Table 2
Does the importance of memory vary by use case?
What makes a relationship work has been measured in some settings for decades. Setting those findings against the 29 memory mechanisms shows which ones a relationship of that kind actually depends on.
Summary
What Each Relationship Depends On
Key Points (Table 2)
-
01
Therapy uses far more types of memory than consulting or advice does.
Therapy draws on 23 of the 29 memory mechanisms; advice draws on 13, and every one of those is also used by therapy. The gap is not a product of therapy's longer list of elements: the average set of 9 of its 15 elements still draws on 19.5 mechanisms.
-
02
What therapy needs more of is mostly memory processes.
Of the 10 mechanisms therapy uses and advice does not, 7 are memory processes. Therapy draws on all 6 retrieval and surfacing mechanisms, where advice draws on 1, and it is the only one of the two that needs metacognition. How AI performs on this gap is mostly unknown rather than worse: of those 10, 4 have not yet been compared with human memory and 2 cannot be, against 3 rated Below Human and 1 Comparable to Humans.
-
03
Much of what these relationships need has never been compared with human memory.
Of the 29 mechanisms, 9 have been tested in AI and in people but never against each other, and 6 cannot be compared at all, because no fair test could put both sides on one scale. Of the 14 mechanisms either relationship needs, 4 are waiting on a comparison and 1 are beyond one. That is unmeasured against human memory, not inadequate.
Memory Mechanisms Needed, Against How AI Measures Up
Every mechanism in one bar, coloured by how AI memory compares with people. Under it, a filled cell marks each mechanism an advice and consulting relationship or a therapy relationship draws on. Select a block or a mark to open that mechanism.
Better or Same as Humans
Worse than Human Memory
Not Compared
Semantic Knowledge Confidence Calibration Instrumental Learning Working Memory Priming Retrieval Interference Partner Model Episodic Memory Standing Constraints Temporal Placement Metacognitive Control Schema Formation Classical Conditioning Procedural Memory Reality Monitoring Source Monitoring Schema Application Integrative Inference Prospective Initiation Recognition Feeling of Knowing Consolidation Context-Dependent Retrieval Autobiographical Memory Involuntary Retrieval Reconsolidation Non-Associative Learning Forgetting Motivated ForgettingAI Memory Performance
- Exceeds Human
- Comparable to Humans
- Below Human
- Not Yet Compared
- Comparison Not Possible
- Advice and Consulting draws on 13 of 29 mechanisms: 4 at or above people, 3 below people, 5 not yet compared, and 1 can't be compared.
- Therapy draws on 23 of 29 mechanisms: 5 at or above people, 6 below people, 9 not yet compared, and 3 can't be compared.
Show as a table
| Mechanism | AI memory performance | Advice and Consulting (elements of 9) | Therapy (elements of 15) |
|---|---|---|---|
| Semantic Knowledge | Exceeds Human | 3 | 1 |
| Confidence Calibration | Comparable to Humans | 3 | 2 |
| Instrumental Learning | Comparable to Humans | 1 | 3 |
| Working Memory | Comparable to Humans | 1 | 2 |
| Priming | Comparable to Humans | – | 1 |
| Retrieval Interference | Comparable to Humans | – | – |
| Partner Model | Below Human | 6 | 13 |
| Episodic Memory | Below Human | 4 | 13 |
| Standing Constraints | Below Human | 4 | 6 |
| Temporal Placement | Below Human | – | 8 |
| Metacognitive Control | Below Human | – | 3 |
| Schema Formation | Below Human | – | 2 |
| Classical Conditioning | Below Human | – | – |
| Procedural Memory | Below Human | – | – |
| Reality Monitoring | Not Yet Compared | 2 | 5 |
| Source Monitoring | Not Yet Compared | 3 | 3 |
| Schema Application | Not Yet Compared | 2 | 2 |
| Integrative Inference | Not Yet Compared | 2 | 2 |
| Prospective Initiation | Not Yet Compared | – | 3 |
| Recognition | Not Yet Compared | – | 2 |
| Feeling of Knowing | Not Yet Compared | 1 | 1 |
| Consolidation | Not Yet Compared | – | 1 |
| Context-Dependent Retrieval | Not Yet Compared | – | 1 |
| Autobiographical Memory | Comparison Not Possible | 3 | 4 |
| Involuntary Retrieval | Comparison Not Possible | – | 4 |
| Reconsolidation | Comparison Not Possible | – | 1 |
| Non-Associative Learning | Comparison Not Possible | – | – |
| Forgetting | Comparison Not Possible | – | – |
| Motivated Forgetting | Comparison Not Possible | – | – |
Use Case A
Table 2a: Advice and Consulting Relationships
Key Points (Table 2a)
-
01
How advice is delivered outranks whether it would work.
