Fascinating piece! I just have a clarificatory question:
You say that in your paper you used a generative model to “test whether personality disorders could be more accurately understood as relational disorders.” But what actually constitutes this “test”? Did you, for example, check if your model outcomes fit better with existing data than those of competing models of personality disorders? In this post, you only mention running simulations but those alone cannot have been the test because their first basic assumption is that personality disorders are relational disorders*, and so they would have left that claim untested. However, if this is what you did then I think that it is more accurate to say that you simulated possible explanations of how a personality disorder works on the assumption that it is relational, rather than actually test that claim.
Very happy to be corrected, of course 👍
(*I’m a complete subject neophyte but this surely follows from the claim that part of personality’s fundamental function is relational?)
Thanks for the thoughtful response! The paper was purely based on simulations so there were no empirical tests of any sort. I agree that this part of the text could be better phrased. To clarify, by ''test'' I do not mean an empirical test of the claim that personality disorders are relational disorders. Instead, ''test'' in the simulation sense is more about examining the logic behind this theory and checking that its assumptions are internally consistent and can reproduce the phenomenon of interest in silico (than in vitro).
The field of PD is entirely based on the former empirical ''tests'' and that's why we are left with various theories that are completely internally inconsistent (e.g., the idea that the entire tapestry of personality can define disorders of ''personality'', which leads to the paradoxical and unavoidable conclusion that all mental disorders are personality disorders; see https://www.psychiatrymargins.com/p/either-all-psychopathology-is-personality). The aim of this paper is to show that we need more ''logical'' or ''theoretical'' tests to clarify first and foremost (with minimal assumptions) what kind of thing PD is before we actually test it empirically.
Hope this makes sense. But I agree with you I could have phrased this better!
I feel like this post is unnecessarily uncharitable towards other approaches. Beyond formulations such as "reifying solipsistic entities" (wouldn't be caught doing that!), the main aspect I don't understand here, and have wondered about before is the framing that other models are "just descriptive".
For one, getting a better description goes hand in hand with understanding how something works. But more importantly, even descriptive factor models seem to have (multiple, at times competing) explanatory models to make sense of them. Even over time, even dynamically. I really don't see what this critique is about, am I missing something?
The post is fundamentally about the generative and computational model and what it shows. The critique of trait-based description is the just the background and you don’t necessarily have to accept or agree with that critique to engage with the model itself.
I could engage with the model itself, but not meaningfully, as i am not an expert.
On one hand, i think simulations can show you anything you want, and what matters way more is actual data, especially when causality is concerned. For example, there seems to me to be some circularity: Assumption three already presupposes the result of simulation 2. Additionally, it is sometimes more interesting to question what is left out of a model, but again, not enough of an expert to tell.
On the other hand, I think what is proposed here is not controversial and mostly in line with other (modern) approaches to psychopathology, which loops back to my initial comment: I don't see the big difference, but it is made out to be really big. Big enough to revitalize an entire field that is stagnant.
Maybe I am not seeing the difference because I think of another class of theories? I am comparing mostly to HiTOP, in combination with explanatory approaches like those of Randolph Nesse or Eiko Fried, not so much DSM-V.
Thanks for the deep engagement! I agree with you that descriptive models have their place in psychopathology. I would respectfully disagree, however, that these can ''explain'' anything. To briefly illustrate, consider HiTOP (which I am actually a big fan of). Although it does a great job at organizing various psychopathologies and showing symptoms and traits, it does not tell clinicians what kinds of *mechanisms* generate patients' psychopathology.
Importantly, even if HiTOP were to incorporate a bunch of mechanistic variables what it would miss is a precise formal understanding of how those variables operate. For instance, what is mentalizing, splitting, detachment, etc? How do they operate mechanistically to generate suffering? These questions are left completely unaddressed in current literature and it is these questions that this modeling framework aims to address.
I said more or less as it is more adjacent to the network approach in modeling how symptoms interface with one another in creating "nodes" as opposed to factors. That being said, it sounds like it is also incorporative of agent-based modeling techniques (or something similar in the generative phase). I'll have a look at the paper.
Regarding your respectful disagreement, I am starting to wonder if part of the critique that descriptive models "don't explain anything" is a semantic one.
To me, the explanations around descriptive models are "why things are the way they are". The inaugural HiTOP papers list a multitude of strands of evidence that come with such explanations. For example, they might summarize evidence for an underlying genetic framework, which would then lead to explanations like natural selection of complex traits, or cliff-edged fitness models.
It seems to me that you use the phrase to refer to intra-patient processes. I am not aware of those (mostly psychoanalytic) processes have been sufficiently demonstrated empirically, but either way, to me those come in when finding a way for patients to conceptualize themselves, or their interactions with the world.
From my personal experience of moderating a decently large mental health forum, I find that people naturally gravitate towards one way of conceptualizing versus another. I personally find the former frame much more informative than the latter, and would be highly skeptical of a therapist forcing the latter frame on me (or not being able to switch frames, I guess). But to the extent that someone gravitates towards a more psychoanalytic frame, I also think it should be employed, regardless of it's empirical status, but in accordance with the larger body of empirical knowledge.
I guess I'm skeptical that there exists a general mechanistic understanding that is both true in a scientific sense and meaningful to all potential patients. Much like for physical ills, I expect my doctor to diagnose, tell me what to do, and maybe give me a good-enough understanding of what is happening, but not give me an in-depth lecture on the actual mechanistic processes involved. A sufficient layer of abstraction, fitted to the individual.
