Speakers

Learning from others is an important adaptation. However, the evolution of social learning and its role in the spread of socially transmitted information are not well understood. Few models of social learning account for the fact that socially transmitted information must be reconstructed by the learner, based on the learner’s previous knowledge and cognition. To represent the reconstructive nature of social learning, I present a modeling framework that incorporates the evolution of a neural network and a simple yet biologically realistic learning mechanism. The framework encompasses various forms of individual and social learning and allows the investigation of their interplay. Individual-based simulations reveal that an effective neural network structure rapidly evolves, leading to adaptive inborn behavior in static environments, pure individual learning in highly variable environments, and a combination of individual and social learning in environments of intermediate stability. However, the evolutionary outcome depends strongly on the type of social learning (social guidance versus social instruction) and the order of individual and social learning. Moreover, the evolutionary dynamics of social learning can be surprisingly complex, with replicate simulations converging to alternative outcomes.

Building open space virtual environments for captive apes is proving to be a powerful tool for studying spatial navigation, decision making, cooperation, competition and other questions that have been difficult to study in the wild. Powerful game development platforms such as Unity provide tools to create realistic 3D environments allowing us to recreate, for example, a rainforest with a simulated two square mile area. Such environments can be populated with non-player prey characters or other actors under AI control. I will talk about lessons learned in building these environments including how to create realistic scenes, create ape friendly intuitive controls, choose the best point of view, and how to capture, reanimate and analyze the collected data. I will talk about using AI tools such as A* pathfinding to create highly reactive cooperative and competitive game agents, to analyze results and to use machine learning to compare the output of AI trained agents to apes on the same experimental tasks. I will also address future plans to use multiplayer setups for interspecies testing and comparisons. Finally, as first argued by David Washburn. computer-task testing can be an effective environmental enrichment for promoting psychological well-being and may serve to reveal previously undocumented competencies.

Why can collectives outperform individuals when solving problems? Fundamentally, they leverage greater computational resources with more sensory information, more memory, more processing capacity, more ways to act. Inspired by Marr’s levels of analysis, we show how these “hardware” advantages lead directly to well-known forms of collective intelligence – the wisdom of the crowd, collective sensing, division of labour, and cultural learning – as well as less studied behaviours such as distributed reasoning and deliberation. Through case studies, we show how collectives solve real-world problems by combining multiple forms of collective intelligence. Our framework provides a unifying perspective that brings clarity to existing findings and opens new directions for research in collective intelligence.

Cooperation experiments have long been used to explore the cognition underlying animals’ coordination towards a shared goal. While the ability to understand the need for a partner has been demonstrated in a number of species, far fewer studies have explored the communicative strategies animals use to coordinate their behavior in such tasks. To explore this in bottlenose dolphins, we ran a series of studies using a cooperative button-pressing task that required precise behavioral synchronization. In the first of these studies, members of each dyad were required to swim across a lagoon and each press their own underwater button simultaneously (within a 1 s time window), whether sent together or with a delay between partners of 1–20 s. Later, these dolphins were paired with either their previous partner, a novel partner who also knew the task, or a naïve partner, and given access to the button-pressing apparatus in the absence of any trainers. Throughout, we recorded the dolphins’ physical and vocal behavior to determine what strategies they use to coordinate with each other during this task, and to communicate their intention to enter into such a collaborative task on their own. The dolphins coordinated their behavior with extreme precision when both partners had previous knowledge of the task, but were unsuccessful when a knowledgeable dolphin was paired with a naïve dolphin. When they did succeed, results showed that dyads used vocal communication and/or physical synchrony to coordinate their behaviors during the task, but visual cues to signal their original intention to cooperate. This pattern of success, along with the differing vocal and physical signals used for different aspects of this collaborative task, sheds further light on the mechanisms underlying cooperation in the animal kingdom.

Recent advances in computer science have led to an explosion of AI systems for scene understanding and video prediction. However, despite these successes there has been relatively less work studying whether these artificial systems make predictions like natural intelligence. I will present work on the Physion dataset that was designed to compare the predictions that artificial systems and humans make about physical events. This work demonstrates that (1) AI with engineered structure tends to outperform systems that learn from data with few inductive biases, and (2) the only model that attains human levels of accuracy and generalization is a stochastic “physics engine in the head” model based on findings from cognitive science. These findings suggest a path towards more human-like AI by focusing on structured perception and physical reasoning, and less on raw statistical learning. Finally, I will discuss how these ideas can be extended to compare AI with not just humans, but with other biological intelligence as well.