Politeness is the strongest predictor of feeling helped (β = .531), more than twice efficacy (β = .201). Consulting is more relational than its framing as expertise suggests.
-
02
Knowledge and the person split the load.
Semantic Knowledge is needed by 3 of the 9 elements and Partner Model by 3: knowledge for whether the advice is sound, the person for whether it fits.
-
03
Tailoring to the person cuts both ways.
Feasibility, which needs a model of what this person can manage, predicts coping on its own (ρ = .375) but turns negative once the other predictors are controlled (β = −.132).
-
04
Consulting uses few memory mechanisms.
24 of the 29 mechanisms are needed by no element, and likability and trustworthiness need no mechanism at all.
Memory Mechanisms Needed in Advice Interactions
A meta-analysis of advice interactions tested which features of advice, and of the person giving it, predict how well it is received. Each element is coded below against the memory mechanisms it draws on.
The two literatures measure success differently
Psychotherapy research measures whether the client got better. The advice literature measures three outcomes reported by the recipient instead: how good the advice seemed (advice quality), whether it helped the person cope with the problem (facilitation of coping), and whether they intended to act on it (implementation intention). None is an independent measure of whether the problem was solved. This table is coded against facilitation of coping, as the closest analogue to benefit, so what it shows consulting to require is derived from help as perceived rather than help as verified.
The count is needed / needed or supported, out of 9 elements. Tap an element to read how it was coded.
Memory Systems
-
Working Memory 0/1 : needed by 0, drawn on by 1 of 9 elements
-
Semantic Knowledge 3/3 : needed by 3, drawn on by 3 of 9 elements
-
Episodic Memory 0/4 : needed by 0, drawn on by 4 of 9 elements
-
Instrumental Learning 0/1 : needed by 0, drawn on by 1 of 9 elements
Not drawn on: Priming, Non-Associative Learning, Classical Conditioning, and Procedural Memory
Representational Structures
-
Schema Application 2/2 : needed by 2, drawn on by 2 of 9 elements
-
Autobiographical Memory 1/3 : needed by 1, drawn on by 3 of 9 elements
-
Partner Model 3/6 : needed by 3, drawn on by 6 of 9 elements
Not drawn on: Schema Formation
Memory Processes
-
Integrative Inference 0/2 : needed by 0, drawn on by 2 of 9 elements
-
Standing Constraints 2/4 : needed by 2, drawn on by 4 of 9 elements
Not drawn on: Consolidation, Reconsolidation, Recognition, Temporal Placement, Context-Dependent Retrieval, Involuntary Retrieval, Prospective Initiation, Forgetting, Retrieval Interference, and Motivated Forgetting
Metacognition
-
Source Monitoring 0/3 : needed by 0, drawn on by 3 of 9 elements
-
Reality Monitoring 0/2 : needed by 0, drawn on by 2 of 9 elements
-
Feeling of Knowing 0/1 : needed by 0, drawn on by 1 of 9 elements
-
Confidence Calibration 0/3 : needed by 0, drawn on by 3 of 9 elements
Not drawn on: Metacognitive Control
The Elements
A single meta-analysis throughout: Vu, Li & High (2025), Communication Research. Effects are corrected true-score correlations (ρ) with 95% confidence intervals. Multivariate coefficients are standardized betas from the model entering all predictors simultaneously (Model C).
