**Abstract:**

Archeological findings indicate that around 400KYA-200KYA in the eastern levant, ancient humans followed a new type of habitation style, basecamp cave dwelling. This entailed multiple small groups living in the same environment for long stretches of time, foraging for resources in different parts of the environment during the day and meeting back at the cave for shelter. We suggest that this habitation pattern could support the gradual development of complex social learning skills. Initially, basecamp habitation allowed tracking different group members’ expertise in resource attainment, and the development of prestige bias. In addition, based on their knowledge of the common environment, individuals could infer where other individuals foraged that day, and gain counterfactual information about the richness of the environment. Intentional signaling using distinct local markers could increase information transmission fidelity. Using simulations of computational models, we show that material benefits gradually increase with the level of sophistication of social learning strategies. This increase was more distinct in larger environments with multiple foraging opportunities. Our work highlights the way habitation patterns and physical settings which humans occupied could support the development of complex social learning skills, and the way environmental characteristics could serve as a common framework for communication.

**Speaker's social media:**

@uhertz

RECORDING - Ancient humans’ basecamp habitation pattern and the transition from experience-based to information-based social learning

**Abstract:**

Mathematical and computational approaches have made important contributions to cultural evolutionary theory by highlighting the effect of factors such as population size, social network structure, and transmission fidelity on cultural evolution. However, some cultural phenomena have so far been more difficult to capture using agent-based and formal models. In particular, determining how personality, background, and cultural context shape how individuals select, transform and transmit social information has remained out of the reach of traditional modelling approaches, and has thus mainly been studied empirically. We here propose that leveraging the capacity of Large Language Models (LLMs) to “role-play” and impersonate specific profiles may be fruitful to address this gap. Indeed, these generative models have been found to exhibit quite satisfying accuracy and relatively good consistency when used to simulate different characters. We present a framework for simulating and analyzing cultural evolution in populations of role-playing LLMs, along with results from use-cases that illustrate the kind of hypotheses that can be generated using this approach. The software we developed for conducting these simulations is open-source and features an intuitive, code-free user-interface, which we hope will help to build bridges between the fields of cultural evolution and generative artificial intelligence.

**Speaker's social media:**

@Jeremy__Perez

LIVE - Extending cultural evolutionary theory to machines

**Abstract:**

Previous studies suggest that social learning strategies can be adaptive, promoting collective intelligence because it filters out seemingly unfavourable behavioural options from individual exploration. Such adaptive filtering functions well, as long as individuals share the same fitness landscape or are tasked with the same decision problems. However, this is rarely the case in real life, with human populations having diverse needs, circumstances, and goals. In such heterogeneous populations, exactly copying others may not guarantee acquiring adaptive behaviours, and socially-guided exploration may backfire. To address this issue, we added social correlations to a spatially-correlated multi-armed bandit task, which lets us operationalize differences in taste while maintaining a common ground truth that applies to all individuals. We introduce a novel model, Social Generalization (SG), which generalises social observations as noisier information than privately obtained payoff information. Evolutionary simulations show that SG out-competes existing simpler social learning models including both the decision biassing and value shaping. The results from online experiments suggest that humans are best fit by the SG model, and benefited from socially-guided exploration despite the individual differences. Our study shows that social learning is adaptive in a wider range of circumstances than previously assumed, even in a heterogeneous population.

**Speaker's social media:**

Twitter: @WataruToyokawa BlueSky @watarutoyokawa.bsky.social

LIVE - Social learning generates collective intelligence even among heterogeneous individuals

**Abstract:**

The whisper game is often used for studying culture transmission and how variation can arise from transmission. However, it has two drawbacks from the perspective of its computer mediation: it is analog, i.e. the message is the culture to transmit, and its variation comes from unfaithful transmission. First, computer-based transmission may be faithful. Second, culture is transmitted 'semantically': the message is first integrated within preexisting culture from which a message is generated for transmission.

We aim at showing that these integration/generation operations produce sufficient variation.

For that purpose, we introduce a game in which agents induce a card classification from the dispatch of randomly selected cards. They then generate from their classification the next dispatch of a random card sample. The two operations are implemented deterministically, but depending on the sample, the generalisation may be incomplete. The agents may know or suspect that and complete it. This brings variation back. Depending on the size of the sample with respect to that of the classification, more variation should be observed, inducing a degradation or preservation of faithfulness.