Recent neurobiological experiments indicate that almost all areas of neocortex, even those traditionally labelled as sensory, are modulated by upcoming actions. Parallel evidence from neuroanatomical studies points to major outputs from neurons across cortical areas to subcortical motor centers. Active predictive coding (APC) is a new theory of predictive coding in the neocortex that combines actions and hierarchical sensory-motor dynamics. Simulation studies suggest that the same APC architecture can answer questions that at first blush seem to require very different solutions: (1) how do we recognize an object and its parts using eye movements? (2) why does perception seem stable despite eye movements? (3) how do we learn compositional representations, e.g., part-whole hierarchies, and nested reference frames? (4) how do we plan actions in a complex domain by composing sequences of sub-goals and simpler actions, and (5) how do we form episodic memories of our sensory-motor experiences and learn abstract concepts such as a family tree? Our results from the APC model illustrate the critical role played by actions, both external and internal to the brain, in mediating perception and cognition.

Cooperation is central to our ability to manage sustainability challenges. We know from decades of experiments and in situ case studies that humans can, under certain conditions, develop cooperative governance systems to sustain resources collectively (e.g., Ostrom et al., 1994). Despite this, we do not yet have a clear picture of the cultural, developmental, and evolutionary origins of these cooperative sustainability strategies. The only published comparative literature on cooperative sustainability in children and non-human apes explored dyadic cooperative patterns (Koomen & Herrmann, 2018a,b). If we are to harness insights on the origins of our cooperative sustainability skills to inform fair and effective environmental policies, we must address open questions about group-level cooperative sustainability across comparative populations. In ongoing research, we are testing the cooperative strategies of chimpanzee and bonobo groups in a novel, open diffusion common-pool resource dilemma paradigm. I will present preliminary data from four groups of non-human apes. Group- and individual-level strategies for cooperative sustainability will be discussed, along with predictions for group-level differences yet to be uncovered by future research, and what this might say about the evolution of human cooperative sustainability strategies.

Humans’ unparalleled success in adapting to diverse environments is often attributed to our capacity for cumulative cultural evolution—the ability to progressively build on prior knowledge across generations. A prominent hypothesis suggests that only humans can socially acquire behaviors that lie beyond their individual learning capacities. Here, we leverage the unique tandem running recruitment behavior of the rock ant (Temnothorax rugatulus) to investigate this hypothesis by asking the following two questions: (1) can individuals innovate solutions to previously unsolvable problems through incremental learning? And (2) can these innovations be transmitted to naïve individuals via social learning? In our experiments, individual ants were gradually trained to cross a bridge containing an aversive stimulus (Vaseline), with stimulus intensity increasing over trials. Once trained, these ants were returned to their colony and presented with a bridge presenting the strong stimulus. Our results so far show that trained ants successfully crossed the bridge and later recruited naïve nestmates to cross the bridge together using tandem running. These recruited individuals then became capable of crossing the bridge on their own. Control experiments confirmed that naïve ants could not cross the same bridge unaided, confirming that the task was beyond their independent capabilities. This study establishes a powerful new paradigm for investigating cumulative culture in non-human systems and offers insights into the general principles of collective intelligence, with broad implications for understanding the evolutionary roots of human cultural cognition.

How do captive apes’ spatial cognition, exploration to learn routes in complex environments, and spatial memory compare with wild apes’ dynamic challenges (competitors, seasonality, food availability) when navigating to foraging sites? What effect do captive environments have on the development of spatial cognitive abilities? We investigated these questions with captive bonobos simulating wild counterparts’ real-world habitats. Navigating a virtual “rainforest” presents a unique opportunity to assess cognitive abilities in exploring novel complex environments, attention to landmarks, and capacity to generate and encode novel routes. Captive bonobos explored a virtual “rainforest” comprising three regions: dense, moderate dense, and savannah-like area. Virtual fruits (mango, cherry, grape), region-specific, were associated with specific bird calls. Moving prey (e.g., red river hogs) shifted between regions. Food rewards matched selected virtual foods. Three 12-minute trials/day presented daybreak to sundown. Individualized start locations began where each bonobo trial previously ended. Bonobos were highly attentive during virtual exploration. We discuss spatial memory over time for routes based on environmental features, landmarks, and goal-oriented navigation toward preferred fruits, in order of preference. Within trials, bonobos were observed to turn towards items previously seen but not presently visible, demonstrating object permanence of visual+directional memory within a virtual space. We compare results with wild chimpanzee navigation to foraging sites by route length, efficiency, re-use of routes, and goal-directed navigation to preferred foraging sites. We highlight novel use of virtual reality to compare species and populations with different experiential and developmental trajectories.