Thank you for clarifying your position. I suppose there are two things to separate and note.
First, I agree with your practical, clinical point: I would also want a therapist who is more flexible and adapts to my needs (and personally I would never force one way of looking at things to a patient; and I would also not mention computational models!).
Second, this is not the perspective I took in the previous comment. At that comment, I was talking mainly from a scientific point-of-view and was trying to say that what I believe to be the best form of explanation is within-person, mechanistic, generative mathematical models. Of course, you are right that one can review the literature and integrate genetic, biological, psychological, and evolutionary perspectives into a rich, multi-level explanation of mental disorders. But, however comprehensive, this linguistic synthesis is not equivalent in power to a set of well-specified mathematical models of the phenomena of interest.
The reason these things are not equivalent is this: even if we explain things verbally (in real life and in our manuscripts), at the end of the day we need to operationalize them formally in order to use them empirically—for prediction, explanation, and test of theories. This is what happened in every “mature science.” In physics, Newton described his ideas about motion and gravity in words, but they only became truly powerful when he wrote them as equations. Once that happened, people could calculate how planets move, how fast objects fall, and how to build machines that actually worked as intended.
The same applies to evolutionary theory. Darwin’s account was verbal, but it became predictive when population genetics formalized inheritance and selection, and later when evolutionary game theory and the Price equation quantified strategic interactions and changes in trait frequencies.
The essence of the point then is that anything verbal or underspecified is weak; whereas anything formalized, mechanistic, and generative is strong because it forces precision, constrains interpretation, and yields predictions that can actually be wrong.
Again, this differs from the earlier clinical point in that the clinical stance is about flexibility of communication, while the scientific stance is about precision of formulation.
Thank you again for the in-depth response. It makes sense to me.
Ofc, verbal accounts are weaker, all else being equal. I am not quite convinced that verbally connecting multiple fields of knowledge that have themselves matured to empirical rigour is weaker than an empirically mature field in isolation, and (after skimming your recent meta-analysis) I remain skeptical at the prospect of reaching that degree of strength of evidence for mechanistically describing intrapsychic processes, but I am open to being convinced otherwise over time. Either way, it certainly is interesting to think about and keep in mind.
Ofc, my opinion doesn't matter at all, as I am just an interested layman. :) Again, thank you for indulging my questioning, and I wish you all the best in your efforts.
I appreciate the ambition of this piece and the effort to push the field toward richer, more explanatory models of personality disorders. Most clinicians share this frustration with purely descriptive approaches, and perhaps the computational framing outlined could offer interesting new angles. That said, the article’s portrayal of how clinicians currently conceptualize PDs feels like a thin caricature -- one I don’t recognize from my own work or from my colleagues in clinical, neuropsychological, and personality assessment settings.
When my colleagues and I think about PDs, our case conceptualizations almost always involve a dynamic, interactive unfolding of multiple factors that together disrupt a person’s ability to adaptively relate to others, themselves, and their life goals. To illustrate, here’s how I might think through a composite “client” (not a real person, but similar to many I assess):
This client struggles with:
* View of self: He alternates between grandiose fantasies of perfect success with a deep fear of being found incompetent.
* View of others: He believes others have near-malevolent power -- he simultaneously craves their approval, fears their judgment, is intimidated by their competence, and resents their achievements.
* Impact on goals: Any action feels intolerable because it will be imperfect, “confirming” his pathological self-beliefs; and swings between wanting dependence, feeling vexatious toward others, and wishing he didn’t care about their opinions (which he acts out when emotionally compromised).
I would understand this through several interwoven lenses:
1. Dimensional / Neurobiological: Temperament (high neuroticism; strong negative urgency), cognitive profile (gifted but asynchronous with chronological age; rigidity; perhaps orbitofrontal cortex involvement), and body factors (e.g., neuroinflammation, somatic sensitivity).
2. Narrative / Environmental / Family Dynamic: Life events (single-incident traumas, over-attentive caregivers, gifted-school context reinforcing both his grandiosity and self-doubt).
3. Behavioral: Coping strategies that help short-term but harm long-term goals (withdrawal, anger outbursts, numbing with screens, revenge fantasies) and the environmental reinforcers that sustain them.
4. Skills Gaps: Lagging social-emotional problem-solving skills that were masked by his high intellect.
5. Seasonal / Developmental: Unsuccessful resolution of the adolescent struggles to balance dependence and independence needs, compounded by isolation and fear of rejection.
6. "Humanness": Common struggles of being a person, such as experiential avoidance and the challenge of balancing autonomy with relatedness in close relationships, which are exaggerated in him (due to the above factors) and lead to a strong and self-perpetuating shame response.
7. Broader Context: Family of means (both helpful and enabling); modern-life pressures (e.g., transition crises among high-achieving youth).
8. Maturity / Growth Edge: Resistance to accepting meaningful friction in the service of his values.
9. Diagnostic: PD due to (A) impairments in self-functioning (identity and self-direction) and interpersonal functioning (empathy and intimacy) and (B) traits of negative affectivity, antagonism, detachment, disinhibition (to use the Alternative Model of PDs language), "classically" resembling narcissistic and avoidant patterns.
