This is an automated rejection. No LLM generated, heavily assisted/co-written, or otherwise reliant work.
Read full explanation
Epistemic Status: Strong opinion held with high confidence regarding the mechanics of communication, based on clinical psychology literature on mentalization and hypermentalizing. Open to refinement on edge cases in crisis intervention.
TL;DR: When we express certainty about others' internal states ("I know how you feel"), we're treating our simulation as direct access. This creates pressure on them to validate our model rather than correct it. The clinical literature on hypermentalizing supports this concern. Proposed alternative: express empathic hypotheses with explicit uncertainty, what I call "Scientific Kindness."
The Core Problem
We need to talk about the machinery of connection.
You know that moment where someone tells you exactly how you are feeling. They look you in the eye. They are warm. They are sure. And they are completely wrong.
But here is the trap. Because they are so nice about it, and so confident, you feel like the broken one for correcting them.
This is not just a calibration error. It is a structural failure.
I'm not talking about kindness. Kindness is good. I am talking about "Empathy" as a specific mechanical claim. The claim that one person can access the inside of another person's head. The difference between saying "I am trying to understand you" and "I understand you."
You call it deep connection. I would argue it functions as an access violation.
Tabooing "Empathy"
Words can be very slippery. Let's taboo and look at the gears underneath the label. (For a broader survey of empathy research, see David Gross's excellent Notes on Empathy, which covers the conceptual landscape in depth.)
When people practice this high-stakes empathy, three things happen:
The Target: The other person's internal state. The black box.
The Stance: Achievement. Not "I am looking," but "I have arrived."
The Expression: Confidence. "I feel your pain." "I get it."
Strip away the confidence. Strip away the claim of access. So what do we actually have remaining? You are left with curiosity or kindness. But we reserve the special word "Empathy" for that moment where the gap between minds seems to vanish.
My sense is that the gap never vanishes. We just pretend it does. And that pretense causes damage.
The Structural Risk: The Three-State Trap
Here is the physics of the problem. Minds are opaque. You cannot see my internal state (The Territory). You can only build a model of it based on what I do (The Map).
When you confidently say "I know how you feel," you are taking your Map and overlaying it on my Territory.
If your map is perfect, nobody notices. But human maps are noisy. We guess wrong. And when you guess wrong with confidence, you force the other person into a bind I call the Three-State Trap.
Look at Alice and Bob.
State 1 (The Reality): Alice feels vague exhaustion. Just a heavy fog.
State 2 (The Model): Bob resonates with her tone. He decides she is angry. He says, "You feel betrayed by your boss."
State 3 (The Conflict): Alice now has to hold two things that do not fit. Her fog. And Bob's certainty.
Because Bob came in hot (confident, caring, "empathic"), Alice cannot easily dismiss him. She has to simulate his model of her. She might even overwrite her own vague feelings with his precise ones just to keep the connection alive.
This is not merely theoretical. The neuroscience of empathy literature distinguishes between three dissociable components: neural resonance (visceral mirroring), prosocial motivation (desire to help), and mentalizing/theory of mind (inferring internal states). As the Empathic Inference notes, empathic inference uses "your own brain as reference" to emulate another's mental state, but this method "doesn't reveal the underlying mechanisms behind the reactions and emotional states."
The more you "care," the more you want to signal that you are right there with them. So you signal certainty about a place you have never been.
The Validation Flip
This creates a backward economic transaction. I call it the Validation Flip.
In a healthy system, the listener supports the speaker. But watch the gears in confident Empathy. The speaker (Bob) feels good because he thinks he "gets it." He needs Alice to nod. If Alice says "No, that's not it," Bob's feeling of connection collapses.
So Alice, the one who is actually in pain, is recruited to validate Bob's model. I've seen this happen so many times.
The psychotherapy literature on alliance ruptures describes exactly this dynamic. Safran, Muran, and Eubanks-Carter (2011) define ruptures as "breakdowns in the collaborative process" that often stem from "failures of empathy" or misattunement. When a therapist confidently misreads a client, the client faces pressure to either correct the therapist (risking the relationship) or comply with the misreading (sacrificing their own reality). The clinical consensus is that repair requires the empathizer to acknowledge their model was wrong, not to double down on their confidence (Safran et al., 2011).
The Dissolution Test
You can test if the magic is real. I call this the Dissolution Test.
If you think you are truly connected, try adding a disclaimer that quantifies your uncertainty.