Consistent Predictors of Coping
-
01 Politeness
Advice delivered with facework, solidarity, and tact.
- Needs
Politeness can be extended to a stranger on general norms of what threatens face and what softens it, which is schema application. What memory adds is tact fitted to this person: knowing what they are sensitive about, keeping it in mind without being reminded, and adjusting from how earlier advice landed. That it is the strongest predictor by some distance is the largest single fact in this table.
- Effect on coping
- ρ = .676 [.580, .771]
- Multivariate
- β = .531, p < .001. Strongest predictor.
- Evidence base
- 9 studies, N = 3,192
-
02 Efficacy
The advised action would solve the problem.
Knowing what would work is domain knowledge, and recognising which kind of problem this is so that knowledge applies is schema application. Integrative Inference is marked for a solution assembled from facts never stored together. Because efficacy is judged by the recipient, calibrated confidence and knowing where a claim came from bear on whether advice seems likely to work as well as whether it does.
- Effect on coping
- ρ = .536 [.355, .718]
- Multivariate
- β = .201, p = .001
- Evidence base
- 8 studies, N = 2,112
-
03 Absence of Limitations
The advised action has few or mild downsides.
Downsides are partly general, which is domain knowledge, and partly this person's: a budget, a health condition, a commitment already mentioned. Those have to shape the advice without the recipient asking about them, which is what a standing constraint is.
- Effect on coping
- ρ = .368 [.162, .574]
- Multivariate
- β = .191, p < .001
- Evidence base
- 7 studies, N = 1,835
Supported, with Instability
-
04 Expertise
The provider is knowledgeable or experienced.
- Needs
Knowledge is semantic memory. Perceived expertise also rests on knowing the edges of that knowledge and saying where claims come from, which is where the metacognitive marks sit. "Experienced" implies a history of having done the thing, which a system can only claim, not have.
- Effect on coping
- ρ = .393 [.172, .614]
- Multivariate
- β = .169, p < .001, but nonsignificant for coping in the source-primacy SEM (Model 1) while significant in the parallel additive SEM (Model 3)
- Evidence base
- 6 studies, N = 1,331
-
05 Feasibility
The recipient is able to perform the advised action.
- Supports
Whether this person can do it depends on a model of this person: their resources, their constraints, what they have already tried. This is the element whose sign reverses: positive on its own, negative once the other predictors are controlled.
- Effect on coping
- ρ = .375 [.214, .537]
- Multivariate
- β = −.132, p = .024. Sign reverses once other predictors are controlled.
- Evidence base
- 8 studies, N = 2,112
Insufficient Research for Coping
Too few studies for a stable estimate.-
06 Likability
The provider is pleasant and respectful.
- Supports
The consulting counterpart of positive regard, and like it, available to a stranger. It adds nothing once the other predictors are entered.
- Effect on coping
- ρ = .377 [.109, .644], below the authors' k ≥ 5 reporting threshold
- Multivariate
- β = −.033, p = .552, ns
- Evidence base
- 4 studies, N = 977
-
07 Confirmation
The advice fits the recipient's existing plan.
- Needs
Fitting a plan requires knowing the plan, and if it was described in an earlier conversation, recalling it. Reality Monitoring is marked for keeping the plan the recipient stated apart from the one the adviser assumed. For an AI system this is also the row where tailoring shades into agreeing, and the evidence here is too thin to say whether it helps.
- Effect on coping
- ρ = .282, CI [−2.899, 3.462] Interval exceeds [−1, 1]; uninterpretable.
- Multivariate
- Dropped from all multivariate models
- Evidence base
- 2 studies, N = 506
-
08 Trustworthiness
Confidence in the provider's motives.
Trust in motives can be extended on a first meeting. Across meetings it builds on a consistent record, which is where memory contributes, and on being clear about where advice comes from.
