Managing extreme AI risks amid rapid progress.

Preparation requires technical research and development, as well as adaptive, proactive governance.

Evolving Multimodal Robot Behavior via Many Stepping Stones with the Combinatorial Multiobjective Evolutionary Algorithm.

An important challenge in reinforcement learning is to solve multimodal problems, where agents have to act in qualitatively different ways depending on the circumstances. Because multimodal problems are often too difficult to solve directly, it is often helpful to define a curriculum, which is an ordered set of subtasks that can serve as the stepping stones for solving the overall problem. Unfortunately, choosing an effective ordering for these subtasks is difficult, and a poor ordering can reduce the performance of the learning process. Here, we provide a thorough introduction and investigation of the Combinatorial Multiobjective Evolutionary Algorithm (CMOEA), which allows all combinations of subtasks to be explored simultaneously. We compare CMOEA against three algorithms that can similarly optimize on multiple subtasks simultaneously: NSGA-II, NSGA-III, and ε-Lexicase Selection. The algorithms are tested on a function-optimization problem with two subtasks, a simulated multimodal robot locomotion problem with six subtasks, and a simulated robot maze-navigation problem where a hundred random mazes are treated as subtasks. On these problems, CMOEA either outperforms or is competitive with the controls. As a separate contribution, we show that adding a linear combination over all objectives can improve the ability of the control algorithms to solve these multimodal problems. Lastly, we show that CMOEA can leverage auxiliary objectives more effectively than the controls on the multimodal locomotion task. In general, our experiments suggest that CMOEA is a promising algorithm for solving multimodal problems.

First return, then explore.

Reinforcement learning promises to solve complex sequential-decision problems autonomously by specifying a high-level reward function only. However, reinforcement learning algorithms struggle when, as is often the case, simple and intuitive rewards provide sparse1 and deceptive2 feedback. Avoiding these pitfalls requires a thorough exploration of the environment, but creating algorithms that can do so remains one of the central challenges of the field. Here we hypothesize that the main impediment to effective exploration originates from algorithms forgetting how to reach previously visited states (detachment) and failing to first return to a state before exploring from it (derailment). We introduce Go-Explore, a family of algorithms that addresses these two challenges directly through the simple principles of explicitly 'remembering' promising states and returning to such states before intentionally exploring. Go-Explore solves all previously unsolved Atari games and surpasses the state of the art on all hard-exploration games1, with orders-of-magnitude improvements on the grand challenges of Montezuma's Revenge and Pitfall. We also demonstrate the practical potential of Go-Explore on a sparse-reward pick-and-place robotics task. Additionally, we show that adding a goal-conditioned policy can further improve Go-Explore's exploration efficiency and enable it to handle stochasticity throughout training. The substantial performance gains from Go-Explore suggest that the simple principles of remembering states, returning to them, and exploring from them are a powerful and general approach to exploration-an insight that may prove critical to the creation of truly intelligent learning agents.

Improving the accessibility and transferability of machine learning algorithms for identification of animals in camera trap images: MLWIC2.

Motion-activated wildlife cameras (or "camera traps") are frequently used to remotely and noninvasively observe animals. The vast number of images collected from camera trap projects has prompted some biologists to employ machine learning algorithms to automatically recognize species in these images, or at least filter-out images that do not contain animals. These approaches are often limited by model transferability, as a model trained to recognize species from one location might not work as well for the same species in different locations. Furthermore, these methods often require advanced computational skills, making them inaccessible to many biologists. We used 3 million camera trap images from 18 studies in 10 states across the United States of America to train two deep neural networks, one that recognizes 58 species, the "species model," and one that determines if an image is empty or if it contains an animal, the "empty-animal model." Our species model and empty-animal model had accuracies of 96.8% and 97.3%, respectively. Furthermore, the models performed well on some out-of-sample datasets, as the species model had 91% accuracy on species from Canada (accuracy range 36%-91% across all out-of-sample datasets) and the empty-animal model achieved an accuracy of 91%-94% on out-of-sample datasets from different continents. Our software addresses some of the limitations of using machine learning to classify images from camera traps. By including many species from several locations, our species model is potentially applicable to many camera trap studies in North America. We also found that our empty-animal model can facilitate removal of images without animals globally. We provide the trained models in an R package (MLWIC2: Machine Learning for Wildlife Image Classification in R), which contains Shiny Applications that allow scientists with minimal programming experience to use trained models and train new models in six neural network architectures with varying depths.

