AI Wall Bouncing Series: Inferring the Flow of Thought
I'll introduce how to use Gemini to predict the flow of thoughts.
2025.09.06
Hello. I'm TB, working in the Organization Development department, in charge of organization development.
In workplace interactions, what others say isn't always the complete picture. In reality, there's a thought process behind their statements - some people carefully explain their thinking while others only communicate conclusions.
So, I'd like to introduce a method using Gemini to infer these thought processes.
The Ladder of Inference
As a foundation for understanding thought processes, let me introduce the Ladder of Inference.
The Ladder of Inference is a model that shows in seven stages how we move from observed facts to conclusions and actions.
By climbing the ladder of inference one step at a time, we can trace the thinking process that connects objective facts to subjective actions. Understanding this model allows us to objectively reflect on "why we thought and acted in certain ways" and make healthier judgments. This applies not only to our own thinking but to understanding others as well.
Below, I'll explain each stage with examples:
1. Observing Facts
These are the most objective data or events.
Example: Person A didn't say anything during the meeting.
2. Selecting Facts
We unconsciously select specific information from vast amounts of data.
Example: Focusing on the fact that Person A didn't speak.
3. Adding Meaning
We add our own interpretations or meaning to the selected facts.
Example: Interpreting that Person A "isn't interested in this meeting."
4. Building Assumptions
Based on our interpretations, we make further assumptions.
Example: Assuming Person A didn't prepare for the meeting.
5. Drawing Conclusions
We treat assumptions as facts and draw final conclusions.
Example: Concluding that Person A "has low motivation for work."
6. Adopting Beliefs
The conclusions become established beliefs about the person or situation.
Example: Holding the belief that Person A is unmotivated.
7. Taking Action
We take specific actions based on established beliefs.
Example: Deciding not to assign important work to Person A in the future.
This is an example of a quite biased, unfavorable interpretation, but the thought process varies from person to person even for the same event.## Thought Process-free Gemini Output
To contrast with examples that reveal thought processes, I'll first look at an example of a prompt that doesn't explicitly show the process.
Input
Today, we had a meeting to discuss team issues.
Mr. A didn't say anything during the meeting. Please explain three possible reasons for this behavior.
Output
Here are three possible explanations for why Mr. A didn't speak during the meeting.
---
### 1. Lack of preparation or information
Mr. A may not have fully understood the agenda or lacked the necessary information.
Even if he wanted to say something during the meeting, he might have chosen to remain silent to avoid making irrelevant comments if he couldn't grasp the content.
It's also possible that new information emerged during the meeting, and he was focusing on processing it.
---
### 2. Personality factors or psychological barriers
Mr. A might naturally not be good at speaking in public or have an introverted personality.
Especially when discussions are lively, he may have trouble finding the right timing to speak up and hesitate to join the conversation.
He might also have been anxious about being wrong, or felt overwhelmed by other members' opinions, leading him to refrain from speaking.
---
### 3. Strategic reasons or team role
Mr. A may have deliberately chosen not to speak.
For example, he might have been carefully listening to other members' opinions to grasp the overall picture.
As part of his team role, he might have been serving as a facilitator, note-taker, or consolidator, focusing on monitoring or recording the discussion rather than being a speaker.
In this case, Mr. A's silence could be seen as a strategic action to enhance team productivity.
```## Method 1 - Gemini Output Using Zero-shot Prompting + Chain-of-Thought Prompting
Zero-shot Prompting is a way of writing prompts that give instructions without providing examples.
For topics where simple instructions tend to produce the intended behavior, Zero-shot Prompting is sufficient. Conversely, when some degree of example-based tuning is necessary, Few-shot Prompting (introduced later) becomes necessary.
Chain-of-Thought Prompting is a prompting technique that encourages LLMs to output not just final answers when solving complex problems, but also the thought process and intermediate steps that lead to those answers.
Based on these, I'll explore the thought process in depth.
Input
```markdown
Today, we had a meeting to discuss our team's challenges.
Person A didn't say anything during the meeting. Please explain three possible reasons for this behavior.
Also, please explain step by step, including the thought process.