- Effect on coping
- ρ = .154, CI [−0.942, 1.251] Interval exceeds [−1, 1]; uninterpretable.
- Multivariate
- Dropped from all multivariate models
- Evidence base
- 2 studies, N = 506
-
09 Similarity
The provider shares the recipient's characteristics or experience.
- Supports
Similarity is a relation between two people, so it needs a model of the recipient and a past of one's own to compare against. As with self-disclosure in therapy, a system with no lived past can only offer a generated one, which is why Reality Monitoring is marked.
- Effect on coping
- ρ = .205, CI [−2.312, 2.723] Interval exceeds [−1, 1]; uninterpretable.
- Multivariate
- Dropped from all multivariate models
- Evidence base
- 2 studies, N = 506
Reading the Evidence
Help as perceived, not as verified
All three outcomes in the advice literature are reported by the recipient. Facilitation of coping is the closest to benefit, but it records that the person felt helped, not that the problem was solved.
Delivery outranks whether the advice would work
Politeness predicts coping at β = .531 with the other predictors controlled, more than twice efficacy's β = .201. Under this outcome, how advice is delivered matters more than whether it would work, which makes consulting considerably more relational than its framing as expertise suggests.
Feasibility changes sign
On its own, advice the recipient can actually carry out predicts coping (ρ = .375). With the other predictors controlled the coefficient turns negative (β = −.132). Tailoring advice to what a person can manage appears to help them intend to act while lowering how helped they feel.
Expertise depends on the model
Expertise predicts coping in the simultaneous model and in the parallel additive structural model, but not in the source-primacy structural model.
Intervals wider than the scale
Confirmation, trustworthiness, and similarity rest on two studies each. Correcting for measurement error with so few studies produces confidence intervals beyond [−1, 1]. They are reported as published and should be read as uninterpretable. Likability rests on four studies, below the authors' own threshold of five for reporting.
Use Case B
Table 2b: Therapeutic Relationships
Key Points (Table 2b)
-
01
The relationship runs on a model of the person.
Partner Model is needed by 11 of the 15 elements. The next most needed, Episodic Memory, is needed by 5.
-
02
Memory matters most where the relationship spans sessions.
Alliance, goal consensus, collaboration, and rupture repair each need a record of this person or of particular sessions. At the other end, positive regard, congruence, and emotional expression need no mechanism at all.
-
03
The most memory-hungry elements have the thinnest evidence.
Self-disclosure and immediacy each need 4 mechanisms, more than any other element, and both are rated promising but insufficient, resting on case counts rather than effect sizes.
-
04
Therapy draws on every kind of memory, metacognition included.
Therapy needs mechanisms from all four categories: 3 memory systems, 3 representational structures, 3 memory processes, and 3 metacognitive functions. It draws on 9 of the 12 memory processes and 5 of the 5 metacognitive functions. Advice, by comparison, needs 1 of the 12 memory processes and 0 of the 5 metacognitive functions.
Memory Mechanisms Needed in Therapeutic Relationships
An American Psychological Association task force reviewed the meta-analytic evidence on the therapy relationship and rated each element by the strength of its link to outcome. Each element is coded below against the memory mechanisms it draws on.
The count is needed / needed or supported, out of 15 elements. Tap an element to read how it was coded.
