Prey capture requires the zebrafish to select, pursue, and ultimately consume fast moving single-celled organisms swimming through its environment. The larval zebrafish, a teleost withВ ~100,000 neurons, is a tractable model organism amenable to an array of modern neuroscience techniques ( Avella et al., 2012 Dunn et al., 2016).

Here, we examine these questions through characterization and computational modeling of 3D prey capture sequences executed by the larval zebrafish. Therefore, characterizing the goals, algorithms, and advantages of animals possessing implicit models should provide insight into the evolution of more advanced forms of predictive knowledge ( Spelke and Hespos, 2018). Beneficially stochastic foraging has been observed in animals ranging down to micro-organisms, while predated animals use mixed stochastic strategies to avoid predictability by predators ( Jensen, 2018).Д¬an internal models that mediate predictive abilities be reduced to a set of simple rules that are benefited by the stochasticity of neural systems? We posit that explicit physical intelligence is likely built upon a framework of more primitive sensorimotor behaviors that constitute an implicit model of how objects exist and move within the world (i.e. Crayfish and paddlefish, for instance, both take advantage of stochastic resonance to detect sparse prey and predators ( Douglass et al., 1993 Russett et al., 1999).

However, there is precedent for noisy sensory detection and stochastic movements working to the benefit of many animals. Noise in any computing system is usually considered inconvenient and a nuisance to be overcome ( KГ¶rding and Wolpert, 2004). Moreover, many theories of mind rely on deterministic digital logic, while the brain generates intelligence using noisy, stochastic units in neurons (e.g. Much progress has been made on the neurobiological and computational principles that could mediate these complex abilities, but how exactly internal models are built from component behavioral parts is unknown. Indeed, humans and animals have evolved internal models that allow us to both predict ongoing dynamics in the environment and anticipate how our own actions give rise to consequences in the world ( Battaglia et al., 2013 Ullman et al., 2017 Baillargeon, 1987 Mischiati et al., 2015 Borghuis and Leonardo, 2015). Intelligence itself can been defined as the practice of model-building about ourselves and our surroundings ( Lake et al., 2017). However, a precise mechanistic explanation of how intelligent behavior is generated has remained elusive. The synthesis of intelligence from the interaction of вЂmindlessвЂ™ behavioral modules has long been a staple of computer science and artificial intelligence ( Brooks, 1991 Minsky, 1988 Braitenberg, 1986). Bees learning to pull strings for hidden rewards appear to be displaying insight and ingenuity, but are in fact instituting a set of observational and associative learning rules in sequence ( Alem et al., 2016). For example, seemingly complicated behaviors like schooling in fish arise from stereotypic visuomotor rules being executed by members of the group, without reference to the emerging global pattern ( Couzin and Krause, 2003). It is becoming clear from recent ethological and neuroscience studies that remarkable capabilities in animals are often built by combining sets of more basic behaviors. In sum, our study reveals that zebrafish are equipped with a recursive prey capture algorithm, built up from simple stochastic rules, that embodies an implicit predictive model of the world. This surprising result coalesces with recent findings that illustrate the benefits of biological stochasticity to adaptive behavior.

Importantly, the stochasticity in the fishвЂ™s sensorimotor transformations provides a considerable advantage over equivalent noise-free strategies. Using a novel naturalistic 3D setup, we show that the zebrafish combines position and velocity perception to construct a future positional estimate of its prey, indicating an ability to project trajectories forward in time. For this task, we chose the larval zebrafish, a tractable vertebrate that pursues and captures swimming microbes. Here, we wondered whether predictive models mediating prey capture could be reduced to a simple set of sensorimotor rules performed by a primitive organism. The computational principles underlying predictive capabilities in animals are poorly understood.

Elements 3d