Mechanisms
Representational Structures · Person Models
Partner Model
Example
In a friend group, you hold that Cory likes tulips and has spin class on Wednesdays while Kaye likes reading and does yoga. The two are separate, non-interfering models, retrieved by person rather than by topic, so buying a birthday gift starts from the right person's model without any risk of handing Kaye the tulips.
Partner Model in Humans
Definition
A persisting model of a specific other person, covering what they know, what they like and what they do, held separately for each person and updated across encounters. The model is what gets consulted, rather than the underlying facts being re-derived each time. Its contents are person-specific rather than generic: partners agree about which of them holds which kind of knowledge more often than paired strangers do, though strangers paired only by sex and age still do better than chance, so a partner model layers individual knowledge on top of stereotype (Wegner, Erber, & Raymond, 1991). The same layering shows in preferences, where partners predict each other's likes above chance and are most accurate for strong, stable and widely shared preferences (Scheibehenne, Mata, & Todd, 2011).
Scope: Isolates the individual-side component of Wegner's transactive memory, that is, one partner's directory of the other, rather than the pair-level system the directory participates in.
Neural Substrate
Person knowledge is associated with anterior temporal cortex, particularly ventral temporopolar regions implicated in representing social knowledge and linking faces to what is known about that person (Ranganath & Ritchey, 2012).
The human evidence for that attribution is convergent and lesion-led. Damage to the anterior temporal lobes produces a form of prosopagnosia in which the perception of faces is spared but the recognition of people is not, and the deficit crosses modalities to voice and gait, which places the loss at the level of stored person knowledge rather than face processing (Olson, Plotzker, & Ezzyat, 2007). Surgical resection separates the two hemispheres' contributions: left anterior temporal resection impairs naming famous faces and learning new face-name pairs, while right resection impairs recognizing famous and personally familiar faces at all. Progressive right anterior temporal atrophy produces the same picture as a slow decline, with loss of person identification often the earliest presenting complaint.
Two qualifications belong with this. Olson and colleagues argue that the region is necessary for person recognition but not specialized for it, since the most severely affected patients also fail on other unique items such as landmarks, and they place person knowledge inside a broader function of binding highly processed perception to emotional response. The region is therefore where person knowledge appears to be stored and addressed, not a dedicated partner-model store.
AI at a Glance
People hold a separate, persisting model of each individual they know well, consulted rather than rebuilt, and accurate enough to beat both chance and stereotype while being far less accurate than its holder believes (Wegner et al., 1991; Scheibehenne et al., 2011). AI systems build a comparable model of the one user in front of them and get roughly half of it right, but when several people must be held apart the binding between a person and what is true of them is broken while the memory is being written, and the closest named mechanisms, the memory layer and the context window, are where it breaks (Jiang et al., 2026; Yang et al., 2026).
- Memory layer: A store outside the model that extracts facts from a conversation, keeps them between sessions, and returns a few of them at question time (Yang et al., 2026).
- Context window: The span of text a model can hold at once, so anything longer must be summarized, stored elsewhere, or dropped (Jiang et al., 2026).
Partner Model in AI
How Far It Has Been Measured
Measurement in AI: Indirectly Measured
AI Memory Performance: Below Human approximate
Where It Lives in the System
What Has Been Measured
- No published study gives an AI system the human partner-switch test across separate encounters, or compares with people an AI that keeps separate, lasting models of several individuals; AI memory benchmarks remain almost entirely one-user-at-a-time, so the evidence below covers pieces of the ability in different places: adapting to one partner within a conversation the model can still see, saved memory of one user, and memory for several people without any human baseline (Yang et al., 2026).
- Within one conversation, five multimodal AI models (models that read images as well as text) replayed 54 human-human tangram-style games from 108 online participants, with the whole game history kept in their context window, meaning what they can see in the current conversation; without explicit instruction GPT-4, Gemini 1.0 Pro Vision and Claude 3 Opus (2023-2024) made their descriptions longer rather than shorter across rounds, IDEFICS and LLaVA repeated themselves, and as listeners GPT-4 reached 88.9 to 92.5% against about 99.5% for humans while IDEFICS fell to chance (Hua & Artzi, 2024).
- In a live experiment with 103 adults (32 human pairs, 39 human-AI pairs, plus 18 AI-AI pairs) matching objects over four rounds, human pairs rose from 80% to over 90% accuracy while using fewer words each round (slope -74.9 words); GPT-5.2 (2025), given the full dialogue history, fell in accuracy from 90% when paired with itself (slope -5.3 points per round) and from 85% when directing a human (slope -15.3), and its word counts did not fall (Zeng et al., 2026).
- The only study that isolates partner-specificity (a 2026 preprint) compared real pairs with pseudo-pairs stitched together from different pairs' games: in 42 human pairs from a published tangram corpus, shared labels were far more common in real than pseudo pairs (odds ratio 6.85) and descriptions shrank to about 30% of their first length, whereas for GPT-5 (2025) agents, which could see the full history, label sharing was near ceiling with or without a real partner (odds ratio 1.26, 95% CI 0.99 to 1.61) because their long descriptions reused the same stock words, and descriptions stayed near 90% of their first length (Wang et al., 2026).