Output
There are several possible reasons why Person A didn't speak during today's meeting.
Here, I'll explain three examples, including the thought process.
---
### 1. Nervousness or Anxiety
In this case, Person A is **feeling strong pressure about speaking up**.
**Thought Process:**
1. **Before the meeting:** "I wonder if I need to say something today..." and tries to prepare comments in advance.
2. **During the meeting (early stage):** As teammates begin speaking one after another, they become anxious about whether their opinion might be inappropriate or not good enough.
3. **During the meeting (middle stage):** "Should I speak up now? But maybe there will be a better timing" "What if I'm wrong?" They repeatedly question themselves. At this stage, nervousness peaks, and they lose the courage to speak.
4. **During the meeting (late stage):** Time passes without being able to say anything, missing the opportunity to speak. "It's too late now. I'll try next time," they resign themselves.
---
### 2. Observing the Situation and Gathering Information
In this case, Person A is **prioritizing understanding the overall situation rather than speaking**.
**Thought Process:**
1. **Before the meeting:** They understand the general outline of the agenda but feel they lack detailed information. "First, I'll listen carefully to everyone's opinions and grasp the whole picture," they decide.
2. **During the meeting (early stage):** They pay close attention to teammates' opinions and the flow of discussion. They analyze who has what information and how they think.
3. **During the meeting (middle stage):** As the discussion deepens, their own thoughts start to form. However, they judge it's not yet time to speak. "This problem needs more data" or "I might need to talk to Person B privately about this later," they consider next actions.
4. **After the meeting:** Based on information gathered during the meeting, they work on solutions to problems individually or approach specific people for private conversations.
---
### 3. Not Feeling the Need to Speak
In this case, Person A **agrees with the direction of the discussion and feels they have nothing to add**.
**Thought Process:**
1. **Before the meeting:** They have their own thoughts about the agenda, but nothing particularly innovative.
2. **During the meeting (early stage):** Listening to teammates' opinions, they realize these are very similar to their own thoughts. "What Person C is saying is exactly what I was thinking," they acknowledge.
3. **During the meeting (middle stage):** As the discussion progresses, they feel it's heading in a good direction toward solving the issue. "If we continue this way, it should work out. There's no need to interrupt the discussion by speaking up," they decide.
4. **After the meeting:** They have no objections to the meeting's decisions and smoothly move on to the next actions. Although they didn't speak, they feel they contributed sufficiently to the team's conclusion.
As shown, there can be **various thoughts and intentions** behind not speaking up.
It might not simply be due to indifference.
```## Method 2 - Gemini Output Using Few-shot Prompting + Chain-of-Thought Prompting
Few-shot Prompting is a technique that provides examples to improve the quality of responses.
In this case, instead of using Zero-shot Prompting for Chain-of-Thought Prompting, we'll provide examples that correspond to the ladder of inference as Few-shot Prompting.
Input
```markdown
Today, we had a meeting to discuss our team's challenges.
Please explain three possible thought processes that might have led to Person A not speaking at all during the meeting.
For each example, please include an explanation based on the ladder of inference process as shown below.
[Example of the ladder of inference]
### 1. Observing facts
These are the most objective data or events.
Example: Person A did not speak during the meeting.
### 2. Selecting facts
We unconsciously select specific information from the vast amount available.
Example: Focusing on the fact that Person A did not speak.
### 3. Assigning meaning
We add our own interpretation or meaning to the facts we've selected.
Example: Interpreting that Person A "is not interested in this meeting."
### 4. Building assumptions
We make further assumptions based on our interpretations.
Example: Assuming that Person A probably didn't prepare for the meeting.
### 5. Drawing conclusions
We treat assumptions as facts and draw final conclusions.
Example: Concluding that Person A "has low motivation for work."
### 6. Adopting beliefs
The conclusions we draw become established beliefs about the person or situation.
Example: Adopting the belief that Person A is unmotivated.
### 7. Taking action
We take specific actions based on our established beliefs.
Example: Deciding not to assign important work to Person A in the future.
Output
Here are three plausible thought processes for why Person A didn't speak during the meeting, explained using the ladder of inference.