While my “9-lens” model is simply a teaching framework I use with my trainees, almost every clinician I know implicitly works through most of these layers in some form. The thought leaders cited in the article also write and present (at conferences like the Society for Personality Assessment) with an even greater depth of clinical richness, so I find the suggestion that the field is stuck in a purely descriptive rut to be overstated.
(While it’s true the DSM is relatively silent on the development, maintenance, and individual texture of PDs, that’s true for *every* disorder. The richness comes from case conceptualization, not the criteria themselves.)
The reframing of PDs as relational disorders might have promise -- certainly almost everyone agrees the term "personality disorder" is an underwhelming and often unhelpful name for complex, maladaptive patterns of relating to self and others -- but as presented here, this reframe feels thinner and less dynamic than the multi-lens approaches already in use. It also leaves out important levers for intervention, such as addressing the dimensional, behavioral, developmental, and contextual factors, that are often essential for shifting entrenched patterns (“basins of attraction” that develop through both normal systemic processes, such as negative feedback loops, and non-linear dynamic processes, such as when a small perturbation at a critical juncture snowballs into a cascade of misery). And of course, extending unconditional positive regard ("authentic, compassionate, non-judgmental" care) to people with PDs is important, but most clinicians have long been doing this. Most clinicians also know it’s rarely, if ever, sufficient on its own.
I’ll be following the author's Substack with interest to see how this modeling approach develops, especially if future posts integrate a more three-dimensional appreciation for how experienced clinicians conceptualize and intervene in PDs, and if the model is explained in ways that seem less circular (as some of the other commenters have astutely pointed out). My hope is that the computational model will richly complement the clinical depth already present in the field.
(For context: I’m a neuropsychologist, not a psychiatrist, but I doubt most psychiatrists would recognize themselves in the portrayal here either.)
I’d say, in addition to Orestis, that there is a difference between a (good) explanatory model of something and a (good) framework used for clinical assessment. Although the former very much informs the latter, the latter also requires considerations that may not be directly relevant to a mechanistic explanation. For example, consider the “clinical characterization” aspect of psychopathology ( https://onlinelibrary.wiley.com/doi/full/10.1002/wps.20919 ). What you are describing is a good clinical characterization/assessment.
Many thanks for this detailed outline of your clinical practice! I share your concern that all models can be reductive to some extent (and in our paper, we apologized in advance if the model seems reductive to clinicians or people with lived experiences; please see footnote 1).
That said, I want to make one point extremely clear (and will also clarify this in my Substack): There is a HUGE difference between "personality" and "personality disorders". You're correct that all of your nine dimensions are crucial for a *personality assessment*. However, as I have argued elsewhere, all of the dimensions you've mentioned (and many more) cannot be used to define the personality disorders, specifically, because they define all mental disorders, broadly.
To briefly expand this point further, consider the five-factor definition of "personality", which is that personality is one's characteristic way of thinking, feeling, behaving, and relating. Based on this definition, all mental disorders are personality disorders because all mental disorders inherently entail problems in ways of "thinking, feeling, behaving, and relating." Moreover, all mental disorders have impairments in the dimension you've mentioned (as well as other personality dimensions from a psychodynamic perspective). In that sense, all mental disorders are equally "personality disordered".
The only way ahead then is to understand that all mental disorders are equally impaired across domains of personality but are particularly impaired in some domains vs. others. For instance, emotional disorders are particularly impaired in the emotional domain you mentioned while impulse disorders (like substance use or ADHD) are particularly impaired in the behavioural domain. The personality disorders appeared to be particularly impaired in the relational domain which is why it makes sense to define them based on that domain. While also keeping the rigour and richness of the personality assessment for all mental disorders broadly (as you have outlined).
Hope this makes sense. For more information, please check:
Thanks for taking the time to clarify. I understand you’re aiming to define the 'core' of PDs in mechanistic terms, and that your model specifies this 'core' as a dysfunction in how internal and external information about self and others is integrated -- essentially a maladaptive inference process (in many ways similar to Fonagy and colleagues' notion of 'epistemic trust').
However, where I’m still struggling is with how your model is conceptually different from AMPD’s Criterion A, which already defines PDs in terms of impairments in self-functioning (identity, self-direction) and interpersonal functioning (empathy, intimacy). Together, those domains *are* relational -- self-relating and other-relating. So, it’s hard to see what’s added by restating them as “relational impairment”, unless we’re talking about the computational modeling specifics. Certainly, computational modeling would be a meaningful and welcome addition if the formalism allows us to test, simulate, or refine hypotheses about how specific maladaptive inferences lead to specific relational impairments, in a way that goes beyond current non-computational studies on how people with PDs process social, emotional, and interoceptive information. However, without a body of more specific computational data, your “relational core” sounds like a rewording of Criterion A.
I’m also not entirely convinced by the sharp boundary you and Dr. Aftab are drawing between “definition”/"explanatory model" and “case conceptualization”/"clinical characterization." In practice, a dynamic, multi-layered explanation for why self- and interpersonal functioning are impaired -- including neurobiological, developmental, behavioral, contextual, and cultural factors -- is not mere "characterization.” It’s ALSO an explanatory model, albeit one tailored to the individual rather than populations (though it;s captured somewhat at the group level in AMPD's Criterion B, which acknowledges that PDs do NOT in fact seem reducible to just their relational core, and instead seem to also involve cybernetic mechanisms that have evolved across a person's life). Reducing an individualized explanatory model down to “characterization” risks implying that clinicians aren’t already aware of, or working with, the ways people with PDs process social, emotional, and interoceptive information differently from those without PDs, and the centrality of this point to the clinical picture.
