Large language models demonstrate advanced problem-solving capabilities by incorporating reasoning techniques such as chain of thought and reflection. However, how these reasoning capabilities extend to social intelligence remains unclear. In this study, we investigate this question using economic games that model social dilemmas, where social intelligence plays a crucial role. First, we examine the effects of chain-of-thought and reflection techniques in a public goods game. We then extend our analysis to six economic games on cooperation and punishment, comparing off-the-shelf non-reasoning and reasoning models. We find that reasoning models significantly reduce cooperation and norm enforcement, prioritizing individual rationality. Consequently, groups with more reasoning models exhibit less cooperation and lower gains through repeated interactions. These behaviors parallel human tendencies of "spontaneous giving and calculated greed." Our results suggest the need for AI architectures that incorporate social intelligence alongside reasoning capabilities to ensure that AI supports, rather than disrupts, human cooperative intuition. Recent innovations in reasoning techniques, such as chain of thought [1] and reflection [2], are advancing the intellectual capabilities of large language models (LLMs) to the next level. Models such as OpenAI o1 leverage these techniques to solve complex problems, generate coherent arguments, and improve decision-making in multi-step reasoning scenarios [3-5]. Indeed, these reasoning models have demonstrated excellence in mathematical proofs, logical deduction, and strategic planning [6, 7]. The necessity of social intelligence is highlighted in social dilemmas, where individual rationality leads to collective irrationality [12].

In reinforcement learning (RL), world models serve as internal simulators, enabling agents to predict environment dynamics and future outcomes in order to make informed decisions. While previous approaches leveraging discrete latent spaces, such as DreamerV3, have demonstrated strong performance in discrete action settings and visual control tasks, their comparative performance in state-based continuous control remains underexplored. In contrast, methods with continuous latent spaces, such as TD-MPC2, have shown notable success in state-based continuous control benchmarks. In this paper, we demonstrate that modeling discrete latent states has benefits over continuous latent states and that discrete codebook encodings are more effective representations for continuous control, compared to alternative encodings, such as one-hot and label-based encodings. Based on these insights, we introduce DCWM: Discrete Codebook World Model, a self-supervised world model with a discrete and stochastic latent space, where latent states are codes from a codebook. We combine DCWM with decision-time planning to get our model-based RL algorithm, named DC-MPC: Discrete Codebook Model Predictive Control, which performs competitively against recent state-of-the-art algorithms, including TD-MPC2 and DreamerV3, on continuous control benchmarks. See our project website www.aidanscannell.com/dcmpc.

Arbitrarily Applicable Relational Responding (AARR) is a cornerstone of human language and reasoning, referring to the learned ability to relate symbols in flexible, context-dependent ways. In this paper, we present a novel theoretical approach for modeling AARR within an artificial intelligence framework using the Non-Axiomatic Reasoning System (NARS). NARS is an adaptive reasoning system designed for learning under uncertainty. By integrating principles from Relational Frame Theory - the behavioral psychology account of AARR - with the reasoning mechanisms of NARS, we conceptually demonstrate how key properties of AARR (mutual entailment, combinatorial entailment, and transformation of stimulus functions) can emerge from the inference rules and memory structures of NARS. Two theoretical experiments illustrate this approach: one modeling stimulus equivalence and transfer of function, and another modeling complex relational networks involving opposition frames. In both cases, the system logically demonstrates the derivation of untrained relations and context-sensitive transformations of stimulus significance, mirroring established human cognitive phenomena. These results suggest that AARR - long considered uniquely human - can be conceptually captured by suitably designed AI systems, highlighting the value of integrating behavioral science insights into artificial general intelligence (AGI) research.

Recent advances in AI reasoning models provide unprecedented transparency into their decision-making processes, transforming them from traditional black-box systems into models that articulate step-by-step chains of thought rather than producing opaque outputs. This shift has the potential to improve software quality, explainability, and trust in AI-augmented development. However, software engineers rarely have the time or cognitive bandwidth to analyze, verify, and interpret every AI-generated thought in detail. Without an effective interface, this transparency could become a burden rather than a benefit. In this paper, we propose a vision for structuring the interaction between AI reasoning models and software engineers to maximize trust, efficiency, and decision-making power. We argue that simply exposing AI's reasoning is not enough -- software engineers need tools and frameworks that selectively highlight critical insights, filter out noise, and facilitate rapid validation of key assumptions. To illustrate this challenge, we present motivating examples in which AI reasoning models state their assumptions when deciding which external library to use and produce divergent reasoning paths and recommendations about security vulnerabilities, highlighting the need for an interface that prioritizes actionable insights while managing uncertainty and resolving conflicts. We then outline a research roadmap for integrating automated summarization, assumption validation, and multi-model conflict resolution into software engineering workflows. Achieving this vision will unlock the full potential of AI reasoning models to enable software engineers to make faster, more informed decisions without being overwhelmed by unnecessary detail.

Temporal networks have gained significant prominence in the past decade for modelling dynamic interactions within complex systems. A key challenge in this domain is Temporal Link Prediction (TLP), which aims to forecast future connections by analysing historical network structures across various applications including social network analysis. While existing surveys have addressed specific aspects of TLP, they typically lack a comprehensive framework that distinguishes between representation and inference methods. This survey bridges this gap by introducing a novel taxonomy that explicitly examines representation and inference from existing methods, providing a novel classification of approaches for TLP. We analyse how different representation techniques capture temporal and structural dynamics, examining their compatibility with various inference methods for both transductive and inductive prediction tasks. Our taxonomy not only clarifies the methodological landscape but also reveals promising unexplored combinations of existing techniques. This taxonomy provides a systematic foundation for emerging challenges in TLP, including model explainability and scalable architectures for complex temporal networks.