The Surprising Creativity of Digital Evolution: A Collection of Anecdotes from the Evolutionary Computation and Artificial Life Research Communities.

Evolution provides a creative fount of complex and subtle adaptations that often surprise the scientists who discover them. However, the creativity of evolution is not limited to the natural world: Artificial organisms evolving in computational environments have also elicited surprise and wonder from the researchers studying them. The process of evolution is an algorithmic process that transcends the substrate in which it occurs. Indeed, many researchers in the field of digital evolution can provide examples of how their evolving algorithms and organisms have creatively subverted their expectations or intentions, exposed unrecognized bugs in their code, produced unexpectedly adaptations, or engaged in behaviors and outcomes, uncannily convergent with ones found in nature. Such stories routinely reveal surprise and creativity by evolution in these digital worlds, but they rarely fit into the standard scientific narrative. Instead they are often treated as mere obstacles to be overcome, rather than results that warrant study in their own right. Bugs are fixed, experiments are refocused, and one-off surprises are collapsed into a single data point. The stories themselves are traded among researchers through oral tradition, but that mode of information transmission is inefficient and prone to error and outright loss. Moreover, the fact that these stories tend to be shared only among practitioners means that many natural scientists do not realize how interesting and lifelike digital organisms are and how natural their evolution can be. To our knowledge, no collection of such anecdotes has been published before. This article is the crowd-sourced product of researchers in the fields of artificial life and evolutionary computation who have provided first-hand accounts of such cases. It thus serves as a written, fact-checked collection of scientifically important and even entertaining stories. In doing so we also present here substantial evidence that the existence and importance of evolutionary surprises extends beyond the natural world, and may indeed be a universal property of all complex evolving systems.

The Emergence of Canalization and Evolvability in an Open-Ended, Interactive Evolutionary System.

Many believe that an essential component for the discovery of the tremendous diversity in natural organisms was the evolution of evolvability, whereby evolution speeds up its ability to innovate by generating a more adaptive pool of offspring. One hypothesized mechanism for evolvability is developmental canalization, wherein certain dimensions of variation become more likely to be traversed and others are prevented from being explored (e.g., offspring tend to have similar-size legs, and mutations affect the length of both legs, not each leg individually). While ubiquitous in nature, canalization is rarely reported in computational simulations of evolution, which deprives us of in silico examples of canalization to study and raises the question of which conditions give rise to this form of evolvability. Answering this question would shed light on why such evolvability emerged naturally, and it could accelerate engineering efforts to harness evolution to solve important engineering challenges. In this article, we reveal a unique system in which canalization did emerge in computational evolution. We document that genomes entrench certain dimensions of variation that were frequently explored during their evolutionary history. The genetic representation of these organisms also evolved to be more modular and hierarchical than expected by chance, and we show that these organizational properties correlate with increased fitness. Interestingly, the type of computational evolutionary experiment that produced this evolvability was very different from traditional digital evolution in that there was no objective, suggesting that open-ended, divergent evolutionary processes may be necessary for the evolution of evolvability.

Automatically identifying, counting, and describing wild animals in camera-trap images with deep learning.