(In particular, I'm concerned that this boundary or framing -- with its implications that clinicians who start with Criterion A but then *go beyond it* to try to understand how that relational impairment and the cybernetic mechanisms that sustain it developed in this specific person are both (1) ignoring or ignorant of the relational core of the PD and (2) doing something less sophisticated and/or less central to our understanding of PDs -- will accidentally alienate the audience you would most like to be receptive to your ideas.)
That being said, I'm quite excited about the possibilities for computational modeling of how relational deficits develop in response to specific social-emotional-interoceptive processing deficits. Indeed, I keep thinking about whether this could be integrated with SlimeMoldTimeMold's call to computationally model governors that regulate social and interpersonal drives (e.g., https://slimemoldtimemold.com/2025/02/27/the-mind-in-the-wheel-part-iii-personality-and-individual-differences/), as part of psychology's ongoing quest for a true paradigm. It may be that the real value here lies in integrating levels: using your computational framework to more precisely model the mechanisms that underlie Criterion A impairments, while recognizing that individualized, multi-factorial conceptualization is not an optional extra, but a necessary bridge between population-level mechanisms, individually-evolved cybernetic mechanisms (Criterion B), and the lived complexity of patients’ lives. I'll be following your work to learn more about how you envision all of this fitting together in practice!
Thanks for the thoughtful response! You raised at least two points with which I agree strongly actually. The first point is about the criterion A and its overlap with our model: As you'll see, in our paper, we explicitly state both in our Introduction and Discussion that our model can be understood as a formal model of personality functioning. I should have perhaps clarified this more in the post here to avoid any miscommunications. In any case, to clarify this now: Our model is a formal representation of the concept of "personality functioning" (aka self-other relating), albeit with fewer assumptions (i.e., no need to reify a general latent entity representing one's "personality").
The second point is clinical and, as far as I can tell, it's about how we formulate clinical problems. I agree with you again on this, if I got it right: the more dimensions of human experience we can integrate and clinically formulate, the richer our explanations will be of a *specific patient's problems*. This is where our model falls short because it only aims to explain one dimension of human experience. In the paper, we acknowledge (in footnote 2) that our model is necessarily reductive and that in no way do we believe that we can reduce any patients' difficulties to a simple relational perspective (like the one we outline). This point, then, begs the question: What's the point of our model?
I think the point of our model is to isolate this specific mechanism and show how it operates on its own. This is a necessary first step before we move to more complex models that integrate more dimensions of human experience (e.g., thinking, feeling, deciding, behaving, reflecting, etc, for a complete cybernetic perspective). In that sense, and to conclude, I agree with your other point that these models should aim to approximate the more complex clinical formulations that clinicians make by showing how multiple dimensions interact to generate human suffering. (From a broad perspective, though, such a modeling work would likely differ for each patient as each patient would have their own model with specific interaction components.)
Hope this clarifies our approach. Personally, when I say disorders of "personality" are disorders of "relating", I don't mean they are *only disorders of "relating"*; instead, I only mean that out of all the problems such patients experience, the relational ones are the most striking and prominent (in the same way that cognitive-perceptual problems are the most striking and most prominent prominent problems in those diagnosed with psychotic disorders). Hope this makes sense.
Best Wishes and thanks for the thoughtful engagement!
Feel free to ignore this comment... Comes from a student in ignorance. Personality disorders confuse me... I was puzzled by the examples of 2 patients, one who felt like they'd failed themself, the other that they'd failed others. Were you meaning to feel either of those states is pathological? I imagine a trainee psychiatrist could easily fit into either of those camps if they fail their board exams, depending on how much outside support they had. Or you could feel like you failed both yourself and others. This seems a normal human response to me. Not worthy of narcissistic or dependent.
Or was it the description of them that follows most significant to the pathology? Or by distress do you mean "distress that brings one to a psychiatrist"? v normal human responses?
Thank you for this! To clarify: by ''distress'', I meant subjective distress and with these examples, I meant to highlight the most striking aspects of some clinical cases. I would agree that these people should not receive a ''personality disorder'' diagnosis (but that's because I think no one should be given such a diagnosis). Finally, the main point of that brief clinical illustration is to show that many classic ''personality cases'' can be better framed in ''relational terms''. I did not create those cases. If you'd like to read more about them, have a look at ''introjective'' vs. ''anaclitic'' depression from Sydney Blatt.
PS: A lot of what we term ''mental health problems'' are normal human responses -- so I would agree with you on that as well. What I'd like to add, however, is that this does not preclude the possibility that psychotherapy might help the person in need.
Fascinating piece! I just have a clarificatory question:
You say that in your paper you used a generative model to “test whether personality disorders could be more accurately understood as relational disorders.” But what actually constitutes this “test”? Did you, for example, check if your model outcomes fit better with existing data than those of competing models of personality disorders? In this post, you only mention running simulations but those alone cannot have been the test because their first basic assumption is that personality disorders are relational disorders*, and so they would have left that claim untested. However, if this is what you did then I think that it is more accurate to say that you simulated possible explanations of how a personality disorder works on the assumption that it is relational, rather than actually test that claim.