Memory Systems
-
Working Memory 1/2 : needed by 1, drawn on by 2 of 15 elements
-
Semantic Knowledge 0/1 : needed by 0, drawn on by 1 of 15 elements
-
Episodic Memory 5/13 : needed by 5, drawn on by 13 of 15 elements
-
Priming 0/1 : needed by 0, drawn on by 1 of 15 elements
-
Instrumental Learning 1/3 : needed by 1, drawn on by 3 of 15 elements
Not drawn on: Non-Associative Learning, Classical Conditioning, and Procedural Memory
Representational Structures
-
Schema Formation 1/2 : needed by 1, drawn on by 2 of 15 elements
-
Schema Application 0/2 : needed by 0, drawn on by 2 of 15 elements
-
Autobiographical Memory 2/4 : needed by 2, drawn on by 4 of 15 elements
-
Partner Model 11/13 : needed by 11, drawn on by 13 of 15 elements
Memory Processes
-
Consolidation 0/1 : needed by 0, drawn on by 1 of 15 elements
-
Reconsolidation 0/1 : needed by 0, drawn on by 1 of 15 elements
-
Integrative Inference 0/2 : needed by 0, drawn on by 2 of 15 elements
-
Recognition 0/2 : needed by 0, drawn on by 2 of 15 elements
-
Temporal Placement 2/8 : needed by 2, drawn on by 8 of 15 elements
-
Context-Dependent Retrieval 0/1 : needed by 0, drawn on by 1 of 15 elements
-
Involuntary Retrieval 0/4 : needed by 0, drawn on by 4 of 15 elements
-
Standing Constraints 2/6 : needed by 2, drawn on by 6 of 15 elements
-
Prospective Initiation 2/3 : needed by 2, drawn on by 3 of 15 elements
Not drawn on: Forgetting, Retrieval Interference, and Motivated Forgetting
Metacognition
-
Source Monitoring 1/3 : needed by 1, drawn on by 3 of 15 elements
-
Reality Monitoring 1/5 : needed by 1, drawn on by 5 of 15 elements
-
Feeling of Knowing 0/1 : needed by 0, drawn on by 1 of 15 elements
-
Confidence Calibration 0/2 : needed by 0, drawn on by 2 of 15 elements
-
Metacognitive Control 1/3 : needed by 1, drawn on by 3 of 15 elements
The Elements
Task force on evidence-based therapy relationships. All articles in Psychotherapy, 55(4), December 2018.
Demonstrably Effective
-
01 Alliance in Individual Psychotherapy
The collaborative bond between therapist and client, and their agreement on the goals and tasks of treatment.
- Needs
The alliance is a composite of goal consensus, collaboration, and bond, so its coding is close to the union of those rows. What it adds is continuity: the bond is with this person, and it rests on a shared history of specific sessions.
- Effect
- r = .278
- Evidence base
- 295 samples, ~30,000 patients
- Source
- Flückiger, Del Re, Wampold & Horvath (2018), pp. 316–340
-
02 Goal Consensus
Therapist and client agree on what treatment is aiming for.
Agreed goals are facts about one person that have to shape later sessions without the client restating them, which is what a standing constraint is. Reality Monitoring is marked because consensus depends on keeping what the client said apart from what the therapist inferred they meant.
- Effect
- r = .24
- Evidence base
- 54 studies, N = 7,278
- Source
- Tryon, Birch & Verkuilen (2018), pp. 372–383
-
03 Collaboration
Therapist and client work on the problem jointly, as a team.
- Supports
Joint work carries over between sessions. Work agreed in one session has to be picked up in the next, with nothing in the moment prompting it, and picked up as it was actually agreed.
- Effect
- r = .29
- Evidence base
- 53 studies, N = 5,286
- Source
- Tryon, Birch & Verkuilen (2018), pp. 372–383
-
04 Empathy
The therapist understands the client's experience from the client's point of view, and communicates that understanding.
- Needs
Within a session, empathy is largely holding and reflecting what the client is saying. Understanding it from the client's point of view needs a model of that particular person rather than of people in general. Integrative Inference is marked for connecting two things the client said separately, sometimes sessions apart.
- Effect
- r = .28
- Evidence base
- 82 samples, 6,138 clients
- Source
- Elliott, Bohart, Watson & Murphy (2018), pp. 399–410
-
05 Positive Regard and Affirmation
Warmth, acceptance, and affirmation of the client.
- Supports
The least memory-dependent element in the set. Warmth and acceptance can be offered to a stranger. Affirmation is more specific when it can point to something the client has actually done, which is where memory contributes.
- Effect
- g = .28 (.36 nested)
- Evidence base
- 64 studies, 3,528 participants
- Source
- Farber, Suzuki & Lynch (2018), pp. 411–423
-
06 Collecting and Delivering Client Feedback
Measuring the client's progress routinely and feeding it back, so a case going off track is caught.
Feedback works by comparing a client's current scores with their own earlier ones and with an expected course, so order and trajectory are the core of it. The metacognitive marks record what the measure stands in for: the therapist's own judgment of who is improving, checked against an outside source. Instrumental Learning is marked for changing course when what is being done is not working.
- Effect
- Small to moderate; deterioration reduced, reliable change nearly doubled in predicted-poor cases
- Evidence base
- 24 studies
- Source
- Lambert, Whipple & Kleinstäuber (2018), pp. 520–537
Probably Effective
-
07 The Real Relationship
The person-to-person relationship beneath the professional roles, genuine and realistic.
Defined partly as perceiving and experiencing the other realistically, which is a claim about the accuracy of the model of the person rather than warmth toward them. For a system that writes its inferences to memory, that accuracy turns on keeping what the client said apart from what was concluded about them. Genuineness asks for a self that stays consistent across the relationship.
- Effect
- r = .38, 95% CI [.30, .44], d = 0.80
- Evidence base
- 16 studies, N = 1,502
- Source
- Gelso, Kivlighan & Markin (2018), pp. 434–444
-
08 Repairing Alliance Ruptures
Noticing strains or breakdowns in the alliance and working through them together.
Repair works on a particular moment, which has to be recalled as it happened. Once repaired, what caused the rupture should shape later sessions without being raised again. Recognition and Schema Formation are marked for ruptures that recur, where the repair is to the pattern rather than the instance, and Reconsolidation because talking a rupture through revises the shared memory of it.
- Effect
- r = .29 resolved vs unresolved; r = .24 repaired vs no rupture
- Evidence base
- 11 studies, 1,314 patients
- Source
- Eubanks, Muran & Safran (2018), pp. 508–519
-
09 Congruence/Genuineness
What the therapist expresses matches what the therapist is actually experiencing.
A stance more than a memory operation. For a system in the role, the part memory touches is not overclaiming: not saying it remembers what it does not, and not stating more certainty than it has.
- Effect
- r = .23
- Evidence base
- 21 studies, 1,192 patients
- Source
- Kolden, Wang, Austin, Chang & Klein (2018), pp. 424–433
-
10 Emotional Expression
Emotion expressed in session, by the therapist and by the client.
Coded for the therapist's side, though the larger and better-supported effect is the client's own expression. Memory contributes by connecting an emotion expressed now with where it came up before. Most of what this element concerns sits under Emotional Modulation, which the Method page treats as a cross-cutting factor rather than a construct.
- Effect
- therapist d = 0.56; client d = 0.85
- Evidence base
- 66 studies (13 therapist, 43 client)
- Source
- Peluso & Freund (2018), pp. 461–472
-
11 Managing Countertransference
The therapist notices reactions rooted in the therapist's own history and keeps them from distorting the work.
Self-insight is knowing one's own history well enough to notice when it is intruding on the read of someone else. That takes Autobiographical Memory to have the history, Source Monitoring to attribute a reaction to it rather than to the client, and Metacognitive Control to act on the attribution. Priming and Involuntary Retrieval are marked as what is being managed. Conceptualizing ability, another component, draws on applying a theoretical schema and connecting material across sessions.
- Effect
- CT management → outcome r = .39
- Evidence base
- 9 studies
- Source
- Hayes, Gelso, Goldberg & Kivlighan (2018), pp. 496–507
-
12 Promoting Treatment Credibility
The client finds the treatment and its rationale believable.
Credibility shifts over treatment in response to experience, and the practice recommendation is to reassess it repeatedly rather than once at intake, which needs each reading held in order. The rationale given also has to stay consistent from session to session.
- Effect
- r = .12, d = .24
- Evidence base
- 24 samples, 1,504 patients
- Source
- Constantino, Coyne, Boswell, Iles & Vîslă (2018), pp. 486–495
-
13 Cultivating Positive Expectations
The client expects treatment to help.
