ABSTRACT: Discussions about artificial intelligence, and neuroscience more generally, are almost exclusively nomothetic, which is like a telescope that can only detect major outcroppings such as average differences between species. Switching to an idiographic microscope that can discern individual differences within a species is especially relevant for intelligence in living organisms unlike silicon-based artificial intelligence. The most notable feature about intelligence in living organisms is that it evolves through natural selection of random mutations in DNA. Reproductive fitness winnows this within-species DNA variation but does not eliminate it, in part because DNA variation is money in the bank for future evolution. Of our three billion base pairs of DNA, 99.9% are identical. This is what makes us human. But the 0.1% that differs is what makes us individuals, including individual differences in intelligence. Inherited DNA differences are a feature, not a bug, of human intelligence. 

I will highlight what genetic research, both quantitative genetics (e.g., twin studies) and molecular genetics (DNA studies), has revealed about individual differences in human intelligence. I conclude that inherited DNA differences are the major systematic force responsible for individual differences in intelligence. The environment is also important but genetic research shows that these environmental influences are not nurture, which connotes systematic effects of family environment. To the contrary, environmental influences are largely random in the philosophy of science sense of being unpredictable, chance rather than choice. 

DNA research has shown that the heritability of intelligence is caused by thousands of DNA differences each with miniscule effects. Nonetheless, it is possible to aggregate these DNA differences to create an index, called a polygenic score, that can predict from birth more than 10% of the variance of intelligence test scores in adulthood. 

ABSTRACT: What underlies the ability to deal with numbers and where did it come from? It has been hypothesized that our ability to accurately represent the number of objects in a set (numerousness), and to carry out numerical comparisons and arithmetic, developed from an evolutionarily conserved system for approximating numerical magnitude. Non-symbolic number cognition based on an approximate sense of magnitude has been documented in a variety of species. However, we know little about its origins (i.e., to what extent experience would shape it) and of its neural and molecular bases. To address the first issue we performed single cell recordings in awake young domestic chicks. We found neurons selective to number in the caudal nidopallium (a higher associative area with functional similarities to the mammalian prefrontal cortex), which suggest that an approximate sense of magnitude can be an inborn feature in the avian brain. To address the issue of circuitry and molecular bases of the sense of magnitude we made use of zebrafish, that in recent years became established as ideal developmental and behavioral genetic model system. Using a combination of early gene expression and in-situ hybridization we identified for the first time a small region in the caudal part of the dorso-central division of the zebrafish pallium that shows selective activation upon change in numerousness of visual stimuli. As pallial regions are implicated in number cognition in mammals and birds, these findings support the existence of an evolutionarily conserved system for approximating magnitudes and provide an avenue for exploring its underlying molecular and genetic correlates. 

ABSTRACT: The ability to make the olfactory world intelligible is a key capability of the Drosophila olfactory system that we shall call olfactory intelligence. Memory in the Mushroom Body of the fruit fly brain depends on the encoding and processing of odorants in the first three stages of the Early Olfactory System: the Antenna, the Antennal Lobe and the Mushroom Body Calyx. The Kenyon Cells (KCs) of the Calyx provide the Mushroom Body compartments with the identity of pure and odorant mixtures encoded as a train of spikes.

Characterizing the structure of the code underlying the KC spike trains is a major challenge in neuroscience.

To address this challenge we start by modeling the space of odorants using constructs of both semantic and syntactic information. Odorant semantics concerns the identity of odorants while odorant syntactics pertains to their concentration amplitude. These odorant attributes are multiplicatively coupled in the process of olfactory transduction. A key question that early olfactory systems must address is how to disentangle the odorant semantic information from the odorant syntactic information.

To that end, based on our previous work on the functional logic of odor signal processing in the Antennal Lobe, we established that the Antennal Lobe and Calyx jointly remove the concentration dependency of the odorant information from the confounding representation of the Antenna. We demonstrated that these circuits separate the odorant semantics from syntax, thereby undoing the multiplicative coupling of these two information streams in the Antenna.