Very happy to be corrected, of course 👍
(*I’m a complete subject neophyte but this surely follows from the claim that part of personality’s fundamental function is relational?)
Thanks for the thoughtful response! The paper was purely based on simulations so there were no empirical tests of any sort. I agree that this part of the text could be better phrased. To clarify, by ''test'' I do not mean an empirical test of the claim that personality disorders are relational disorders. Instead, ''test'' in the simulation sense is more about examining the logic behind this theory and checking that its assumptions are internally consistent and can reproduce the phenomenon of interest in silico (than in vitro).
The field of PD is entirely based on the former empirical ''tests'' and that's why we are left with various theories that are completely internally inconsistent (e.g., the idea that the entire tapestry of personality can define disorders of ''personality'', which leads to the paradoxical and unavoidable conclusion that all mental disorders are personality disorders; see https://www.psychiatrymargins.com/p/either-all-psychopathology-is-personality). The aim of this paper is to show that we need more ''logical'' or ''theoretical'' tests to clarify first and foremost (with minimal assumptions) what kind of thing PD is before we actually test it empirically.
Hope this makes sense. But I agree with you I could have phrased this better!
Yes this makes good sense, thanks! It’s very interesting: you’re using a simulation model to flesh out a theory that you can then test empirically?
Exactly! There is already considerable evidence in support of the model, though, as outlined in this systematic review:
https://www.nature.com/articles/s44220-025-00465-9
I feel like this post is unnecessarily uncharitable towards other approaches. Beyond formulations such as "reifying solipsistic entities" (wouldn't be caught doing that!), the main aspect I don't understand here, and have wondered about before is the framing that other models are "just descriptive".
For one, getting a better description goes hand in hand with understanding how something works. But more importantly, even descriptive factor models seem to have (multiple, at times competing) explanatory models to make sense of them. Even over time, even dynamically. I really don't see what this critique is about, am I missing something?
The post is fundamentally about the generative and computational model and what it shows. The critique of trait-based description is the just the background and you don’t necessarily have to accept or agree with that critique to engage with the model itself.
Thank you for your answer.
I could engage with the model itself, but not meaningfully, as i am not an expert.
On one hand, i think simulations can show you anything you want, and what matters way more is actual data, especially when causality is concerned. For example, there seems to me to be some circularity: Assumption three already presupposes the result of simulation 2. Additionally, it is sometimes more interesting to question what is left out of a model, but again, not enough of an expert to tell.
On the other hand, I think what is proposed here is not controversial and mostly in line with other (modern) approaches to psychopathology, which loops back to my initial comment: I don't see the big difference, but it is made out to be really big. Big enough to revitalize an entire field that is stagnant.
Maybe I am not seeing the difference because I think of another class of theories? I am comparing mostly to HiTOP, in combination with explanatory approaches like those of Randolph Nesse or Eiko Fried, not so much DSM-V.
Thanks for the deep engagement! I agree with you that descriptive models have their place in psychopathology. I would respectfully disagree, however, that these can ''explain'' anything. To briefly illustrate, consider HiTOP (which I am actually a big fan of). Although it does a great job at organizing various psychopathologies and showing symptoms and traits, it does not tell clinicians what kinds of *mechanisms* generate patients' psychopathology.
Importantly, even if HiTOP were to incorporate a bunch of mechanistic variables what it would miss is a precise formal understanding of how those variables operate. For instance, what is mentalizing, splitting, detachment, etc? How do they operate mechanistically to generate suffering? These questions are left completely unaddressed in current literature and it is these questions that this modeling framework aims to address.
This is just the factor model vs. network model debate more or less.
Not at all — (data-driven) network models are also descriptive (see https://www.tandfonline.com/doi/full/10.1080/1047840X.2020.1853461).
The model here is generative meaning that it can generate data to show how "personality disorder" actually operates.
Please have a look at the full paper — the post is a very condensed summary of it so it may not do full justice to the model.
I said more or less as it is more adjacent to the network approach in modeling how symptoms interface with one another in creating "nodes" as opposed to factors. That being said, it sounds like it is also incorporative of agent-based modeling techniques (or something similar in the generative phase). I'll have a look at the paper.
Thank you for your response.
Regarding your respectful disagreement, I am starting to wonder if part of the critique that descriptive models "don't explain anything" is a semantic one.
To me, the explanations around descriptive models are "why things are the way they are". The inaugural HiTOP papers list a multitude of strands of evidence that come with such explanations. For example, they might summarize evidence for an underlying genetic framework, which would then lead to explanations like natural selection of complex traits, or cliff-edged fitness models.
It seems to me that you use the phrase to refer to intra-patient processes. I am not aware of those (mostly psychoanalytic) processes have been sufficiently demonstrated empirically, but either way, to me those come in when finding a way for patients to conceptualize themselves, or their interactions with the world.
From my personal experience of moderating a decently large mental health forum, I find that people naturally gravitate towards one way of conceptualizing versus another. I personally find the former frame much more informative than the latter, and would be highly skeptical of a therapist forcing the latter frame on me (or not being able to switch frames, I guess). But to the extent that someone gravitates towards a more psychoanalytic frame, I also think it should be employed, regardless of it's empirical status, but in accordance with the larger body of empirical knowledge.
I guess I'm skeptical that there exists a general mechanistic understanding that is both true in a scientific sense and meaningful to all potential patients. Much like for physical ills, I expect my doctor to diagnose, tell me what to do, and maybe give me a good-enough understanding of what is happening, but not give me an in-depth lecture on the actual mechanistic processes involved. A sufficient layer of abstraction, fitted to the individual.
Thank you for clarifying your position. I suppose there are two things to separate and note.
First, I agree with your practical, clinical point: I would also want a therapist who is more flexible and adapts to my needs (and personally I would never force one way of looking at things to a patient; and I would also not mention computational models!).
Second, this is not the perspective I took in the previous comment. At that comment, I was talking mainly from a scientific point-of-view and was trying to say that what I believe to be the best form of explanation is within-person, mechanistic, generative mathematical models. Of course, you are right that one can review the literature and integrate genetic, biological, psychological, and evolutionary perspectives into a rich, multi-level explanation of mental disorders. But, however comprehensive, this linguistic synthesis is not equivalent in power to a set of well-specified mathematical models of the phenomena of interest.
The reason these things are not equivalent is this: even if we explain things verbally (in real life and in our manuscripts), at the end of the day we need to operationalize them formally in order to use them empirically—for prediction, explanation, and test of theories. This is what happened in every “mature science.” In physics, Newton described his ideas about motion and gravity in words, but they only became truly powerful when he wrote them as equations. Once that happened, people could calculate how planets move, how fast objects fall, and how to build machines that actually worked as intended.
The same applies to evolutionary theory. Darwin’s account was verbal, but it became predictive when population genetics formalized inheritance and selection, and later when evolutionary game theory and the Price equation quantified strategic interactions and changes in trait frequencies.
The essence of the point then is that anything verbal or underspecified is weak; whereas anything formalized, mechanistic, and generative is strong because it forces precision, constrains interpretation, and yields predictions that can actually be wrong.
Again, this differs from the earlier clinical point in that the clinical stance is about flexibility of communication, while the scientific stance is about precision of formulation.
Hope this clarifies my perspective.
Thank you again for the in-depth response. It makes sense to me.
Ofc, verbal accounts are weaker, all else being equal. I am not quite convinced that verbally connecting multiple fields of knowledge that have themselves matured to empirical rigour is weaker than an empirically mature field in isolation, and (after skimming your recent meta-analysis) I remain skeptical at the prospect of reaching that degree of strength of evidence for mechanistically describing intrapsychic processes, but I am open to being convinced otherwise over time. Either way, it certainly is interesting to think about and keep in mind.
Ofc, my opinion doesn't matter at all, as I am just an interested layman. :) Again, thank you for indulging my questioning, and I wish you all the best in your efforts.
I appreciate the ambition of this piece and the effort to push the field toward richer, more explanatory models of personality disorders. Most clinicians share this frustration with purely descriptive approaches, and perhaps the computational framing outlined could offer interesting new angles. That said, the article’s portrayal of how clinicians currently conceptualize PDs feels like a thin caricature -- one I don’t recognize from my own work or from my colleagues in clinical, neuropsychological, and personality assessment settings.
When my colleagues and I think about PDs, our case conceptualizations almost always involve a dynamic, interactive unfolding of multiple factors that together disrupt a person’s ability to adaptively relate to others, themselves, and their life goals. To illustrate, here’s how I might think through a composite “client” (not a real person, but similar to many I assess):
This client struggles with:
* View of self: He alternates between grandiose fantasies of perfect success with a deep fear of being found incompetent.
* View of others: He believes others have near-malevolent power -- he simultaneously craves their approval, fears their judgment, is intimidated by their competence, and resents their achievements.
* Impact on goals: Any action feels intolerable because it will be imperfect, “confirming” his pathological self-beliefs; and swings between wanting dependence, feeling vexatious toward others, and wishing he didn’t care about their opinions (which he acts out when emotionally compromised).
I would understand this through several interwoven lenses:
1. Dimensional / Neurobiological: Temperament (high neuroticism; strong negative urgency), cognitive profile (gifted but asynchronous with chronological age; rigidity; perhaps orbitofrontal cortex involvement), and body factors (e.g., neuroinflammation, somatic sensitivity).
2. Narrative / Environmental / Family Dynamic: Life events (single-incident traumas, over-attentive caregivers, gifted-school context reinforcing both his grandiosity and self-doubt).
3. Behavioral: Coping strategies that help short-term but harm long-term goals (withdrawal, anger outbursts, numbing with screens, revenge fantasies) and the environmental reinforcers that sustain them.
4. Skills Gaps: Lagging social-emotional problem-solving skills that were masked by his high intellect.
5. Seasonal / Developmental: Unsuccessful resolution of the adolescent struggles to balance dependence and independence needs, compounded by isolation and fear of rejection.
6. "Humanness": Common struggles of being a person, such as experiential avoidance and the challenge of balancing autonomy with relatedness in close relationships, which are exaggerated in him (due to the above factors) and lead to a strong and self-perpetuating shame response.
7. Broader Context: Family of means (both helpful and enabling); modern-life pressures (e.g., transition crises among high-achieving youth).
8. Maturity / Growth Edge: Resistance to accepting meaningful friction in the service of his values.
9. Diagnostic: PD due to (A) impairments in self-functioning (identity and self-direction) and interpersonal functioning (empathy and intimacy) and (B) traits of negative affectivity, antagonism, detachment, disinhibition (to use the Alternative Model of PDs language), "classically" resembling narcissistic and avoidant patterns.