Having accurate, detailed, and up-to-date information about the location and behavior of animals in the wild would improve our ability to study and conserve ecosystems. We investigate the ability to automatically, accurately, and inexpensively collect such data, which could help catalyze the transformation of many fields of ecology, wildlife biology, zoology, conservation biology, and animal behavior into "big data" sciences. Motion-sensor "camera traps" enable collecting wildlife pictures inexpensively, unobtrusively, and frequently. However, extracting information from these pictures remains an expensive, time-consuming, manual task. We demonstrate that such information can be automatically extracted by deep learning, a cutting-edge type of artificial intelligence. We train deep convolutional neural networks to identify, count, and describe the behaviors of 48 species in the 3.2 million-image Snapshot Serengeti dataset. Our deep neural networks automatically identify animals with >93.8% accuracy, and we expect that number to improve rapidly in years to come. More importantly, if our system classifies only images it is confident about, our system can automate animal identification for 99.3% of the data while still performing at the same 96.6% accuracy as that of crowdsourced teams of human volunteers, saving >8.4 y (i.e., >17,000 h at 40 h/wk) of human labeling effort on this 3.2 million-image dataset. Those efficiency gains highlight the importance of using deep neural networks to automate data extraction from camera-trap images, reducing a roadblock for this widely used technology. Our results suggest that deep learning could enable the inexpensive, unobtrusive, high-volume, and even real-time collection of a wealth of information about vast numbers of animals in the wild.

Diffusion-based neuromodulation can eliminate catastrophic forgetting in simple neural networks.

A long-term goal of AI is to produce agents that can learn a diversity of skills throughout their lifetimes and continuously improve those skills via experience. A longstanding obstacle towards that goal is catastrophic forgetting, which is when learning new information erases previously learned information. Catastrophic forgetting occurs in artificial neural networks (ANNs), which have fueled most recent advances in AI. A recent paper proposed that catastrophic forgetting in ANNs can be reduced by promoting modularity, which can limit forgetting by isolating task information to specific clusters of nodes and connections (functional modules). While the prior work did show that modular ANNs suffered less from catastrophic forgetting, it was not able to produce ANNs that possessed task-specific functional modules, thereby leaving the main theory regarding modularity and forgetting untested. We introduce diffusion-based neuromodulation, which simulates the release of diffusing, neuromodulatory chemicals within an ANN that can modulate (i.e. up or down regulate) learning in a spatial region. On the simple diagnostic problem from the prior work, diffusion-based neuromodulation 1) induces task-specific learning in groups of nodes and connections (task-specific localized learning), which 2) produces functional modules for each subtask, and 3) yields higher performance by eliminating catastrophic forgetting. Overall, our results suggest that diffusion-based neuromodulation promotes task-specific localized learning and functional modularity, which can help solve the challenging, but important problem of catastrophic forgetting.

How evolution learns to generalise: Using the principles of learning theory to understand the evolution of developmental organisation.

One of the most intriguing questions in evolution is how organisms exhibit suitable phenotypic variation to rapidly adapt in novel selective environments. Such variability is crucial for evolvability, but poorly understood. In particular, how can natural selection favour developmental organisations that facilitate adaptive evolution in previously unseen environments? Such a capacity suggests foresight that is incompatible with the short-sighted concept of natural selection. A potential resolution is provided by the idea that evolution may discover and exploit information not only about the particular phenotypes selected in the past, but their underlying structural regularities: new phenotypes, with the same underlying regularities, but novel particulars, may then be useful in new environments. If true, we still need to understand the conditions in which natural selection will discover such deep regularities rather than exploiting 'quick fixes' (i.e., fixes that provide adaptive phenotypes in the short term, but limit future evolvability). Here we argue that the ability of evolution to discover such regularities is formally analogous to learning principles, familiar in humans and machines, that enable generalisation from past experience. Conversely, natural selection that fails to enhance evolvability is directly analogous to the learning problem of over-fitting and the subsequent failure to generalise. We support the conclusion that evolving systems and learning systems are different instantiations of the same algorithmic principles by showing that existing results from the learning domain can be transferred to the evolution domain. Specifically, we show that conditions that alleviate over-fitting in learning systems successfully predict which biological conditions (e.g., environmental variation, regularity, noise or a pressure for developmental simplicity) enhance evolvability. This equivalence provides access to a well-developed theoretical framework from learning theory that enables a characterisation of the general conditions for the evolution of evolvability.

Improving HybrID: How to best combine indirect and direct encoding in evolutionary algorithms.

Many challenging engineering problems are regular, meaning solutions to one part of a problem can be reused to solve other parts. Evolutionary algorithms with indirect encoding perform better on regular problems because they reuse genomic information to create regular phenotypes. However, on problems that are mostly regular, but contain some irregularities, which describes most real-world problems, indirect encodings struggle to handle the irregularities, hurting performance. Direct encodings are better at producing irregular phenotypes, but cannot exploit regularity. An algorithm called HybrID combines the best of both: it first evolves with indirect encoding to exploit problem regularity, then switches to direct encoding to handle problem irregularity. While HybrID has been shown to outperform both indirect and direct encoding, its initial implementation required the manual specification of when to switch from indirect to direct encoding. In this paper, we test two new methods to improve HybrID by eliminating the need to manually specify this parameter. Auto-Switch-HybrID automatically switches from indirect to direct encoding when fitness stagnates. Offset-HybrID simultaneously evolves an indirect encoding with directly encoded offsets, eliminating the need to switch. We compare the original HybrID to these alternatives on three different problems with adjustable regularity. The results show that both Auto-Switch-HybrID and Offset-HybrID outperform the original HybrID on different types of problems, and thus offer more tools for researchers to solve challenging problems. The Offset-HybrID algorithm is particularly interesting because it suggests a path forward for automatically and simultaneously combining the best traits of indirect and direct encoding.

Correction: Curiosity Search: Producing Generalists by Encouraging Individuals to Continually Explore and Acquire Skills Throughout Their Lifetime.

[This corrects the article DOI: 10.1371/journal.pone.0162235.].

Curiosity Search: Producing Generalists by Encouraging Individuals to Continually Explore and Acquire Skills throughout Their Lifetime.

Natural animals are renowned for their ability to acquire a diverse and general skill set over the course of their lifetime. However, research in artificial intelligence has yet to produce agents that acquire all or even most of the available skills in non-trivial environments. One candidate algorithm for encouraging the production of such individuals is Novelty Search, which pressures organisms to exhibit different behaviors from other individuals. However, we hypothesized that Novelty Search would produce sub-populations of specialists, in which each individual possesses a subset of skills, but no one organism acquires all or most of the skills. In this paper, we propose a new algorithm called Curiosity Search, which is designed to produce individuals that acquire as many skills as possible during their lifetime. We show that in a multiple-skill maze environment, Curiosity Search does produce individuals that explore their entire domain, while a traditional implementation of Novelty Search produces specialists. However, we reveal that when modified to encourage intra-life behavioral diversity, Novelty Search can produce organisms that explore almost as much of their environment as Curiosity Search, although Curiosity Search retains a significant performance edge. Finally, we show that Curiosity Search is a useful helper objective when combined with Novelty Search, producing individuals that acquire significantly more skills than either algorithm alone.

WebAL Comes of Age: A Review of the First 21 Years of Artificial Life on the Web.

We present a survey of the first 21 years of web-based artificial life (WebAL) research and applications, broadly construed to include the many different ways in which artificial life and web technologies might intersect. Our survey covers the period from 1994-when the first WebAL work appeared-up to the present day, together with a brief discussion of relevant precursors. We examine recent projects, from 2010-2015, in greater detail in order to highlight the current state of the art. We follow the survey with a discussion of common themes and methodologies that can be observed in recent work and identify a number of likely directions for future work in this exciting area.

The Evolutionary Origins of Hierarchy.

Hierarchical organization-the recursive composition of sub-modules-is ubiquitous in biological networks, including neural, metabolic, ecological, and genetic regulatory networks, and in human-made systems, such as large organizations and the Internet. To date, most research on hierarchy in networks has been limited to quantifying this property. However, an open, important question in evolutionary biology is why hierarchical organization evolves in the first place. It has recently been shown that modularity evolves because of the presence of a cost for network connections. Here we investigate whether such connection costs also tend to cause a hierarchical organization of such modules. In computational simulations, we find that networks without a connection cost do not evolve to be hierarchical, even when the task has a hierarchical structure. However, with a connection cost, networks evolve to be both modular and hierarchical, and these networks exhibit higher overall performance and evolvability (i.e. faster adaptation to new environments). Additional analyses confirm that hierarchy independently improves adaptability after controlling for modularity. Overall, our results suggest that the same force-the cost of connections-promotes the evolution of both hierarchy and modularity, and that these properties are important drivers of network performance and adaptability. In addition to shedding light on the emergence of hierarchy across the many domains in which it appears, these findings will also accelerate future research into evolving more complex, intelligent computational brains in the fields of artificial intelligence and robotics.

Robots that can adapt like animals.

Robots have transformed many industries, most notably manufacturing, and have the power to deliver tremendous benefits to society, such as in search and rescue, disaster response, health care and transportation. They are also invaluable tools for scientific exploration in environments inaccessible to humans, from distant planets to deep oceans. A major obstacle to their widespread adoption in more complex environments outside factories is their fragility. Whereas animals can quickly adapt to injuries, current robots cannot 'think outside the box' to find a compensatory behaviour when they are damaged: they are limited to their pre-specified self-sensing abilities, can diagnose only anticipated failure modes, and require a pre-programmed contingency plan for every type of potential damage, an impracticality for complex robots. A promising approach to reducing robot fragility involves having robots learn appropriate behaviours in response to damage, but current techniques are slow even with small, constrained search spaces. Here we introduce an intelligent trial-and-error algorithm that allows robots to adapt to damage in less than two minutes in large search spaces without requiring self-diagnosis or pre-specified contingency plans. Before the robot is deployed, it uses a novel technique to create a detailed map of the space of high-performing behaviours. This map represents the robot's prior knowledge about what behaviours it can perform and their value. When the robot is damaged, it uses this prior knowledge to guide a trial-and-error learning algorithm that conducts intelligent experiments to rapidly discover a behaviour that compensates for the damage. Experiments reveal successful adaptations for a legged robot injured in five different ways, including damaged, broken, and missing legs, and for a robotic arm with joints broken in 14 different ways. This new algorithm will enable more robust, effective, autonomous robots, and may shed light on the principles that animals use to adapt to injury.

Neural modularity helps organisms evolve to learn new skills without forgetting old skills.

A long-standing goal in artificial intelligence is creating agents that can learn a variety of different skills for different problems. In the artificial intelligence subfield of neural networks, a barrier to that goal is that when agents learn a new skill they typically do so by losing previously acquired skills, a problem called catastrophic forgetting. That occurs because, to learn the new task, neural learning algorithms change connections that encode previously acquired skills. How networks are organized critically affects their learning dynamics. In this paper, we test whether catastrophic forgetting can be reduced by evolving modular neural networks. Modularity intuitively should reduce learning interference between tasks by separating functionality into physically distinct modules in which learning can be selectively turned on or off. Modularity can further improve learning by having a reinforcement learning module separate from sensory processing modules, allowing learning to happen only in response to a positive or negative reward. In this paper, learning takes place via neuromodulation, which allows agents to selectively change the rate of learning for each neural connection based on environmental stimuli (e.g. to alter learning in specific locations based on the task at hand). To produce modularity, we evolve neural networks with a cost for neural connections. We show that this connection cost technique causes modularity, confirming a previous result, and that such sparsely connected, modular networks have higher overall performance because they learn new skills faster while retaining old skills more and because they have a separate reinforcement learning module. Our results suggest (1) that encouraging modularity in neural networks may help us overcome the long-standing barrier of networks that cannot learn new skills without forgetting old ones, and (2) that one benefit of the modularity ubiquitous in the brains of natural animals might be to alleviate the problem of catastrophic forgetting.

The evolutionary origins of modularity.