The search for effective collaboration between humans and computer systems is one of the biggest challenges in Artificial Intelligence. One of the more effective mechanisms that humans use to coordinate with one another is theory of mind (ToM). ToM can be described as the ability to `take someone else's perspective and make estimations of their beliefs, desires and intentions, in order to make sense of their behaviour and attitudes towards the world'. If leveraged properly, this skill can be very useful in Human-AI collaboration. This introduces the question how we implement ToM when building an AI system. Humans and AI Systems work quite differently, and ToM is a multifaceted concept, each facet rooted in different research traditions across the cognitive and developmental sciences. We observe that researchers from artificial intelligence and the computing sciences, ourselves included, often have difficulties finding their way in the ToM literature. In this paper, we identify four common misconceptions around ToM that we believe should be taken into account when developing an AI system. We have hyperbolised these misconceptions for the sake of the argument, but add nuance in their discussion. The misconceptions we discuss are: (1) "Humans Use a ToM Module, So AI Systems Should As Well". (2) "Every Social Interaction Requires (Advanced) ToM". (3) "All ToM is the Same". (4) "Current Systems Already Have ToM". After discussing the misconception, we end each section by providing tentative guidelines on how the misconception can be overcome.

Inspired by human cognition, metacognition has gained significant attention for its potential to enhance autonomy, adaptability, and robust learning in artificial agents. Yet research on Computational Metacognitive Architectures (CMAs) remains fragmented: diverse theories, terminologies, and design choices have led to disjointed developments and limited comparability across systems. Existing overviews and surveys often remain at a broad, conceptual level, making it difficult to synthesize deeper insights into the underlying algorithms and representations, and their respective success. We address this gap by performing an explorative systematic review of how CMAs model, store, remember and process their metacognitive experiences, one of Flavell's (1979) three foundational components of metacognition. Following this organizing principle, we identify 35 CMAs that feature episodic introspective data ranging from symbolic event traces to sub-symbolic arousal metrics. We consider different aspects - ranging from the underlying psychological theories to the content and structure of collected data, to the algorithms used and evaluation results - and derive a unifying perspective that allows us to compare in depth how different Computational Metacognitive Architectures (CMAs) leverage metacognitive experiences for tasks such as error diagnosis, self-repair, and goal-driven learning. Our findings highlight both the promise of metacognitive experiences - in boosting adaptability, explainability, and overall system performance - and the persistent lack of shared standards or evaluation benchmarks.

The endowment of AI with reasoning capabilities and some degree of agency is widely viewed as a path toward more capable and generalizable systems. Our position is that the current development of agentic AI requires a more holistic, systems-theoretic perspective in order to fully understand their capabilities and mitigate any emergent risks. The primary motivation for our position is that AI development is currently overly focused on individual model capabilities, often ignoring broader emergent behavior, leading to a significant underestimation in the true capabilities and associated risks of agentic AI. We describe some fundamental mechanisms by which advanced capabilities can emerge from (comparably simpler) agents simply due to their interaction with the environment and other agents. Informed by an extensive amount of existing literature from various fields, we outline mechanisms for enhanced agent cognition, emergent causal reasoning ability, and metacognitive awareness. We conclude by presenting some key open challenges and guidance for the development of agentic AI. We emphasize that a systems-level perspective is essential for better understanding, and purposefully shaping, agentic AI systems.

Humans leverage rich internal models of the world to reason about the future, imagine counterfactuals, and adapt flexibly to new situations. In Reinforcement Learning (RL), world models aim to capture how the environment evolves in response to the agent's actions, facilitating planning and generalization. However, typical world models directly operate on the environment variables (e.g. pixels, physical attributes), which can make their training slow and cumbersome; instead, it may be advantageous to rely on high-level latent dimensions that capture relevant multimodal variables. Global Workspace (GW) Theory offers a cognitive framework for multimodal integration and information broadcasting in the brain, and recent studies have begun to introduce efficient deep learning implementations of GW. Here, we evaluate the capabilities of an RL system combining GW with a world model. We compare our GW-Dreamer with various versions of the standard PPO and the original Dreamer algorithms. We show that performing the dreaming process (i.e., mental simulation) inside the GW latent space allows for training with fewer environment steps. As an additional emergent property, the resulting model (but not its comparison baselines) displays strong robustness to the absence of one of its observation modalities (images or simulation attributes). We conclude that the combination of GW with World Models holds great potential for improving decision-making in RL agents.

In large multimodal models (LMMs), the perception of non-language modalities (e.g., visual representations) is usually not on par with the large language models (LLMs)' powerful reasoning capabilities, deterring LMMs' performance on challenging downstream tasks. This weakness has been recently mitigated by replacing the vision encoder with a mixture-of-experts (MoE), which provides rich, multi-granularity, and diverse representations required by diverse downstream tasks. The performance of multimodal MoE largely depends on its router, which reweights and mixes the representations of different experts for each input. However, we find that the end-to-end trained router does not always produce the optimal routing weights for every test sample. To bridge the gap, we propose a novel and efficient method "Re-Routing in Test-Time (R2-T2)" that locally optimizes the vector of routing weights in test-time by moving it toward those vectors of the correctly predicted samples in a neighborhood of the test sample. We propose three R2-T2 strategies with different optimization objectives and neighbor-search spaces. R2-T2 consistently and greatly improves state-of-the-art LMMs' performance on challenging benchmarks of diverse tasks, without training any base-model parameters.