This setting can be played by human beings or through computer simulations and they can be compared.


**Speaker's social media:**

https://moex.inria.fr

Culture transmission through generalisation and example generation should produce variation

**Abstract:**

Existing models of cumulative cultural evolution suggest that cultural complexity is facilitated by high-fidelity cultural transmission and demographic factors, such as population size. However, models and theory tend to neglect the role of asocial learning in the evolution of cumulative culture, unrealistically treating brains as passive receptacles of socially learned information that increase linearly in size and storage capacity as culture accumulates. Here we address this omission by combining artificial neural network simulations and agent-based models to provide a bottom-up mechanistic understanding of the computational capabilities that neural architectures acquire as they increase in size and alter in morphology. Our more realistic model of brain-culture coevolution generates a biologically feasible nonlinear relationship between brain size, neocortical modularity, and cultural cognition, resembling observed neurological differences between human and nonhuman primates. At optimal network modularity and over a critical size threshold, associative learning and inference give way to new capacities for recursive analogy parsing, generation of arbitrary symbolic associations, and an ability to recall and recombine ideas from uncorrelated domains of cultural knowledge without external cues. Including this within existing models of cumulative culture demonstrates that increasing cultural complexity depends upon recursive imagination, transmission fidelity and demography in equal measure.

**Speaker's social media:**

Twitter: @AseSergiou

RECORDING - The Role of brain size, neural architecture and recursive imagination in the evolution of cumulative culture

**Abstract:**

Migration has played a pivotal role in shaping both cultural and genetic diversity throughout human history. Results from population genetics show that even low rates of migration between populations lead to the homogenization of their genetic compositions. On the cultural side, we know that conformity helps maintain differences between groups. However, equivalent insights into the effect of migration on cultural composition remain limited. To address this gap, I have developed a simulation model that explores the joint evolution of genetic and cultural traits, investigating how varying strengths of conformity and migration rates influence differentiation both within and between populations

**Speaker's social media:**

Twitter: @camilla_cenni

RECORDING - Cultural drivers of stone-tool use in Balinese long-tailed macaques

Strategic social deduction games serve as valuable testbeds for evaluating the understanding and inference skills of language models, offering crucial insights into social science, artificial intelligence, and strategic gaming. This paper focuses on creating proxies of human behavior in simulated environments, with “Among Us” utilized as a tool for studying simulated human behavior. The study introduces a text-based game environment, named AmongAgents, that mirrors the dynamics of “Among Us”. Players act as crew members aboard a spaceship, tasked with identifying impostors who are sabotaging the ship and eliminating the crew. Within this environment, the behavior of simulated language agents is analyzed. The experiments involve diverse game sequences featuring different configurations of Crewmates and Impostor personality archetypes. Our work demonstrates that state-of-the-art large language models (LLMs) can effectively grasp the game rules and make decisions based on the current context. This work aims to promote further exploration of LLMs in goal-oriented games with incomplete information and complex action spaces, as these settings offer valuable opportunities to assess language model performance in socially driven scenarios.

AmongAgents: Evaluating Large Language Models in the Interactive Text-Based Social Deduction Game

Developing a consistent and reliable AI game master for text-based games is a challenging task due to the limitations of large language models (LLMs) and the complexity of the game master's role. This paper presents a novel approach to enhance AI game masters by leveraging function calling in the context of the table-top role-playing game "Jim Henson's Labyrinth: The Adventure Game." Our methodology involves integrating game-specific controls through functions, which we show improves the narrative quality and state update consistency of the AI game master. The experimental results, based on human evaluations and unit tests, demonstrate the effectiveness of our approach in enhancing gameplay experience and maintaining coherence with the game state. This work contributes to the advancement of game AI and interactive storytelling, offering insights into the design of more engaging and consistent AI-driven game masters.