We showed that in the Calyx the sought after semantic information underlies the ranking of the KC dendritic output after the KC dendritic input undergoes the PN-KC random connectivity and the spatio-temporal feed- back provided by the APL neuron. Consequently, expansion recoding in the Calyx characterizes the structure of vector PN responses by computing fixed mean random additive combinations, and use their ranking as a simple yet powerful way of extracting the semantic information of the odorant. More importantly, we addressed the combinatorial complexity of ranking by mapping, for each KC, the concentration-invariant dendritic output into the spike domain. The proposed time code takes the first spike of each active KC and joins them all together at generation time into a single first spike sequence. Clearly, the order of the first spikes across the population of KCs reflects the ranking order at negligible complexity. The existence of such concentration-invariant spike code is supported by increasing evidence in the Antenna and Antennal Lobe.

Time is an intrinsic variable of the concentration waveform, but not of the odorant object identity. Interestingly, the key result of the modeling and characterization of the early olfactory system we advance here asserts that the semantic information is mapped into the time domain by the Calyx circuit, in the form of the first spike sequence code. This allows a low complexity single readout of the semantic information at the downstream MBONs regardless of the exact connectivity between PNs and KCs in individual flies. The code itself is temporally bounded, making it possible for timely memory access in the MB compartments.

Abbreviations: Project Neuron (PN), Anterior Paired Lateral (APL) neuron. 

ABSTRACT: Hummingbirds and bees develop short, repeatable routes around profitable sites (called traplines). However, the size of their brains are not only orders of magnitude different, but their brain morphologies are also vastly dissimilar. And yet, when foraging for small nectar rewards across hundreds of flowers in a single day, their behaviour is equally effective. While hummingbirds and bees seem to converge in the types of information they can learn and use when foraging from one flower, recent experiments suggest that when solving multiple destination problems these animals have different strategies. From the start, hummingbirds simply visit the nearest neighbouring flower linking multiple flowers and can easily reroute in response to changes in the quality of the resources by changing the direction of their routes. Bees, however, tend to develop optimal routes through trial and error improving their routes incrementally and individual bees have been shown to have strong directional biases which make rerouting harder for bees. The problem of finding the shortest possible route between multiple destinations, which is essentially the Travel salesperson problem, is common to many central place foragers. Depending on each species’ cognitive abilities and their behavioural flexibility, the same problem might be solved in a myriad of ways. I will discuss how comparing the foraging traplines that hummingbirds and bees use can help us understand how different brains use information differently. 

ABSTRACT: During high-speed behaviour, animals must predict, detect, process, and respond synchronously to rapid environmental changes, including those caused by their own movements. Yet how neural systems achieve such precision remains unclear. Here, we investigate how the housefly (Musca domestica), renowned for rapid aerial manoeuvres, maintains visual precision during ultrafast movements. Although fast motion typically blurs vision, houseflies retain exceptional visual acuity, particularly during saccadic stimulation; their visual neurons achieve benchmark rates of information sampling (~2,500 bits/s) and synaptic transmission (~4,100 bits/s), far exceeding previous estimates. Using advanced techniques, we measured photomechanical and intracellular photoreceptor responses to sequential light stimuli mimicking saccades, tracking their transmission to large monopolar cells (LMCs)—the primary interneurons of the visual pathway. Photoreceptor–LMC synapses specifically adapt to saccadic input by shifting transmission to higher frequencies, a mechanism we term synaptic frequency jumping, extending visual bandwidth to ~920 Hz, eliminating synaptic delays, and quadrupling previous flicker-fusion estimates (~230 Hz). Ultrafast behavioural assays confirm flies respond to stimuli within ~13–20 ms, even as photoreceptors are still reaching peak response (9–16 ms), challenging classical sequential processing models. Our biophysically realistic photoreceptor-LMC model reveals how photomechanical, quantal, and refractory sampling co-adapt with behaviour, enabling houseflies to actively shape their visual input through self-induced saccades. This interaction drives frequency jumping, efficient coding, hyperacute vision, and neural synchronisation. These findings redefine foundational principles of compound eye function and uncover a universal neural strategy for high-speed, predictive processing. 