Imagine Bob changes his approach. Instead of "I feel your pain," he says:
"I have a strong simulation in my head that you feel betrayed. But I cannot see inside your mind. I am likely projecting my own stuff onto you. Feel free to reject this simulation. It costs you nothing."
Watch what happens. The magic of "feeling fully known" dissolves. It collapses into a scientific hypothesis.
This tells us something uncomfortable. The feeling of "being known" relied on Bob hiding his uncertainty. It relied on a lie. Most people would not consent to that lie if you asked them directly. "Do you want me to pretend I am sure when I am guessing?" The answer is no.
Blocking the Escape Routes
I can hear the objections. You want to save the concept.
Some might say that mirror neurons make empathy automatic. But this confuses the signal with the interpretation. Mirror neurons run a simulation in your brain; they do not open a Wi-Fi connection to theirs.
The neuroscience here is worth stating precisely. Heyes & Catmur's 2022 review "What Happened to Mirror Neurons?" notes that "some descriptions of mirror neurons imply telepathy," but this is precisely the error. Mirror neurons create motor resonance, not direct mind-reading. As they observe: the early hype around mirror neurons was "fueled by less than rational currents of thought (e.g., atomism and telepathy)." The actual mechanism provides a low-resolution signal: "Distress detected." That is it. To turn "Distress" into "Betrayal," you have to guess. You call it resonance. It is just a fast inference.
Others argue that real empathy implies humility. But if you are humble inside while using the language of "I understand," you are misleading them. If you say "I have a guess," you are not doing Empathy anymore. You are doing Science.
And what about raising children? Doesn't this contradict how we attune to them? The data says the opposite. This is the clinical phenomenon of Hypermentalizing.
Sharp et al. (2011) introduced the term "hypermentalizing" to describe making "assumptions about other people's mental states that go so far beyond observable data that others may struggle to see how they are justified" (Sharp et al., 2011, Journal of the American Academy of Child and Adolescent Psychiatry). Their research found that in adolescents with borderline traits, "the loss of mentalization is more apparent in the emergence of unusual alternative strategies (excessive theory of mind or hypermentalizing) than in the loss of the capacity per se."
Follow-up work by Sharp & Fonagy confirmed that healthy parents know they cannot read minds: they get curious, they acknowledge the opacity. The pathological pattern is being too sure you know what your kid is thinking. A 2022 meta-analysis in the American Journal of Psychotherapy (McLaren et al., 2022) found consistent associations between hypermentalizing and borderline personality pathology.
Analogy: Goodharting on Connection
There's an alignment parallel here worth noting, not because the mechanisms are the same, but because the failure mode is.
When we train language models with RLHF, we reward outputs that seem helpful, understanding, aligned. The model learns to produce text that pattern-matches to "I understand you," not because it has access to your internal state, but because that phrasing scores well.
The incentive gradient points toward sounding connected, not being connected.
Human empathy can Goodhart in the same way. If Bob's reward signal is Alice's nod (the feeling that he "got it"), then Bob is optimizing for the appearance of understanding, not the fact of it. The better Bob gets at seeming to understand, the less pressure he feels to actually check.
As discussed in Empathy as a Natural Consequence of Learnt Reward Models: "An expected utility maximizer like AIXI should possess a very sophisticated theory of mind and mentalizing capability, but zero empathy... it only simulates you so as to better exploit you to serve its goals."
The lesson: simulation capacity and genuine care are orthogonal. You can be very good at modeling someone and still be optimizing for the wrong thing. The question is not "can you simulate their state?" but "what are you using that simulation for?"
The Fossil: Stone Cold Empathy
There is a long-term cost to this. I call it Stone Cold Empathy.
This happens when you map someone successfully once, and then you stop looking.
Living Empathy: Bob figures Alice out.
Petrification: Bob gets confident. "I know Alice."
Fossilization: Alice changes. Bob's model does not.
When Alice acts outside Bob's model, Bob does not say "My map is old." He says "That is not the real you." The confidence inherent in his "empathy" has made him immune to new data. He is relating to a fossil.
This connects to a broader rationalist concern: the typical mind fallacy. We use ourselves as the reference model, and once that model "works," we stop updating it.
Nuance: The High-Context Exception
You might say, "But I have been married thirty years. I know when he is tired."
Fair enough. In high-repetition systems, the Map gets very close to the Territory. The error rate drops.
But this is where the danger spikes. Because you are right 99% of the time, you stop checking. The one time he is not tired (he is depressed, or sick, or changing), you will miss it. You will bulldoze him with your history of accuracy.