You Have Thirteen Hours in Which to Solve the Labyrinth: Enhancing AI Game Masters with Function Calling

At the beginning of 2022, a simplistic word-guessing game took the world by storm and was further adapted to many languages beyond the original English version. In this paper, we compare how the strategies of daily word-guessing game players have evolved during a period of over two years. A survey gathered from 25% of frequent players reveals their strategies and motivations for continuing the daily journey. We also explore the capability of several popular open-access large language model systems and open-source models at comprehending and playing the game in two different languages. Results highlight the struggles of certain models to maintain correct guess length, and generate repetitions, as well as hallucinations of non-existent words and inflections.

Strategic Insights in Human and Large Language Model Tactics at Word Guessing Games

Text-based adventure game generation has attracted considerable interest since the advancements in large language models. However, even state-of-the-art (SOTA) large language models (LLMs) face challenges in generating semantically and logically coherent content, and often fail to align with human intentions. In our work, we propose a approach for enhancing content coherence, additionally allowing humans in the loop to better match their intentions. Specifically, we generate game components, including locations, characters, items, actions, and blocks, and integrate these components with semantic and logical constraints. Meanwhile, human game creators can intervene and modify the components to suit their preferences. Evaluation results demonstrate that our procedural world-generation approach both enriches the content and satisfies semantic and logical requirements compared to the SOTA LLMs. Moreover, the human-in-the-loop feature gives game creators more control over generated content.

WorldWeaver: Procedural World Generation for Text Adventure Games using Large Language Models

State-of-the-art Large Multi-Modal Models (LMMs) have demonstrated exceptional capabilities in vision-language tasks. Despite their advanced functionalities, the performances of LMMs are still limited in challenging scenarios that require complex reasoning with multiple levels of visual information. Existing prompting techniques for LMMs focus on either improving textual reasoning or leveraging tools for image preprocessing, lacking a simple and general visual prompting scheme to promote vision-language coordination in LMMs. In this work, we propose Scaffold prompting that scaffolds coordinates to promote vision-language coordination. Specifically, Scaffold overlays a dot matrix within the image as visual information anchors and leverages multi-dimensional coordinates as textual positional references. Extensive experiments on a wide range of challenging vision-language tasks demonstrate the superiority of Scaffold over GPT-4V with the textual CoT prompting.

Scaffolding Coordinates to Promote Vision-Language Coordination in Large Multi-Modal Models

The popularity of role-playing video games (RPGs) continues to increase over time. These types of games often afford players multiple options for how they wish to accomplish any given task, which adds to their appeal. Most commonly, RPGs employ a finite state dialogue management system taking the form of a graph structure, where the set of possible utterances at a given node depend on the current game state. From the player's perspective, these appear as a list of pre-written dialogue options that can be chosen to advance the conversation, thus limiting a player's role-playing to the available options. In this paper, we investigate a hybrid dialogue management approach, which allows players more freedom while staying true to the dialogue graph created by the game's designers. Instead of seeing a list of options, players are provided an open-ended text field. A virtual game master (GM) is then tasked with either mapping the player input to one of the dialogue options, or providing in-character feedback to guide players into typing an utterance which the virtual GM can successfully map to an dialogue option. As a preliminary study, we investigate simulated players with a suite of virtual GMs, including a random baseline within the video game Disco Elysium: The Final Cut.

Underline digital video library

RECORDING - Ancient humans’ basecamp habitation pattern and the transition from experience-based to information-based social learning

LIVE - Extending cultural evolutionary theory to machines

LIVE - Social learning generates collective intelligence even among heterogeneous individuals

Culture transmission through generalisation and example generation should produce variation

RECORDING - The Role of brain size, neural architecture and recursive imagination in the evolution of cumulative culture

RECORDING - Cultural drivers of stone-tool use in Balinese long-tailed macaques

AmongAgents: Evaluating Large Language Models in the Interactive Text-Based Social Deduction Game

You Have Thirteen Hours in Which to Solve the Labyrinth: Enhancing AI Game Masters with Function Calling

Strategic Insights in Human and Large Language Model Tactics at Word Guessing Games

WorldWeaver: Procedural World Generation for Text Adventure Games using Large Language Models

Scaffolding Coordinates to Promote Vision-Language Coordination in Large Multi-Modal Models

Hybrid Dialogue Management for Task-Oriented Dialogue in Video Games

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES