Large language models (LLMs) are increasingly used to evolve solutions to problems. However, most LLM-evolution frameworks solve static optimization problems, overlooking the open-ended adversarial dynamics that characterize real-world evolutionary processes. Here, we study Digital Red Queen (DRQ), a simple self-play algorithm that embraces these "Red Queen" dynamics via a changing objective. DRQ uses an LLM to evolve assembly programs (warriors) which compete for control of a virtual machine in the game of Core War, a Turing-complete environment studied in artificial life and connected to cybersecurity. In each round of DRQ, the model evolves a new warrior to defeat all previous ones. Over many rounds, warriors become increasingly general (relative to a set of held-out human warriors). Interestingly, across independent runs, we observe a convergence pressure toward a single generalpurpose behavioral strategy, much like convergent evolution in nature. Our work positions Core War as a rich, controllable sandbox for studying adversarial adaptation in artificial systems and for evaluating LLM-based evolution methods. More broadly, the simplicity and effectiveness of DRQ suggest that similarly minimal self-play approaches could prove useful in practical multi-agent adversarial domains, like real-world cybersecurity or combating drug resistance.

Reinforcement learning agents must generalize beyond their training experience. Prior work has focused mostly on identical training and evaluation environments. Starting from the recently introducedCrafterbenchmark, a 2-D open world survival game, we introduce a new set of environments suitable for evaluating some agent's ability to generalize on previously unseen (numbers of) objects and to adapt quickly (meta-learning). InCrafter, the agents are evaluated by the number of unlocked achievements (such as collecting resources) when trained for 1 M steps. We show that current agents struggle to generalize, and introduce novel object-centric agents that improve over strong baselines. We also provide critical insights of general interest for future work onCrafterthrough several experiments. We show that careful hyperparameter tuning improves the PPO baseline agent by a large margin and that even feedforward agents can unlock almost all achievements by relying on the inventory display. We achieve a new state-of-the-art performance on the originalCrafterenvironment. In addtion, when trained beyond 1 M steps, our tuned agents can unlock almost all achievements. We show that the recurrent PPO agents improve over feedforward ones, even with the inventory information removed. We introduceCrafterOOD, a set of 15 new environments that evaluate OOD generalization. OnCrafterOOD, we show that the current agents fail to generalize, whereas our novel object-centric agents achieve state-of-the-art OOD generalization while also being interpretable. Our code is public.

Modularity in robotics holds great potential. In principle, modular robots can be disassembled and reassembled in different robots, and possibly perform new tasks. Nevertheless, actually exploiting modularity is yet an unsolved problem: controllers usually rely on inter-module communication, a practical requirement that makes modules not perfectly interchangeable and thus limits their flexibility. Here, we focus on Voxel-based Soft Robots (VSRs), aggregations of mechanically identical elastic blocks. We use the same neural controller inside each voxel, but without any inter-voxel communication, hence enabling ideal conditions for modularity: modules are all equal and interchangeable. We optimize the parameters of the neural controller---shared among the voxels---by evolutionary computation. Crucially, we use a local self-attention mechanism inside the controller to overcome the absence of inter-module communication channels, thus enabling our robots to truly be driven by the collective intelligence of their modules. We show experimentally that the evolved robots are effective in the task of locomotion: thanks to self-attention, instances of the same controller embodied in the same robot can focus on different inputs. We also find that the evolved controllers generalize to unseen morphologies, after a short fine-tuning, suggesting that an inductive bias related to the task arises from true modularity.1

In complex systems, we often observe complex global behavior emerge from a collection of agents interacting with each other in their environment, with each individual agent acting only on locally available information, without knowing the full picture. Such systems have inspired development of artificial intelligence algorithms in areas such as swarm optimization and cellular automata. Motivated by the emergence of collective behavior from complex cellular systems, we build systems that feed each sensory input from the environment into distinct, but identical neural networks, each with no fixed relationship with one another. We show that these sensory networks can be trained to integrate information received locally, and through communication via an attention mechanism, can collectively produce a globally coherent policy. Moreover, the system can still perform its task even if the ordering of its inputs is randomly permuted several times during an episode. These permutation invariant systems also display useful robustness and generalization properties that are broadly applicable. Interactive demo and videos of our results: https://attentionneuron.github.io

Methods for dynamic difficulty adjustment allow games to be tailored to particular players to maximize their engagement. However, current methods often only modify a limited set of game features such as the difficulty of the opponents, or the availability of resources. Other approaches, such as experience-driven Procedural Content Generation (PCG), can generate complete levels with desired properties such as levels that are neither too hard nor too easy, but require many iterations. This paper presents a method that can generate and search for complete levels with a specific target difficulty in only a few trials. This advance is enabled by through an Intelligent Trial-and-Error algorithm, originally developed to allow robots to adapt quickly. Our algorithm first creates a large variety of different levels that vary across predefined dimensions such as leniency or map coverage. The performance of an AI playing agent on these maps gives a proxy for how difficult the level would be for another AI agent (e.g. one that employs Monte Carlo Tree Search instead of Greedy Tree Search); using this information, a Bayesian Optimization procedure is deployed, updating the difficulty of the prior map to reflect the ability of the agent. The approach can reliably find levels with a specific target difficulty for a variety of planning agents in only a few trials, while maintaining an understanding of their skill landscape.

Inattentional blindness is the psychological phenomenon that causes one to miss things in plain sight. It is a consequence of the selective attention in perception that lets us remain focused on important parts of our world without distraction from irrelevant details. Motivated by selective attention, we study the properties of artificial agents that perceive the world through the lens of a self-attention bottleneck. By constraining access to only a small fraction of the visual input, we show that their policies are directly interpretable in pixel space. We find neuroevolution ideal for training self-attention architectures for vision-based reinforcement learning (RL) tasks, allowing us to incorporate modules that can include discrete, non-differentiable operations which are useful for our agent. We argue that self-attention has similar properties as indirect encoding, in the sense that large implicit weight matrices are generated from a small number of key-query parameters, thus enabling our agent to solve challenging vision based tasks with at least 1000x fewer parameters than existing methods. Since our agent attends to only task critical visual hints, they are able to generalize to environments where task irrelevant elements are modified while conventional methods fail.1

Iteratively refining and critiquing sketches are crucial steps to developing effective designs. We introduce Scones, a mixed-initiative, machine-learning-driven system that enables users to iteratively author sketches from text instructions. Scones is a novel deep-learning-based system that iteratively generates scenes of sketched objects composed with semantic specifications from natural language. Scones exceeds state-of-the-art performance on a text-based scene modification task, and introduces a mask-conditioned sketching model that can generate sketches with poses specified by high-level scene information. In an exploratory user evaluation of Scones, participants reported enjoying an iterative drawing task with Scones, and suggested additional features for further applications. We believe Scones is an early step towards automated, intelligent systems that support human-in-the-loop applications for communicating ideas through sketching in art and design.

Deep networks have achieved excellent results in perceptual tasks, yet their ability to generalize to variations not seen during training has come under increasing scrutiny. In this work we focus on their ability to have invariance towards the presence or absence of details. For example, humans are able to watch cartoons, which are missing many visual details, without being explicitly trained to do so. As another example, 3D rendering software is a relatively recent development, yet people are able to understand such rendered scenes even though they are missing details (consider a film like Toy Story). The failure of ma- chine learning algorithms to do this indicates a significant gap in generalization between human abilities and the abilities of deep networks. We propose a dataset that will make it easier to study the detail-invariance problem concretely. We produce a concrete task for this: SketchTransfer, and we show that state-of-the-art domain transfer algorithms still struggle with this task. The state-of-the-art technique which achieves over 95% on MNIST → SVHN transfer only achieves 59% accuracy on the SketchTransfer task, which is much better than random (11% accuracy) but falls short of the 87% accuracy of a classifier trained directly on labeled sketches. This indicates that this task is approachable with today’s best methods but has substantial room for improvement.

Sketching is one of the most accessible techniques for communicating our ideas quickly and for collaborating in real time. Here we present a web-based environment for collaborative sketching of everyday visual concepts. We explore the integration of an artificial agent, instantiated as a recurrent neural network, who is both cooperative and responsive to actions performed by its human collaborator. To evaluate the quality of the sketches produced in this environment, we conducted an experimental user study and found that sketches produced collaboratively carried as much semantically relevant information as those produced by humans on their own. Further control analyses suggest that the semantic information in these sketches were indeed the product of collaboration, rather than attributable to the contributions of the human or the artificial agent alone. Taken together, our findings attest to the potential of systems enabling real-time collaboration between humans and machines to create novel and meaningful content.