I want to explicitly steelman the counter-position here: Perhaps confident empathy serves a social bonding function that outweighs epistemic costs in many contexts. The feeling of "being known" might be valuable even if literally false, a useful fiction that maintains relationships. My response: this may be true for some contexts, but (1) we should be explicit that it's a fiction rather than discovery, and (2) the contexts where precision matters most (therapy, crisis intervention, supporting someone through change) are precisely where the fiction breaks down most harmfully.
The Alternative: Scientific Kindness
If we drop the magic, what is left?
The Scientific Method.
We have a reliable method for understanding objects. Observe. Hypothesize. Test. Update. We should use the same gear for people.
Observe: Look at the data.
Hypothesize: Build a model.
Test: Check it against reality.
Update: If wrong, fix it.
Validate: Once the model holds weight, validate the experience.
I call this Scientific Kindness. It shifts the stance from Achievement ("I got it") to Inquiry ("I am calibrating").
The Dialogue Difference
Standard Empathy:
"I know you are angry." "I am not." "You sound angry. I can feel it." (Diagnostic Dead-End. Bob is stuck defending his map.)
Scientific Kindness:
"I am noticing a tone that my brain flags as anger. Am I misreading that?" "Actually, I am just excited." "Ah. Thanks for the calibration. Updating." (The system clears. Alice feels seen. Bob gets smarter.)
Coda: Falsification vs. Invalidation
This clears up a confusion that ruins relationships.
Falsification targets your model. ("My guess was wrong.")
Invalidation targets their experience. ("Your feeling is wrong.")
When you use the Scientific Method, you welcome Falsification. You want to know your map is wrong so you can fix it. This protects the other person. You take the hit so their reality can stay intact.
The Crux: How I Could Be Wrong
I want to be clear about the load-bearing wall here.
My argument rests on the idea that direct access to other minds is impossible. That we are stuck with simulations.
The argument breaks if: You can prove humans have a biological "Direct Access" channel. If mirror neurons are actually telepathy. If we can transmit qualia without loss.
If I can know your state without guessing, then confidence is fine. But until you show me that hardware (and the current neuroscience consensus, per Heyes & Catmur (2022), is that mirror neurons are not telepathy), I'm going to assume we are all just guessing. (And I include myself in that.)
Conclusion
We should be wary of any tool that encourages us to lie about what we know. Empathy, as commonly practiced, creates an incentive to overstate our access.
Drop the crystal ball. You do not need it. Just ask.
References
Sharp, C., Pane, H., Ha, C., Venta, A., Patel, A. B., Sturek, J., & Fonagy, P. (2011). Theory of mind and emotion regulation difficulties in adolescents with borderline traits. Journal of the American Academy of Child and Adolescent Psychiatry, 50(6), 563-573. PubMed
Sharp, C., Ha, C., Carbone, C., Kim, S., Perry, K., Williams, L., & Fonagy, P. (2013). Hypermentalizing in adolescent inpatients: Treatment effects and association with borderline traits. Journal of Personality Disorders, 27(1), 3-18. PubMed
McLaren, V., Gallagher, M., Hopwood, C. J., & Sharp, C. (2022). Hypermentalizing and Borderline Personality Disorder: A Meta-Analytic Review. American Journal of Psychotherapy, 75(1), 21-31. Link
Heyes, C., & Catmur, C. (2022). What Happened to Mirror Neurons? Perspectives on Psychological Science, 17(1), 153-168. PMC
Sharp, C., & Fonagy, P. (2008). The parent's capacity to treat the child as a psychological agent: Constructs, measures and implications for developmental psychopathology. Social Development, 17, 737-754.
Safran, J. D., Muran, J. C., & Eubanks-Carter, C. (2011). Repairing alliance ruptures. Psychotherapy, 48(1), 80-87. PubMed
Epistemic Status: Strong opinion held with high confidence regarding the mechanics of communication, based on clinical psychology literature on mentalization and hypermentalizing. Open to refinement on edge cases in crisis intervention.
TL;DR: When we express certainty about others' internal states ("I know how you feel"), we're treating our simulation as direct access. This creates pressure on them to validate our model rather than correct it. The clinical literature on hypermentalizing supports this concern. Proposed alternative: express empathic hypotheses with explicit uncertainty, what I call "Scientific Kindness."
The Core Problem
We need to talk about the machinery of connection.
You know that moment where someone tells you exactly how you are feeling. They look you in the eye. They are warm. They are sure. And they are completely wrong.
But here is the trap. Because they are so nice about it, and so confident, you feel like the broken one for correcting them.
This is not just a calibration error. It is a structural failure.
