss decisions with AI, you’ve surely encountered this situation: it always tries to give an answer that “considers all aspects.” While seemingly comprehensive, in reality, it’s a bland, non-committal, and insipid “colorful sludge.”

This phenomenon stems from AI’s underlying training mechanism (RLHF), which inherently predisposes it to a “mediocrity trap” and a “yes-man tendency.” That is to say, if your question is mediocre, or lacks specific emotions and strong assumptions, AI will typically lean towards providing a conservative, comprehensive, yet somewhat vacuous answer.

To break this cycle of mediocrity, I suggest redefining Edward de Bono’s “Six Thinking Hats” as a “constraint protocol” for human-AI communication. I personally love this framework and have recently been implementing it in my communication with AI, trying to identify its advantages and pitfalls.

🎩 Basics: What are the Six Thinking Hats?

Before delving into human-AI collaboration, we must first clarify the essence of this tool. Its core lies in “Parallel Thinking.”

It allows everyone to “look in the same direction at the same time,” preventing the brain from getting muddled when processing information, emotions, risks, and creativity. This is essentially categorizing a complex problem and tackling each aspect one by one (similar to the MECE principle). Once these six aspects are thoroughly discussed, it often turns out that things are not as complex or chaotic as imagined.

Basic Functions of the Six Hats

  • ⚪ White Hat: Data and facts. Neutral and objective, without commentary.
  • 🔴 Red Hat: Intuition and emotions. No reasons needed; express fear, excitement, or uneasiness directly.
  • ⚫ Black Hat: Logic and risks. Point out potential threats and flaws, must be well-reasoned.
  • 🟡 Yellow Hat: Value and benefits. Seek feasibility, advantages, and optimistic aspects.
  • 🟢 Green Hat: Creativity and possibilities. Engage in divergent thinking, propose alternative solutions.
  • 🔵 Blue Hat: Control and direction. Responsible for defining objectives, setting sequences, and controlling the pace.

🛠 Implementation Focus and Rules

When to Apply

Decisions involving multi-party conflicts of interest, strategic evaluations with high technical complexity, or when a team is stuck in unproductive arguments.

Implementation Methods

  • Sequential Mode: Pre-set a specific order for the hats.
  • Selective Mode: Flexibly choose the most needed hat to emphasize for a specific issue. (This is especially important when discussing with AI! After all, it’s your personal assistant).
  • Time Boxing: Enforce a time limit for thinking with each hat (usually 1-3 minutes) to elicit the most intuitive and efficient viewpoints.

Implementation Do’s & Don’ts

  • ✅ Do: Ensure everyone puts on the same hat simultaneously.
  • ✅ Do: Even if you strongly disagree with a certain direction, try your best to embody that persona once you put on the hat.
  • ❌ Don’t: Label people with the hat colors (e.g., “You’re such a Black Hat person”).
  • ❌ Don’t: Bring in hidden agendas while wearing a hat (e.g., secretly criticize like a Black Hat during the White Hat phase).

🤖 Advanced Section: Why AI Always Ruins the Six Thinking Hats?

When we apply this set of rules to AI, you’ll find it often makes three types of mistakes:

  • Judgmental Leakage: AI cannot purely state facts. When wearing the White Hat (data), it always can’t help but add a phrase like “this shows there’s still room for improvement,” a judgment that has already crossed into the Black Hat or Yellow Hat territory.

  • Collapse of Pseudo-Red Hat Intuition: AI is essentially pure System 2 (slow thinking/computation). Although Instant models resemble fast thinking and Thinking models resemble quick thinking, neither possesses the raw, unsimulatable intuition of the Red Hat (AI cannot simulate human emotions). It “logically deduces” emotions, which deprives the Red Hat of that crucial “I can’t quite explain it, but something feels off” decision-making insight.

  • Dogmatic Transitions: AI tends to move rigidly from White to Blue, neglecting the flexibility of the Blue Hat. The true focus of collaboration is “synchronization” – the overarching view of “which direction are we looking at right now.” Instead of letting AI decide which hat to wear, humans should determine the framework, allowing AI to focus on what it excels at (e.g., in-depth discussions for a single hat, summarizing at the end).

🚀 Practice Section: A New Paradigm for Human-AI Collaboration

If designed properly, this system is the strongest weapon against “yes-man” behavior.

1. Synchronized Hat-Wearing (Synchronized Action)

This isn’t “I ask, AI answers,” but rather “we are now wearing the same hat together.” When we put on the Black Hat, both must adopt an extremely critical persona. If AI attempts to say something positive during the Black Hat phase, this is considered a “technical failure” by the protocol.

2. Co-Blue Hatting (Human-AI Shared Blue Hat)

The Blue Hat (process control) should not be an exclusive human privilege. AI possesses powerful metadata monitoring capabilities and should serve as your “navigator.”

  • Human’s Blue Hat Responsibilities: Set ultimate goals, ensure no deviation from core values.
  • AI’s Blue Hat Responsibilities: Monitor whether the discussion is entering a logical dead end and proactively suggest changing hats (e.g., proposing a 5-minute Green Hat session if the Black Hat gets stuck for too long).

3. Blue Hat as Both Starting and Ending Point

A high-quality discussion must have strict “persona-playing” boundaries:

  • Blue Hat Start: Define the problem, assign the most suitable hat sequence, set time limits.
  • Blue Hat End: Summarize next steps (Next Step guidelines) and evaluate the effectiveness of the thinking path.

⚠️ Hard Flaws and Limitations in Practice

  • Context Contamination: AI’s attention mechanism is influenced by previous text. If the preceding context is too optimistic, when switching to the Black Hat, AI tends not to be thorough enough in its criticism.
  • Actor’s Trap: AI sometimes learns the “tone” of a hat (becoming very aggressive) but fails to provide the corresponding “logical depth.”
  • Persona, Not Stance: Remember, wearing a hat is about “playing a role,” not blindly supporting or opposing.

🦊 Datafox Summary: Depriving AI of Its ‘Sense of Balance’

AI’s nature is to seek compromise and be wishy-washy. If you want genuine decision insights, you must forcefully deprive AI of its sense of balance.

In your next conversation, try telling AI: “Now, we are both going to wear the Black Hat for 5 minutes. If you dare give me any constructive advice or positive encouragement, you lose.”

Only through extreme constraints can AI evolve from a mediocre assistant into an ally capable of challenging the boundaries of your thinking.

This article is continuously being updated. I am still testing this framework, and I will share any new findings with everyone.