A central biological question is how natural organisms are so evolvable (capable of quickly adapting to new environments). A key driver of evolvability is the widespread modularity of biological networks--their organization as functional, sparsely connected subunits--but there is no consensus regarding why modularity itself evolved. Although most hypotheses assume indirect selection for evolvability, here we demonstrate that the ubiquitous, direct selection pressure to reduce the cost of connections between network nodes causes the emergence of modular networks. Computational evolution experiments with selection pressures to maximize network performance and minimize connection costs yield networks that are significantly more modular and more evolvable than control experiments that only select for performance. These results will catalyse research in numerous disciplines, such as neuroscience and genetics, and enhance our ability to harness evolution for engineering purposes.

Ontogeny tends to recapitulate phylogeny in digital organisms.

Biologists have long debated whether ontogeny recapitulates phylogeny and, if so, why. Two plausible explanations are that (i) changes to early developmental stages are selected against because they tend to disrupt later development and (ii) simpler structures often precede more complex ones in both ontogeny and phylogeny if the former serve as building blocks for the latter. It is difficult to test these hypotheses experimentally in natural systems, so we used a computational system that exhibits evolutionary dynamics. We observed that ontogeny does indeed recapitulate phylogeny; traits that arose earlier in a lineage's history also tended to be expressed earlier in the development of individuals. The relative complexity of traits contributed substantially to this correlation, but a significant tendency toward recapitulation remained even after accounting for trait complexity. This additional effect provides evidence that selection against developmental disruption also contributed to the conservation of early stages in development.

Selective pressures for accurate altruism targeting: evidence from digital evolution for difficult-to-test aspects of inclusive fitness theory.

Inclusive fitness theory predicts that natural selection will favour altruist genes that are more accurate in targeting altruism only to copies of themselves. In this paper, we provide evidence from digital evolution in support of this prediction by competing multiple altruist-targeting mechanisms that vary in their accuracy in determining whether a potential target for altruism carries a copy of the altruist gene. We compete altruism-targeting mechanisms based on (i) kinship (kin targeting), (ii) genetic similarity at a level greater than that expected of kin (similarity targeting), and (iii) perfect knowledge of the presence of an altruist gene (green beard targeting). Natural selection always favoured the most accurate targeting mechanism available. Our investigations also revealed that evolution did not increase the altruism level when all green beard altruists used the same phenotypic marker. The green beard altruism levels stably increased only when mutations that changed the altruism level also changed the marker (e.g. beard colour), such that beard colour reliably indicated the altruism level. For kin- and similarity-targeting mechanisms, we found that evolution was able to stably adjust altruism levels. Our results confirm that natural selection favours altruist genes that are increasingly accurate in targeting altruism to only their copies. Our work also emphasizes that the concept of targeting accuracy must include both the presence of an altruist gene and the level of altruism it produces.

Natural selection fails to optimize mutation rates for long-term adaptation on rugged fitness landscapes.

The rate of mutation is central to evolution. Mutations are required for adaptation, yet most mutations with phenotypic effects are deleterious. As a consequence, the mutation rate that maximizes adaptation will be some intermediate value. Here, we used digital organisms to investigate the ability of natural selection to adjust and optimize mutation rates. We assessed the optimal mutation rate by empirically determining what mutation rate produced the highest rate of adaptation. Then, we allowed mutation rates to evolve, and we evaluated the proximity to the optimum. Although we chose conditions favorable for mutation rate optimization, the evolved rates were invariably far below the optimum across a wide range of experimental parameter settings. We hypothesized that the reason that mutation rates evolved to be suboptimal was the ruggedness of fitness landscapes. To test this hypothesis, we created a simplified landscape without any fitness valleys and found that, in such conditions, populations evolved near-optimal mutation rates. In contrast, when fitness valleys were added to this simple landscape, the ability of evolving populations to find the optimal mutation rate was lost. We conclude that rugged fitness landscapes can prevent the evolution of mutation rates that are optimal for long-term adaptation. This finding has important implications for applied evolutionary research in both biological and computational realms.