- Why the comparison is approximate but still clear: these games test building a partner-specific way of talking within one encounter rather than a lasting model reused across encounters, and the AI could still read the whole history while people relied on memory; because that advantage favoured the AI and it still fell short of people on the signatures of a partner model, the result counts as Below Human (Wang et al., 2026).
- Spread across systems: in a 2026 preprint GPT-5 pairs reached 99% accuracy against 93% for human pairs, but their descriptions barely shortened (58 to 56 words, against 20 to 6 for humans) and GPT-5 paired with people stayed below human pairs in accuracy and word overlap even when prompted to sound humanlike; explicit instructions to shorten (Hua & Artzi, 2024) or extra training (Vaduguru et al., 2025, up to 41% shorter messages) narrow the length gap, but prompted models kept changing their words, and the higher AI-AI accuracy seen with the history visible is not counted (Jones et al., 2026).
- Across sessions, saved memory of a single user has been tested only without human baselines: answering questions about a user's earlier chats, ChatGPT's memory feature scored 57.7% and Coze 33.0% (both on GPT-4o, 2024) against 91.8% when the same model read the chat history directly; related benchmarks of a user's changing preferences (PersonaMem, PrefEval) likewise report no human scores, so none of this can be placed on a human scale (Wu et al., 2025).
- Keeping separate records for several people is also unmatched by human data: in a 2026 preprint on group chats, eight memory methods averaged 20.6% to 46.0%, a simple keyword search scored 43.2%, and none exceeded 27.1% on keeping each person's changing information up to date; FANToM (Kim et al., 2023), where GPT-4 trailed humans on who knows what within one conversation, tests theory of mind rather than a lasting person model and is not used for this rating (Yang et al., 2026).
Where a Mechanism Can Live
04 Product
The whole configuration, and the only thing a benchmark observes directly.
01
Weights
Across deployments
02
Context Window
Within a session
03
Memory Layer
Across sessions
Which of the three a result belongs to is inferred from it, not seen in it.
Read more
Anterior temporal lobe
Lobe in the temporal lobe
The forward portion of the temporal lobe, converging on the temporal pole and heavily connected to orbitofrontal and limbic cortex.
The anterior temporal lobe covers the front of the superior, middle and inferior temporal gyri together with the temporal pole and the anterior fusiform gyrus. It sits at the end of the ventral visual and auditory streams and connects strongly to orbitofrontal cortex, amygdala and entorhinal cortex through the uncinate fasciculus. Its degeneration in semantic dementia produces progressive loss of word and object meaning across modalities, which motivated the "semantic hub" account. Whether the region is genuinely modality-general or an assembly of separable category-specific areas remains contested, partly because MRI signal here is badly distorted.
Temporal pole
Cortex in the anterior temporal lobe
The rounded front tip of the temporal lobe, sitting in the middle cranial fossa in front of the amygdala.
The temporal pole is the anterior end of the temporal lobe, where the superior, middle and inferior temporal gyri converge; it corresponds approximately to Brodmann area 38 and sits directly above the sphenoid ridge in the middle cranial fossa. It connects to orbitofrontal cortex through the uncinate fasciculus and to the amygdala and entorhinal cortex nearby. It atrophies early in semantic dementia, and is a frequent site of contusion in head injury and of surgical resection in temporal lobe epilepsy. Functional imaging of it is difficult because of susceptibility artefact from the adjacent air spaces.
Hua & Artzi (2024)
Talk Less, Interact Better: Evaluating In-context Conversational Adaptation in Multimodal LLMs
In
arXiv
Abstract
Humans spontaneously use increasingly efficient language as interactions progress, by adapting and forming ad-hoc conventions. This phenomenon has been studied extensively using reference games, showing properties of human language that go beyond relaying intents. It remains unexplored whether multimodal large language models (MLLMs) similarly increase communication efficiency during interactions, and what mechanisms they may adopt for this purpose. We introduce ICCA, an automated framework to evaluate such conversational adaptation as an in-context behavior in MLLMs. We evaluate several state-of-the-art MLLMs, and observe that while they may understand the increasingly efficient language...
Find it
Jiang et al. (2026)
PersonaMem-v3: Toward Omni-Platform Personal Intelligence for Holistic User Understanding, Recommendation, and Agentic Tasks
In
arXiv
Abstract
Personal intelligence is becoming a central frontier for user-facing AI agents. To be helpful in everyday life, agents must understand users across the digital contexts where their preferences, intents, habits, social relationships, and needs unfold over time. Today's systems can personalize within individual apps or tasks, but personal intelligence as a whole remains under-measured: how agents build cross-context user understanding, support steerable recommendation systems, act proactively across platforms, and avoid over-personalization. We introduce PersonaMem-v3, a real-world-grounded benchmark and evaluation harness for omni-platform personal intelligence. PersonaMem-v3 is seeded...