---
### Example 1: Waiting for the Right Opportunity to Speak
In this case, Person A is carefully observing the meeting's progress and trying to determine the best timing to speak.
1. **Observing facts:** Person A did not speak during the meeting.
2. **Selecting facts:** Focusing on the participants present, the complexity of the topics, and the pace of the discussion.
3. **Assigning meaning:** Interpreting that "this topic hasn't reached a conclusion yet" or "my opinion might confuse the discussion if I share it now."
4. **Building assumptions:** Assuming it's best to wait for a more appropriate time to make a constructive suggestion after understanding the full picture of the discussion.
5. **Drawing conclusions:** Concluding that "not speaking at this point is the best choice for the team."
6. **Adopting beliefs:** Holding the belief that one should prioritize the team's overall benefit.
7. **Taking action:** Looking for an opportunity to speak, but ultimately ending the meeting without speaking.
---
### Example 2: Feeling Anxious About Speaking
In this case, Person A lacks confidence in speaking due to past experiences and feels nervous.
1. **Observing facts:** Person A did not speak during the meeting.
2. **Selecting facts:** Focusing on past experiences where their opinions were rejected or they couldn't explain themselves well during meetings.
3. **Assigning meaning:** Interpreting that "my opinion might be wrong" or "I might embarrass myself if I can't express myself clearly."
4. **Building assumptions:** Assuming there's a risk of failing again if they speak, and that it's safer not to speak.
5. **Drawing conclusions:** Concluding that "not speaking is less mentally taxing for me."
6. **Adopting beliefs:** Holding the belief that speaking in meetings is a high-risk activity.
7. **Taking action:** Refraining from speaking during the meeting and focusing on listening to others' opinions.
---
### Example 3: Already Having a Solution or Agreeing with Others
In this case, Person A already has their own answer to the topic and is listening to confirm their thoughts against others' opinions.
1. **Observing facts:** Person A did not speak during the meeting.
2. **Selecting facts:** Focusing on prior knowledge of the topic or solutions they've already formulated.
3. **Assigning meaning:** Interpreting that "I've already reached a conclusion on this topic" or "other members' opinions seem to align with my thinking."
4. **Building assumptions:** Assuming the discussion will head in the right direction without needing to express their opinion.
5. **Drawing conclusions:** Concluding "there's no need to state my opinion."
6. **Adopting beliefs:** Holding the belief that one should trust the team's discussion and avoid unnecessary comments.
7. **Taking action:** Listening to other members' opinions during the meeting, confirming they align with their own thoughts, and concluding the meeting without speaking.
```## Situations for Organizing Others' Thinking
In situations where you need to infer what others are thinking, ideally you would ask questions to clarify uncertainties when topics arise, or verify intentions rather than making assumptions about the other person's words and actions. However, this isn't always possible in every situation.
It's beneficial to revisit and organize your thoughts in situations where you couldn't confirm details or when it feels awkward to ask questions after the fact.
## Additional Notes
Using the ladder of inference to organize thought processes is useful not only for inferring others' thoughts but also for self-reflection.
It makes it easier to verify whether snap judgments were appropriate, whether there were biases, or if there were misinterpretations.
Also, repeatedly organizing both others' thought processes and your own thought processes enhances your understanding of thought processes in general.
As a result, you become more conscious of the details of your own thinking, less likely to be trapped by biases, and better at imagining the details of others' thinking and asking effective questions to confirm your understanding.
## Related Information
* [AI Dialogue Series: Career Consultation | DevelopersIO](https://dev.classmethod.jp/articles/consulting-ai-about-your-career/)
* [AI Dialogue Series: Growth Challenges | DevelopersIO](https://dev.classmethod.jp/articles/growth-issues-with-ai/)
* [AI Dialogue Series: Adaptive Challenges | DevelopersIO](https://dev.classmethod.jp/articles/adaptive-challenges-with-gemini/)
* [AI Dialogue Series: Predicting Decision Outcomes | DevelopersIO](https://dev.classmethod.jp/articles/decision-making-with-gemini/)
* [Chain-of-Thought Prompting | Prompt Engineering Guide](https://www.promptingguide.ai/jp/techniques/cot)