Dramatic advances in generative models have resulted in near photographic quality for artificially rendered faces, animals and other objects in the natural world. In spite of such advances, a higher level understanding of vision and imagery does not arise from exhaustively modeling an object, but instead identifying higher-level attributes that best summarize the aspects of an object. In this work we attempt to model the drawing process of fonts by building sequential generative models of vector graphics. This model has the benefit of providing a scale-invariant representation for imagery whose latent representation may be systematically manipulated and exploited to perform style propagation. We demonstrate these results on a large dataset of fonts crawled from the web and highlight how such a model captures the statistical dependencies and richness of this dataset. We envision that our model can find use as a tool for graphic designers to facilitate font design.

Planning has been very successful for control tasks with known environment dynamics. To leverage planning in unknown environments, the agent needs to learn the dynamics from interactions with the world. However, learning dynamics models that are accurate enough for planning has been a long-standing challenge, especially in image-based domains. We propose the Deep Planning Network (PlaNet), a purely model-based agent that learns the environment dynamics from images and chooses actions through fast online planning in latent space. To achieve high performance, the dynamics model must accurately predict the rewards ahead for multiple time steps. We approach this using a latent dynamics model with both deterministic and stochastic transition components. Moreover, we propose a multi-step variational inference objective that we name latent overshooting. Using only pixel observations, our agent solves continuous control tasks with contact dynamics, partial observability, and sparse rewards, which exceed the difficulty of tasks that were previously solved by planning with learned models. PlaNet uses substantially fewer episodes and reaches final performance close to and sometimes higher than strong model-free algorithms.

Much of model-based reinforcement learning involves learning a model of an agent's world, and training an agent to leverage this model to perform a task more efficiently. While these models are demonstrably useful for agents, every naturally occurring model of the world of which we are aware---e.g., a brain---arose as the byproduct of competing evolutionary pressures for survival, not minimization of a supervised forward-predictive loss via gradient descent. That useful models can arise out of the messy and slow optimization process of evolution suggests that forward-predictive modeling can arise as a side-effect of optimization under the right circumstances. Crucially, this optimization process need not explicitly be a forward-predictive loss. In this work, we introduce a modification to traditional reinforcement learning which we call observational dropout, whereby we limit the agents ability to observe the real environment at each timestep. In doing so, we can coerce an agent into learning a world model to fill in the observation gaps during reinforcement learning. We show that the emerged world model, while not explicitly trained to predict the future, can help the agent learn key skills required to perform well in its environment. Videos of our results available at https://learningtopredict.github.io/

Not all neural network architectures are created equal, some perform much better than others for certain tasks. But how important are the weight parameters of a neural network compared to its architecture? In this work, we question to what extent neural network architectures alone, without learning any weight parameters, can encode solutions for a given task. We propose a search method for neural network architectures that can already perform a task without any explicit weight training. To evaluate these networks, we populate the connections with a single shared weight parameter sampled from a uniform random distribution, and measure the expected performance. We demonstrate that our method can find minimal neural network architectures that can perform several reinforcement learning tasks without weight training. On a supervised learning domain, we find network architectures that achieve much higher than chance accuracy on MNIST using random weights. Interactive version of this paper at https://weightagnostic.github.io/

In many reinforcement learning tasks, the goal is to learn a policy to manipulate an agent, whose design is fixed, to maximize some notion of cumulative reward. The design of the agent's physical structure is rarely optimized for the task at hand. In this work, we explore the possibility of learning a version of the agent's design that is better suited for its task, jointly with the policy. We propose an alteration to the popular OpenAI Gym framework, where we parameterize parts of an environment, and allow an agent to jointly learn to modify these environment parameters along with its policy. We demonstrate that an agent can learn a better structure of its body that is not only better suited for the task, but also facilitates policy learning. Joint learning of policy and structure may even uncover design principles that are useful for assisted-design applications.

A generative recurrent neural network is quickly trained in an unsupervised manner to model popular reinforcement learning environments through compressed spatio-temporal representations. The world model's extracted features are fed into compact and simple policies trained by evolution, achieving state of the art results in various environments. We also train our agent entirely inside of an environment generated by its own internal world model, and transfer this policy back into the actual environment. Interactive version of this paper is available at https://worldmodels.github.io