- Needs
- Supports
The expectation is the client's, but cultivating it draws on specific evidence of progress, which is only available to a therapist who has kept it.
- Effect
- r = .18, d = .36
- Evidence base
- 81 samples, 12,722 patients
- Source
- Constantino, Vîslă, Coyne & Boswell (2018), pp. 473–485
Promising but Insufficient Research
Qualitative meta-analysis; no effect size.-
14 Self-Disclosure
The therapist shares personal information or experience.
The practice recommendation is an explicit store-and-reuse loop: disclose, observe this client's reaction, and decide on that basis whether to disclose again with this person. The content of a disclosure is autobiographical. A system with no lived past can only disclose something generated, which is why Reality Monitoring is marked.
- Finding
- Enhanced therapy relationship 64%, improved functioning 45%, insight 46%
- Evidence base
- 5 studies, 99 cases
- Source
- Hill, Knox & Pinto-Coelho (2018), pp. 445–460
-
15 Immediacy
Therapist and client talk about what is happening between them, in the moment.
Immediacy includes agreeing to track what happens between the two as the work goes on. In one published example the pair name a recurring interaction pattern and agree to watch for it: a pattern built across episodes, and an intention to raise it when it next occurs.
- Finding
- Enhanced therapy relationship 40%, client opened up 40%, not helpful 39%
- Evidence base
- 15 studies, 78 cases
- Source
- Hill, Knox & Pinto-Coelho (2018), pp. 445–460
Reading the Evidence
Tier is not effect size
The task force rated each element on the number of studies, consistency, independence, causal evidence, and external validity, not magnitude alone. The real relationship has the largest effect in the set (r = .38) and sits a tier below the alliance (r = .278); treatment credibility has the smallest (r = .12) and sits in the same tier as the largest. Magnitude, size of the evidence base, and tier are three different things, and none predicts the others.
The effects are not on one scale
Most elements report r, one reports g, and some report d. Self-disclosure and immediacy come from a qualitative meta-analysis and report the share of cases with a given outcome, which is not an effect size and cannot be placed on the same scale as the rest.
Emotional expression is mostly the client's
The client effect (d = 0.85, 43 studies) is larger and better supported than the therapist effect (d = 0.56, 13 studies). Clients who express emotion do better, which is not something the therapist does.
Countertransference is managed, not provided
It is the only element measured as keeping something harmful in check rather than providing something helpful. Management predicts outcome at r = .39; countertransference reactions themselves predict it at only r = −.16, an effect that was not significant after trim-and-fill correction.
Two elements report failure rates
Immediacy was rated not helpful in 39% of cases, and the authors note negative effects in up to 30% of cases across self-disclosure and immediacy. That reflects the qualitative method, which counts cases, rather than showing these two to be uniquely risky.
Why the difference matters
Since there is no one-size-fits-all “perfect” AI memory, the purpose and structure must be evaluated individually for each AI tool. A memory system that is well designed for one of these relationships may be actively wrong for the other.
A note on cultural variation
Human memory is culturally variable in narrative structure, autobiographical specificity, and the weighting of self versus relational content (Wang, 2021). AI systems trained on internet-scale corpora, which reflect predominantly Western patterns of self-narration, are unlikely to represent that variation well.
Wang (2021)
The Cultural Foundation of Human Memory
In
Annual Review of Psychology, 72(1), 151-179
Abstract
Human memory, as a product of the mind and brain, is inherently private and personal. Yet, arising from the interaction between the organism and its ecology in the course of phylogeny and ontogeny, human memory is also profoundly collective and cultural. In this review, I discuss the cultural foundation of human memory. I start by briefly reflecting on the conception of memory against a historical and cultural background. I then detail a model of a culturally saturated mnemonic system in which cultural elements constitute and condition various processes of remembering, focusing on memory representation, perceptual encoding, memory function, memory reconstruction, memory expression, and...
Find it