I'm not talking about kindness. Kindness is good. I am talking about "Empathy" as a specific mechanical claim. The claim that one person can access the inside of another person's head. The difference between saying "I am trying to understand you" and "I understand you."
You call it deep connection. I would argue it functions as an access violation.
Tabooing "Empathy"
Words can be very slippery. Let's taboo and look at the gears underneath the label. (For a broader survey of empathy research, see David Gross's excellent Notes on Empathy, which covers the conceptual landscape in depth.)
When people practice this high-stakes empathy, three things happen:
Strip away the confidence. Strip away the claim of access. So what do we actually have remaining? You are left with curiosity or kindness. But we reserve the special word "Empathy" for that moment where the gap between minds seems to vanish.
My sense is that the gap never vanishes. We just pretend it does. And that pretense causes damage.
The Structural Risk: The Three-State Trap
Here is the physics of the problem. Minds are opaque. You cannot see my internal state (The Territory). You can only build a model of it based on what I do (The Map).
When you confidently say "I know how you feel," you are taking your Map and overlaying it on my Territory.
If your map is perfect, nobody notices. But human maps are noisy. We guess wrong. And when you guess wrong with confidence, you force the other person into a bind I call the Three-State Trap.
Look at Alice and Bob.
Because Bob came in hot (confident, caring, "empathic"), Alice cannot easily dismiss him. She has to simulate his model of her. She might even overwrite her own vague feelings with his precise ones just to keep the connection alive.
This is not merely theoretical. The neuroscience of empathy literature distinguishes between three dissociable components: neural resonance (visceral mirroring), prosocial motivation (desire to help), and mentalizing/theory of mind (inferring internal states). As the Empathic Inference notes, empathic inference uses "your own brain as reference" to emulate another's mental state, but this method "doesn't reveal the underlying mechanisms behind the reactions and emotional states."
The more you "care," the more you want to signal that you are right there with them. So you signal certainty about a place you have never been.
The Validation Flip
This creates a backward economic transaction. I call it the Validation Flip.
In a healthy system, the listener supports the speaker. But watch the gears in confident Empathy. The speaker (Bob) feels good because he thinks he "gets it." He needs Alice to nod. If Alice says "No, that's not it," Bob's feeling of connection collapses.
So Alice, the one who is actually in pain, is recruited to validate Bob's model. I've seen this happen so many times.
The psychotherapy literature on alliance ruptures describes exactly this dynamic. Safran, Muran, and Eubanks-Carter (2011) define ruptures as "breakdowns in the collaborative process" that often stem from "failures of empathy" or misattunement. When a therapist confidently misreads a client, the client faces pressure to either correct the therapist (risking the relationship) or comply with the misreading (sacrificing their own reality). The clinical consensus is that repair requires the empathizer to acknowledge their model was wrong, not to double down on their confidence (Safran et al., 2011).
The Dissolution Test
You can test if the magic is real. I call this the Dissolution Test.
If you think you are truly connected, try adding a disclaimer that quantifies your uncertainty.
Imagine Bob changes his approach. Instead of "I feel your pain," he says:
Watch what happens. The magic of "feeling fully known" dissolves. It collapses into a scientific hypothesis.
This tells us something uncomfortable. The feeling of "being known" relied on Bob hiding his uncertainty. It relied on a lie. Most people would not consent to that lie if you asked them directly. "Do you want me to pretend I am sure when I am guessing?" The answer is no.
Blocking the Escape Routes
I can hear the objections. You want to save the concept.
Some might say that mirror neurons make empathy automatic. But this confuses the signal with the interpretation. Mirror neurons run a simulation in your brain; they do not open a Wi-Fi connection to theirs.
The neuroscience here is worth stating precisely. Heyes & Catmur's 2022 review "What Happened to Mirror Neurons?" notes that "some descriptions of mirror neurons imply telepathy," but this is precisely the error. Mirror neurons create motor resonance, not direct mind-reading. As they observe: the early hype around mirror neurons was "fueled by less than rational currents of thought (e.g., atomism and telepathy)." The actual mechanism provides a low-resolution signal: "Distress detected." That is it. To turn "Distress" into "Betrayal," you have to guess. You call it resonance. It is just a fast inference.
Others argue that real empathy implies humility. But if you are humble inside while using the language of "I understand," you are misleading them. If you say "I have a guess," you are not doing Empathy anymore. You are doing Science.
And what about raising children? Doesn't this contradict how we attune to them? The data says the opposite. This is the clinical phenomenon of Hypermentalizing.
