Agreement Is Not the Same as Understanding

A conversation can appear to be going well right up until somebody tries to go backward. Two people are talking easily, each response seems to follow from the one before it, and they may even be agreeing. Then one person says, “Wait, that's not quite what I meant.”

Sometimes the correction is easy. The other person asks what they missed, the earlier interpretation changes and the conversation continues from there. Nothing particularly dramatic happened; one model of what the other person meant was replaced by a better one. Other times the correction feels strangely expensive. The person has to defend why they're correcting themselves, earlier statements get quoted back as though consistency matters more than clarification, and both people begin spending more effort protecting the structure of the exchange than examining whatever they were originally trying to understand.

They may still eventually agree, which is one reason I'm no longer sure agreement tells us very much about whether a conversation worked. Agreement is simply a conclusion two people happen to share, and they can reach it through careful examination, social pressure, exhaustion, misunderstanding, politeness or coincidence. Two people can also understand one another remarkably well and remain in genuine disagreement.

The more interesting question is whether the conversation can survive revision. Can I tell you that you've misunderstood me without making the previous twenty minutes collapse? Can you change your mind without needing to pretend you never held the earlier position? Can either of us introduce information that complicates the emerging explanation? If we can, the conversation still has somewhere to go.

I've sometimes used the word “alignment” for this, although I think I was asking that word to carry too much. Alignment can sound as though two people have achieved some deeper shared state beneath their explicit disagreement. We don't need to claim that. What we can observe is whether the interaction remains revisable.

That matters because conversation is full of compression. I say something, you interpret it, and your response necessarily depends on a smaller model of what I meant than everything I might have meant. I hear your response and do the same thing. Most of the time this works astonishingly well, which is why we can have ordinary conversations without stopping every few seconds to define our terms.

Sometimes the compression is wrong. Tone gets interpreted as intent. Fluency gets interpreted as understanding. Familiar language makes two people think they're using a word in the same way when they aren't. A quick “yes” gets carried forward as agreement with a much larger proposition than the person intended.

None of those mistakes is especially dangerous while correction remains cheap. The trouble begins when the conversation builds on the misunderstanding faster than the misunderstanding can be repaired. Every additional exchange adds structure above the original error, so correcting it eventually means changing more than one interpretation. We have to reconsider everything subsequently built upon it.

This can produce a peculiar kind of conversation in which the exchange remains fluent while becoming progressively less accurate. Both people are responding intelligently, the sentences make sense and the reasoning may even be internally coherent. They're simply no longer talking about quite the same thing.

Fluency isn't enough to tell us whether that has happened, and neither is discomfort. A useful conversation can become uncomfortable because somebody has encountered evidence they don't like. Clarification can sound defensive because sometimes it is defensive. A pause can indicate careful thought, confusion, annoyance or nothing at all. No single signal gives us privileged access to the state of the interaction.

What we can do is leave room to test our interpretations. “Is that what you meant?” “I think I carried that farther than you intended.” “I agreed with the first part, not the conclusion.” “I don't know yet.” These aren't interruptions to understanding. They're part of the mechanism by which understanding gets corrected.

This is also why I don't think a good conversation is necessarily slow. Some conversations move extremely quickly because the participants share years of context and can compress safely. A word, expression or unfinished sentence can carry an enormous amount because both people have repeatedly tested what those shortcuts mean. Other conversations need to move more slowly because the shared context isn't there yet.

Speed isn't the important property. What matters is whether the conversation remains recoverable when the shortcuts fail. If we move quickly and discover that we misunderstood something three exchanges ago, can we return there without treating the return itself as a problem?

The same distinction changes how I think about disagreement. Disagreement isn't automatically evidence that two people failed to understand one another. Sometimes understanding makes the disagreement sharper because each person finally sees exactly where the difference lies. They may discover that they share the same observations but assign different importance to them, agree about the desired outcome while disagreeing about the likely effects of an action, or realize that a word they've been using casually contains two different definitions.

Now the disagreement has structure. Neither person has to disappear for the conversation to become coherent.

The opposite can happen with agreement. Someone may agree because contradicting a supervisor is costly. A friend may say “you're right” because they're tired of arguing. Two people may use the same philosophical or political label while holding substantially different beliefs underneath it. Agreement can conceal difference just as disagreement can reveal it.

So the goal can't simply be to reduce the distance between conclusions. It has to include improving the representation each person has of the other, while remembering that representation will always be incomplete. I don't get direct access to what you mean. I infer it from words, tone, history, behavior and whatever else the interaction makes available, and you do the same with me.

Conversation works because those models can be corrected.

This is where the idea begins to extend beyond conversation. Organizations build models of what customers need. Journalists build models of what happened. Teachers build models of what students understand. Software systems represent states that originated in much messier human activity. Every representation becomes useful partly because it leaves something out, and in each case the important question is whether what was left out has a route back in when it turns out to matter.

Conversation has the same problem at human scale. I form an interpretation of what you said, and that interpretation lets me respond without asking you to define every word from first principles. If I guessed correctly enough, we continue. If I didn't, you need some way to alter my interpretation.

A conversation becomes brittle when that correction becomes expensive. Status may make contradiction risky. An audience may make changing position embarrassing. Emotional investment may turn clarification into a threat. Sometimes we've simply spent so long building an argument that neither person wants to discover that its foundation was a misunderstanding. Whatever the cause, the conversation gradually becomes better at preserving its current shape than learning from the people inside it.

That, I think, is a more useful way to describe drift. It isn't disagreement, speed or discomfort. It's the gradual loss of inexpensive correction.

Once I see it that way, alignment becomes less mysterious. We don't need to sense whether two people occupy the same conceptual field or share some hidden orientation. We can ask whether misunderstanding can be corrected, whether uncertainty can remain uncertainty for a while, whether somebody can say “I was wrong,” and whether a conclusion can reopen when new information arrives.

None of those properties guarantees a good conversation. People can remain open to revision and still misunderstand one another. They can behave generously while reasoning badly. They can understand one another perfectly well and discover that there is simply a disagreement they cannot resolve.

That's fine because a conversation doesn't have to manufacture agreement to justify itself. Sometimes its most useful result is a better disagreement. Sometimes it's a question neither person had when they began. Sometimes one person changes their mind three days later, and sometimes nothing changes except that each person now knows where the other actually stands.

Perhaps that's what I was reaching for when I distinguished alignment from agreement. I wasn't really describing a hidden state between people. I was describing an interaction capable of changing without treating change as damage.

Agreement may emerge from that, or it may not. Understanding doesn't require us to arrive at the same place.

It requires enough room to discover when we haven't.