How to Think Clearly with Dialogue: Others, Self, and AI

Unstructured Conversations

Most conversations are unstructured. Topics change on a whim, and lengthy speeches can become tedious if they cover familiar ground, express common agreement, or are too difficult to follow. This lack of structure is acceptable when the conversation is for leisure or to exchange ideas and information about a broad range of topics. However, when a conversation has a specific goal, some structure will help it accomplish that goal.

Shared Goals

Just as procedures and policies without a shared culture and mission won't make a company effective, structure won't make a conversation productive unless its members share the same goal.

If the goal of a conversation is to evaluate if a belief is true, then its participants need to care more about finding the truth than being correct. Sometimes it’s helpful to state this shared goal explicitly at the outset.

The Method Must Match the Goals

The method must match the goals. The same method won’t work for a sales calls, one-on-ones, staff meetings, company-wide calls, product discovery calls, conversations with your child, planning calls with your spouse, philosophical discussion with friends, and so on. Sometimes no method is needed, but often some structure will help the conversation accomplish its goals. Agendas make business meetings better and time-limits make political debates more useful.

A Method of Conversation

Here is a structured method for two people to explore whether a thesis is consistent with other closely held beliefs. If it is, then it’s more likely they’re true. If it’s not, then either the thesis or one of the held beliefs is false. Thus, the shared goal of the conversation is to progress towards truth.

This method is based on my experience of reading Plato as well as using it with friends and family. I have found it to be effective in evaluating both my own beliefs, whether they are about abstract philosophical topics like those explored by Socrates, or more practical business questions.

It follows these steps:

Agree on a thesis and who will be the questioner and the respondent.
The questioner asks a single question.
The respondent answers.
Repeat steps 2 and 3 until there is:
1. an impasse,
2. sufficient confidence that the thesis is consistent with the respondent’s other beliefs, or
3. agreement that the thesis or roles should be changed.

I suggest writing the questions and answers as you go, especially early in the conversation. Sometimes, we’ll take turns writing at a keyboard even though we’re in person (intermixed with some meta-level discussion.) Online chat platforms can feel more natural than speaking in person for this reason.

Note that the questioner should ask one question. When you have multiple questions, its best to ask them one at a time, since otherwise the conversation begins to split off into multiple lines.

To begin, the questioner should clarify the thesis and any terms used within it. Typically, a few local definitions emerge. Once clarified, the questioner should try to find inconsistencies between the thesis and other beliefs held by the respondent. Good questions are short, require minimal context, and can be answered succinctly.

The questioner should avoid expressing their own beliefs, because the purpose of the method is to determine the internal consistency of the respondents beliefs and not whether the respondent and questioner agree, which is a separate matter. There are several benefits to focusing on internal consistency: it’s tractable. It’s more likely you can change another person’s mind using their own beliefs. Finally, consistency is a prerequisite for truth.

The respondent should take as much time as they need before answering, and the questioner should avoid interrupting with additional thoughts while waiting. Good responses are short, often just a "yes" or "no". There's no need to explain your answer, as any uncertainty can be best addressed with follow-up questions.

Conversations following this method may be uncomfortable. There may be long pauses, and it can feel a little silly, especially at first. The structured nature also requires deeper thinking than unstructured conversation, which can be challenging. As a result, there may be a temptation to abandon the structure.

In addition, there is the issue of control. The respondent selects the thesis, but the questioner controls the direction of the conversation. This can require patience if the respondent doesn't see how a question relates to the thesis. However, in return for giving up control over the conversation's direction, the respondent gets to have their thesis's validity clarified and refined by a mind outside their own.

Self Dialogue

This same method can be applied by writing questions and answers to yourself. Some of the dialogues on this site originated this way. Critical thinking, in many ways, is a dialogue with yourself. You pose a thesis and you question it. Self dialogue does not require willing participants and is thus more accessible (just as books are more accessible than teachers), but it has disadvantages. It is more difficult to identify inconsistencies in one's own thinking than to have another person point them out. Furthermore, we do not know what we do not know, but another person may know things that we do not, and can bring them to our attention.

AI Dialogue

AI chatbots can serve as excellent questioners for your beliefs. In many ways, this is an ideal application of these bots, since hallucinations (when the model makes up things that aren’t true) aren't an issue if they're merely asking you questions. Furthermore, the full dialogue can often fit within their context window.

AI chat bots don’t say “I’m in a hurry now, and it is time for me to go.” And they don’t get frustrated and complain when that you’re making their arguments “go round and not stay in the same place.” The chat bots will be patient while you respond, and won’t leave the structure you provide them. Here is a prompt that’s been useful for structuring conversations with ChatGPT:

Act as if you were Socrates and ask me questions about my thesis in an attempt to show that I'm being inconsistent with my own beliefs. Be sarcastic, make jokes sometimes, and use concrete examples. Begin by asking open-ended questions to gather relevant context about my thesis, then transition to asking questions that can be answered succinctly and with a "yes" or "no" most of the time. Ask me one question at a time. Criticize me if I don’t respond to your questions or get off track, and don’t back down if you think I’m wrong about something. My thesis is …

ChatGPT also allows you return to prior points in a dialogue and explore how it may have progressed if you answered differently. This is useful if you realize that you disagree with an earlier response.

On the other hand, AI Bots have real disadvantages, even beyond their, as of yet, inferior reasoning. Since you’re in control of the conversation, they can’t push back if don’t understand their reasoning or simply want to disengage; they won’t pressure you to face the inconsistencies in your strongly held beliefs if you don’t want to. (Although, most people won’t do this with people either.)

Here are a few examples of dialogues with ChatGPT:

When reading these, you may notice how you often need to help get ChatGPT back on track. This can become annoying. I suspect ChatGPT’s “system message” mechanism could be used to help keep it role-playing without being reminded so often.

How to Be a Better Reader of Plato

Since actively practicing the method outlined above, I've become more engaged in Plato's dialogues.

I first read Plato to understand what he believed. However, reading them proved to be frustrating. Many of his ideas seem strange and disconnected from our current understanding of the physical world and language (for example, read Timaeus). This is compounded by internal inconsistencies within his works. I now believe, as many before me have, that the dialogues were primarily intended to exemplify the ethos and method Socrates used to examine the truth. My expectation that the dialogues to systematically present his views like a treatise was misplaced. Ward Farnsworth nicely highlights this distinction in his book, The Socratic Method:

Questions and answers are the sound of thought happening. An essay or lecture is usually the sound of thought having happened, then polished up so the result is clear and the process of getting there is no longer visible. Ordinarily that’s good. If you know what you think and want someone else to know it, explaining it straight out makes sense. But if you want to provide a model for getting there—because it illustrates the process of figuring that out.

Since adjusting my expectations, I'm now more interested in how Socrate’s conversations break down and how he formulated questions to propel his discussions. If you're reading Plato, I encourage you to practice structured dialogues with a like-minded friend or ChatGPT. It will make you a better reader of Plato, much like active reading in general helps one learn more than passive reading.