Moore (Philosophical Quarterly, 2017) argues that Gricean communication is cognitively simpler than many suppose, such that infants and great apes may be (minimally) Gricean communicators. In a recent response, Scott-Phillips and Heintz (PNAS, 2023) reject this claim. They introduce a distinction between Gricean and ‘Ladyginian’ communication and use the latter to characterise great ape communication. Here it is argued that Scott-Phillips and Heintz offer no plausible account of the Gricean/Ladyginian distinction. A better account holds that great apes are Gricean communicators, albeit ones of limited ability. Drawing upon recent by Graham, Rossano, and Moore (Biological Reviews, 2024) I will present a new account of the pragmatic limitations of great ape communication, consistent with the hypothesis that great apes are minimally Gricean communicators.
Keywords: Gricean communication; Ladyginian communication; humans; great apes; language evolution; pragmatics

Cooperation has long been recognised as an evolutionary paradox, as helping others seems counterintuitive to the competitive nature of Darwinian selection. This apparent contradiction has led to extensive research into the conditions that enable and maintain cooperative behaviour, particularly in dyadic interactions. However, cooperation often occurs beyond pairs, with benefits that can extend to entire social groups. Despite this, little attention has been given to the question of how individual cooperation contributes to the stability of group living. This talk seeks to explore that often-overlooked perspective by considering how cooperative behaviour might help maintain the cohesion and endurance of animal groups. Based on a literature review of both empirical studies and theoretical work, we examine how cooperative behaviours, such as mutualisms, reciprocity, task specialisation and punishment, can influence group stability. These dynamics may give rise to social structures that reduce conflict and promote resilience, suggesting that individual cooperation might play a crucial but underappreciated role in group cohesion. By combining insights from cooperation theory with empirical research on social evolution, we propose that a more integrative approach may offer a more comprehensive understanding of how cooperative behaviour and group living coevolve.

Recent paradigm shifts in the study of animal consciousness, sentience, and affective states now enable the consideration of altered mental states in animals (i.e., changes in consciousness, awareness, or cognition compared to normal). ‘Flow state’ is a widely experienced altered mental state in humans characterised by intense focus, ease of action, and a loss of time and self-awareness that is linked to the development and execution of high-level technical skills. Interestingly, flow fosters a balance between task engagement, performance, and well-being, so it is an ‘optimal experience’ that is not driven solely by peak performance. To date, evidence for flow in animals has not been found, although some components of it have been observed. Cognitive and affective parallels across phylogeny mean it is logical to search for flow in great apes and cetaceans first. I will discuss flow in a wider context of testing for ‘human hallmark’ phenomena in animals, and present a framework for studying animal flow.

This talk explores two frontiers where machine learning meets natural intelligence: first, in decoding animal communication, and second, in designing more biologically grounded AI systems.

The first part presents the Earth Species Project, a non-profit initiative aiming to understand non-human communication using advanced machine learning methods. Animals communicate through complex, often multimodal signals embedded in rich social and temporal contexts. Models like NatureLM-audio, a foundation model for bioacoustics, seek to accelerate our ability to decode these signals, offering new insights into the cognitive worlds of other species and reshaping our relationship with the natural world.

This effort to understand nature naturally leads to the question: what can nature teach us about intelligence itself?

The second part of the talk introduces biologically inspired AI architecture based on oscillatory dynamics. Unlike conventional feedforward models, these networks operate in time, using frequency, phase, and amplitude to process information in a wave-based manner. They show strong performance on time-series tasks, robustness to noise, and promising hardware efficiency. This approach points toward a future of AI that is not only more energy-efficient, but also more aligned with how brains compute, namely by means of dynamics, resonance, and synchronization.

Together, these perspectives show how biological and artificial intelligence move closer together.