While my “9-lens” model is simply a teaching framework I use with my trainees, almost every clinician I know implicitly works through most of these layers in some form. The thought leaders cited in the article also write and present (at conferences like the Society for Personality Assessment) with an even greater depth of clinical richness, so I find the suggestion that the field is stuck in a purely descriptive rut to be overstated.
(While it’s true the DSM is relatively silent on the development, maintenance, and individual texture of PDs, that’s true for *every* disorder. The richness comes from case conceptualization, not the criteria themselves.)
The reframing of PDs as relational disorders might have promise -- certainly almost everyone agrees the term "personality disorder" is an underwhelming and often unhelpful name for complex, maladaptive patterns of relating to self and others -- but as presented here, this reframe feels thinner and less dynamic than the multi-lens approaches already in use. It also leaves out important levers for intervention, such as addressing the dimensional, behavioral, developmental, and contextual factors, that are often essential for shifting entrenched patterns (“basins of attraction” that develop through both normal systemic processes, such as negative feedback loops, and non-linear dynamic processes, such as when a small perturbation at a critical juncture snowballs into a cascade of misery). And of course, extending unconditional positive regard ("authentic, compassionate, non-judgmental" care) to people with PDs is important, but most clinicians have long been doing this. Most clinicians also know it’s rarely, if ever, sufficient on its own.
I’ll be following the author's Substack with interest to see how this modeling approach develops, especially if future posts integrate a more three-dimensional appreciation for how experienced clinicians conceptualize and intervene in PDs, and if the model is explained in ways that seem less circular (as some of the other commenters have astutely pointed out). My hope is that the computational model will richly complement the clinical depth already present in the field.
(For context: I’m a neuropsychologist, not a psychiatrist, but I doubt most psychiatrists would recognize themselves in the portrayal here either.)
I’d say, in addition to Orestis, that there is a difference between a (good) explanatory model of something and a (good) framework used for clinical assessment. Although the former very much informs the latter, the latter also requires considerations that may not be directly relevant to a mechanistic explanation. For example, consider the “clinical characterization” aspect of psychopathology ( https://onlinelibrary.wiley.com/doi/full/10.1002/wps.20919 ). What you are describing is a good clinical characterization/assessment.
Many thanks for this detailed outline of your clinical practice! I share your concern that all models can be reductive to some extent (and in our paper, we apologized in advance if the model seems reductive to clinicians or people with lived experiences; please see footnote 1).
That said, I want to make one point extremely clear (and will also clarify this in my Substack): There is a HUGE difference between "personality" and "personality disorders". You're correct that all of your nine dimensions are crucial for a *personality assessment*. However, as I have argued elsewhere, all of the dimensions you've mentioned (and many more) cannot be used to define the personality disorders, specifically, because they define all mental disorders, broadly.
To briefly expand this point further, consider the five-factor definition of "personality", which is that personality is one's characteristic way of thinking, feeling, behaving, and relating. Based on this definition, all mental disorders are personality disorders because all mental disorders inherently entail problems in ways of "thinking, feeling, behaving, and relating." Moreover, all mental disorders have impairments in the dimension you've mentioned (as well as other personality dimensions from a psychodynamic perspective). In that sense, all mental disorders are equally "personality disordered".
The only way ahead then is to understand that all mental disorders are equally impaired across domains of personality but are particularly impaired in some domains vs. others. For instance, emotional disorders are particularly impaired in the emotional domain you mentioned while impulse disorders (like substance use or ADHD) are particularly impaired in the behavioural domain. The personality disorders appeared to be particularly impaired in the relational domain which is why it makes sense to define them based on that domain. While also keeping the rigour and richness of the personality assessment for all mental disorders broadly (as you have outlined).
Hope this makes sense. For more information, please check:
https://osf.io/preprints/psyarxiv/4xpdg_v1
https://www.psychiatrymargins.com/p/either-all-psychopathology-is-personality?hide_intro_popup=true
https://scholar.google.com/scholar?hl=en&as_sdt=0%2C5&q=orestis+Zavlis+the+illusion+of+personality+&btnG=#d=gs_qabs&t=1754815852198&u=%23p%3DZk0yiDD0fsoJ
Hi Orestis,
Thanks for taking the time to clarify. I understand you’re aiming to define the 'core' of PDs in mechanistic terms, and that your model specifies this 'core' as a dysfunction in how internal and external information about self and others is integrated -- essentially a maladaptive inference process (in many ways similar to Fonagy and colleagues' notion of 'epistemic trust').
However, where I’m still struggling is with how your model is conceptually different from AMPD’s Criterion A, which already defines PDs in terms of impairments in self-functioning (identity, self-direction) and interpersonal functioning (empathy, intimacy). Together, those domains *are* relational -- self-relating and other-relating. So, it’s hard to see what’s added by restating them as “relational impairment”, unless we’re talking about the computational modeling specifics. Certainly, computational modeling would be a meaningful and welcome addition if the formalism allows us to test, simulate, or refine hypotheses about how specific maladaptive inferences lead to specific relational impairments, in a way that goes beyond current non-computational studies on how people with PDs process social, emotional, and interoceptive information. However, without a body of more specific computational data, your “relational core” sounds like a rewording of Criterion A.