Sharp et al. (2011) introduced the term "hypermentalizing" to describe making "assumptions about other people's mental states that go so far beyond observable data that others may struggle to see how they are justified" (Sharp et al., 2011, Journal of the American Academy of Child and Adolescent Psychiatry). Their research found that in adolescents with borderline traits, "the loss of mentalization is more apparent in the emergence of unusual alternative strategies (excessive theory of mind or hypermentalizing) than in the loss of the capacity per se."
Follow-up work by Sharp & Fonagy confirmed that healthy parents know they cannot read minds: they get curious, they acknowledge the opacity. The pathological pattern is being too sure you know what your kid is thinking. A 2022 meta-analysis in the American Journal of Psychotherapy (McLaren et al., 2022) found consistent associations between hypermentalizing and borderline personality pathology.
Analogy: Goodharting on Connection
There's an alignment parallel here worth noting, not because the mechanisms are the same, but because the failure mode is.
When we train language models with RLHF, we reward outputs that seem helpful, understanding, aligned. The model learns to produce text that pattern-matches to "I understand you," not because it has access to your internal state, but because that phrasing scores well.
The incentive gradient points toward sounding connected, not being connected.
Human empathy can Goodhart in the same way. If Bob's reward signal is Alice's nod (the feeling that he "got it"), then Bob is optimizing for the appearance of understanding, not the fact of it. The better Bob gets at seeming to understand, the less pressure he feels to actually check.
As discussed in Empathy as a Natural Consequence of Learnt Reward Models: "An expected utility maximizer like AIXI should possess a very sophisticated theory of mind and mentalizing capability, but zero empathy... it only simulates you so as to better exploit you to serve its goals."
The lesson: simulation capacity and genuine care are orthogonal. You can be very good at modeling someone and still be optimizing for the wrong thing. The question is not "can you simulate their state?" but "what are you using that simulation for?"
The Fossil: Stone Cold Empathy
There is a long-term cost to this. I call it Stone Cold Empathy.
This happens when you map someone successfully once, and then you stop looking.
When Alice acts outside Bob's model, Bob does not say "My map is old." He says "That is not the real you." The confidence inherent in his "empathy" has made him immune to new data. He is relating to a fossil.
This connects to a broader rationalist concern: the typical mind fallacy. We use ourselves as the reference model, and once that model "works," we stop updating it.
Nuance: The High-Context Exception
You might say, "But I have been married thirty years. I know when he is tired."
Fair enough. In high-repetition systems, the Map gets very close to the Territory. The error rate drops.
But this is where the danger spikes. Because you are right 99% of the time, you stop checking. The one time he is not tired (he is depressed, or sick, or changing), you will miss it. You will bulldoze him with your history of accuracy.
I want to explicitly steelman the counter-position here: Perhaps confident empathy serves a social bonding function that outweighs epistemic costs in many contexts. The feeling of "being known" might be valuable even if literally false, a useful fiction that maintains relationships. My response: this may be true for some contexts, but (1) we should be explicit that it's a fiction rather than discovery, and (2) the contexts where precision matters most (therapy, crisis intervention, supporting someone through change) are precisely where the fiction breaks down most harmfully.
The Alternative: Scientific Kindness
If we drop the magic, what is left?
The Scientific Method.
We have a reliable method for understanding objects. Observe. Hypothesize. Test. Update. We should use the same gear for people.
I call this Scientific Kindness. It shifts the stance from Achievement ("I got it") to Inquiry ("I am calibrating").
The Dialogue Difference
Standard Empathy:
Scientific Kindness:
Coda: Falsification vs. Invalidation
This clears up a confusion that ruins relationships.
When you use the Scientific Method, you welcome Falsification. You want to know your map is wrong so you can fix it. This protects the other person. You take the hit so their reality can stay intact.
The Crux: How I Could Be Wrong
I want to be clear about the load-bearing wall here.
My argument rests on the idea that direct access to other minds is impossible. That we are stuck with simulations.
The argument breaks if: You can prove humans have a biological "Direct Access" channel. If mirror neurons are actually telepathy. If we can transmit qualia without loss.
If I can know your state without guessing, then confidence is fine. But until you show me that hardware (and the current neuroscience consensus, per Heyes & Catmur (2022), is that mirror neurons are not telepathy), I'm going to assume we are all just guessing. (And I include myself in that.)
Conclusion
We should be wary of any tool that encourages us to lie about what we know. Empathy, as commonly practiced, creates an incentive to overstate our access.
Drop the crystal ball. You do not need it. Just ask.
References