Find it
Jones et al. (2026)
LLMs and people both learn to form conventions -- just not with each other
In
arXiv
Abstract
Humans align to one another in conversation -- adopting shared conventions that ease communication. We test whether LLMs form the same kinds of conventions in a multimodal communication game. Both humans and LLMs display evidence of convention-formation (increasing the accuracy and consistency of their turns while decreasing their length) when communicating in same-type dyads (humans with humans, AI with AI). However, heterogenous human-AI pairs fail -- suggesting differences in communicative tendencies. In Experiment 2, we ask whether LLMs can be induced to behave more like human conversants, by prompting them to produce superficially humanlike behavior. While the length of their...
Find it
Olson et al. (2007)
The Enigmatic temporal pole: a review of findings on social and emotional processing
In
Brain, 130(7), 1718-1731
Abstract
The function of the anterior-most portion of the temporal lobes, the temporal pole, is not well understood. Anatomists have long considered it part of an extended limbic system based on its location posterior to the orbital frontal cortex and lateral to the amygdala, along with its tight connectivity to limbic and paralimbic regions. Here we review the literature in both non-human primates and humans to assess the temporal pole's putative role in social and emotional processing. Reviewed findings indicate that it has some role in both social and emotional processes, including face recognition and theory of mind, that goes beyond semantic memory. We propose that the temporal pole binds...
Find it
Ranganath & Ritchey (2012)
Two cortical systems for memory-guided behaviour
In
Nature Reviews Neuroscience, 13(10), 713-726
Abstract
Although the perirhinal cortex (PRC), parahippocampal cortex (PHC) and retrosplenial cortex (RSC) have an essential role in memory, the precise functions of these areas are poorly understood. Here, we review the anatomical and functional characteristics of these areas based on studies in humans, monkeys and rats. Our Review suggests that the PRC and PHC–RSC are core components of two separate large-scale cortical networks that are dissociable by neuroanatomy, susceptibility to disease and function. These networks not only support different types of memory but also appear to support different aspects of cognition.
Find it
Scheibehenne et al. (2011)
Older but not wiser—Predicting a partner's preferences gets worse with age
In
Journal of Consumer Psychology, 21(2), 184-191
Abstract
To test the influence of relationship length on ability to predict a partner's preferences, 58 younger (M = 24.1 years) and 20 older (M = 68.7 years) couples made predictions in three domains that varied in daily importance. While prediction accuracy was generally better than chance, longer relationship length correlated with lower prediction accuracy and greater overconfidence. The difference in accuracy between older and younger couples increased for strong preferences and when controlling for preference reliability over time. Independent of relationship length, prediction accuracy was higher for important domains, for strong, reliable, and stereotypical preferences, and when couples...
Find it
Wegner et al. (1991)
Transactive memory in close relationships.
In
Journal of Personality and Social Psychology, 61(6), 923-929
Abstract
Studied memory performance of 118 individuals who had been in close dating relationships for at least 3 mo. For a memory task ostensibly to be performed by pairs, some Ss were paired with their partners and some were paired with an opposite-sex partner from another couple. For some pairs, a memory structure was assigned (e.g., 1 partner should remember food items, another should remember history items, etc.), whereas for others no structure was mentioned. Pairs studied together without communication, and recall was tested in individuals. Memory performance of the natural pairs was better than that of impromptu pairs without assigned structure, whereas the performance of natural pairs was...
Find it
Wu et al. (2024)
LongMemEval: Benchmarking Chat Assistants on Long-Term Interactive Memory
In
arXiv
Abstract
Recent large language model (LLM)-driven chat assistant systems have integrated memory components to track user-assistant chat histories, enabling more accurate and personalized responses. However, their long-term memory capabilities in sustained interactions remain underexplored. We introduce LongMemEval, a comprehensive benchmark designed to evaluate five core long-term memory abilities of chat assistants: information extraction, multi-session reasoning, temporal reasoning, knowledge updates, and abstention. With 500 meticulously curated questions embedded within freely scalable user-assistant chat histories, LongMemEval presents a significant challenge to existing long-term memory...
Find it
Yang et al. (2026)
GroupMemBench: Benchmarking LLM Agent Memory in Multi-Party Conversations
In
arXiv
Abstract
Large Language Model (LLM) agents increasingly serve as personal assistants and workplace collaborators, where their utility depends on memory systems that extract, retrieve, and apply information across long-running conversations. However, both existing memory systems and benchmarks are built around the dyadic, single-user setup, even though real deployments routinely span groups and channels with multiple users interacting with the agent and with each other. This mismatch leaves three properties of group memory unmeasured: (i) group dynamics that go beyond concatenated one-on-one chats, (ii) speaker-grounded belief tracking, where the per-user memory modeling is needed, and (iii)...
Find it
Zeng et al. (2026)
LVLMs and Humans Ground Differently in Referential Communication
In
arXiv
Abstract
For generative AI agents to partner effectively with human users, the ability to accurately predict human intent is critical. But this ability to collaborate remains limited by a critical deficit: an inability to model common ground. We present a referential communication experiment with a factorial design involving director-matcher pairs (human-human, human-AI, AI-human, and AI-AI) that interact with multiple turns in repeated rounds to match pictures of objects not associated with any obvious lexicalized labels. We show that LVLMs cannot interactively generate and resolve referring expressions in a way that enables smooth communication, a crucial skill that underlies human language use....
Find it