I’m also not entirely convinced by the sharp boundary you and Dr. Aftab are drawing between “definition”/"explanatory model" and “case conceptualization”/"clinical characterization." In practice, a dynamic, multi-layered explanation for why self- and interpersonal functioning are impaired -- including neurobiological, developmental, behavioral, contextual, and cultural factors -- is not mere "characterization.” It’s ALSO an explanatory model, albeit one tailored to the individual rather than populations (though it;s captured somewhat at the group level in AMPD's Criterion B, which acknowledges that PDs do NOT in fact seem reducible to just their relational core, and instead seem to also involve cybernetic mechanisms that have evolved across a person's life). Reducing an individualized explanatory model down to “characterization” risks implying that clinicians aren’t already aware of, or working with, the ways people with PDs process social, emotional, and interoceptive information differently from those without PDs, and the centrality of this point to the clinical picture.
(In particular, I'm concerned that this boundary or framing -- with its implications that clinicians who start with Criterion A but then *go beyond it* to try to understand how that relational impairment and the cybernetic mechanisms that sustain it developed in this specific person are both (1) ignoring or ignorant of the relational core of the PD and (2) doing something less sophisticated and/or less central to our understanding of PDs -- will accidentally alienate the audience you would most like to be receptive to your ideas.)
That being said, I'm quite excited about the possibilities for computational modeling of how relational deficits develop in response to specific social-emotional-interoceptive processing deficits. Indeed, I keep thinking about whether this could be integrated with SlimeMoldTimeMold's call to computationally model governors that regulate social and interpersonal drives (e.g., https://slimemoldtimemold.com/2025/02/27/the-mind-in-the-wheel-part-iii-personality-and-individual-differences/), as part of psychology's ongoing quest for a true paradigm. It may be that the real value here lies in integrating levels: using your computational framework to more precisely model the mechanisms that underlie Criterion A impairments, while recognizing that individualized, multi-factorial conceptualization is not an optional extra, but a necessary bridge between population-level mechanisms, individually-evolved cybernetic mechanisms (Criterion B), and the lived complexity of patients’ lives. I'll be following your work to learn more about how you envision all of this fitting together in practice!
Warmly,
Stephanie
Hi Stephanie,
Thanks for the thoughtful response! You raised at least two points with which I agree strongly actually. The first point is about the criterion A and its overlap with our model: As you'll see, in our paper, we explicitly state both in our Introduction and Discussion that our model can be understood as a formal model of personality functioning. I should have perhaps clarified this more in the post here to avoid any miscommunications. In any case, to clarify this now: Our model is a formal representation of the concept of "personality functioning" (aka self-other relating), albeit with fewer assumptions (i.e., no need to reify a general latent entity representing one's "personality").
The second point is clinical and, as far as I can tell, it's about how we formulate clinical problems. I agree with you again on this, if I got it right: the more dimensions of human experience we can integrate and clinically formulate, the richer our explanations will be of a *specific patient's problems*. This is where our model falls short because it only aims to explain one dimension of human experience. In the paper, we acknowledge (in footnote 2) that our model is necessarily reductive and that in no way do we believe that we can reduce any patients' difficulties to a simple relational perspective (like the one we outline). This point, then, begs the question: What's the point of our model?
I think the point of our model is to isolate this specific mechanism and show how it operates on its own. This is a necessary first step before we move to more complex models that integrate more dimensions of human experience (e.g., thinking, feeling, deciding, behaving, reflecting, etc, for a complete cybernetic perspective). In that sense, and to conclude, I agree with your other point that these models should aim to approximate the more complex clinical formulations that clinicians make by showing how multiple dimensions interact to generate human suffering. (From a broad perspective, though, such a modeling work would likely differ for each patient as each patient would have their own model with specific interaction components.)
Hope this clarifies our approach. Personally, when I say disorders of "personality" are disorders of "relating", I don't mean they are *only disorders of "relating"*; instead, I only mean that out of all the problems such patients experience, the relational ones are the most striking and prominent (in the same way that cognitive-perceptual problems are the most striking and most prominent prominent problems in those diagnosed with psychotic disorders). Hope this makes sense.
Best Wishes and thanks for the thoughtful engagement!
Feel free to ignore this comment... Comes from a student in ignorance. Personality disorders confuse me... I was puzzled by the examples of 2 patients, one who felt like they'd failed themself, the other that they'd failed others. Were you meaning to feel either of those states is pathological? I imagine a trainee psychiatrist could easily fit into either of those camps if they fail their board exams, depending on how much outside support they had. Or you could feel like you failed both yourself and others. This seems a normal human response to me. Not worthy of narcissistic or dependent.
Or was it the description of them that follows most significant to the pathology? Or by distress do you mean "distress that brings one to a psychiatrist"? v normal human responses?
Thank you for this! To clarify: by ''distress'', I meant subjective distress and with these examples, I meant to highlight the most striking aspects of some clinical cases. I would agree that these people should not receive a ''personality disorder'' diagnosis (but that's because I think no one should be given such a diagnosis). Finally, the main point of that brief clinical illustration is to show that many classic ''personality cases'' can be better framed in ''relational terms''. I did not create those cases. If you'd like to read more about them, have a look at ''introjective'' vs. ''anaclitic'' depression from Sydney Blatt.
PS: A lot of what we term ''mental health problems'' are normal human responses -- so I would agree with you on that as well. What I'd like to add, however, is that this does not preclude the possibility that psychotherapy might help the person in need.