ABSTRACT: Three of the most striking features of primates—particularly the great apes—are their behavioural variability, flexibility, and adaptability. For example, certain species demonstrate the ability to use a single tool for multiple functions, and conversely, employ multiple tools to achieve a single goal. A similar pattern emerges in gestural communication, where one gesture may serve multiple purposes, and multiple gestures may serve a single purpose. In this talk, I will argue that this kind of means-ends dissociation is a fundamental mechanism underpinning primates’ behavioural versatility. In the second part of my presentation, I will delve into the motivational and cognitive foundations of means-ends dissociation across domains such as problem solving, social learning, and goal attribution. Specifically, I will propose that an intrinsic drive to acquire information, the capacity to extract rules and relationships from both social and non-social stimuli, and the use of inference and prediction equip primates with a robust and generalizable psychological toolkit—one that enables them to navigate a wide range of environmental and social challenges. 

ABSTRACT: Performance in autonomous robotics is improving. Robot dogs are now able to do things that look a bit doggy.  But as performance is converging, the gulf between how robot and biological agents work is widening.  Here we argue that this is because robots and animals operate under fundamentally different resource constraints.  To illustrate we focus on the action selection problem: the deceptively difficult problem of deciding what to do next. Robots can utilise solutions to the action selection problem that are physically and energetically impossible for biological brains.  We discuss how action selection is solved by animals, and why there still might be benefits for roboticists from taking a more biological approach. 

ABSTRACT: Contemporary AI-based approaches such as transformers promise a step change for robotic autonomy. However, their hardware, data, training, and inference resource costs are all very substantial, and risk placing an additional burden on our already strained environment. Furthermore, their robustness and adaptability also remains to be confirmed. In contrast, the brain is tremendously data, energy, and resource efficient, yet provides highly robust and flexible general autonomy. In this talk I will review how the study of some of the smallest brains on the planet, the insects’, can teach us about the path to more efficient and robust robot autonomy, and how the resulting new class of AI is already making its way into the market.  

ABSTRACT:  Navigation, whether over a short or long range, requires an animal to track its location in a consistent frame of reference, over a behaviourally relevant time scale. Sensory information resulting from translation through space needs to be mapped into this frame of reference, and movement decisions extracted from it, often requiring transformation between egocentric and allocentric encodings. The central complex circuit in the insect brain appears to perform exactly this function for the insect, not only maintaining a compass-like heading estimate, but also integrating self-motion, sensory information and goals into the same reference frame, and producing steering output.  For insects using the sun as their main compass cue, such as honeybees following a previously learned vector to a food location, the time scale of behaviour necessitates that some part of this circuit is time-compensated to maintain a consistent reference as the sun moves. We propose a neural circuit by which such compensation could occur. For honeybees performing dances on a vertical honeycomb in a dark hive, transformation to a new frame of reference (gravity) is required. We have modelled how the follower bee could relate its own orientation to the dancer within this frame of reference, and thus recover a vector which subsequently allows them to fly to the food location. 

ABSTRACT:  Evidence for cognitive abilities in animals traditionally comes from controlled laboratory or field experiments, yet movement ecology offers a complementary approach to cognition. Movement-derived measures can capture, at broader spatial and temporal scales, the cognitive sophistication underlying decision-making in an animal’s natural habitat by characterising the spatial organisation of movement patterns. Integrating cognitive traits assessed through experimental assays with movement measures complements the species’ cognitive profile and ultimately improves our ability to compare cognition across taxa. Here, we combine laboratory-based cognitive assays with high-resolution movement tracking to investigate interspecific differences in wild ravens of the Judean Desert. We trapped and tested two sympatric species, the Fan-tailed raven (Corvus rhipidurus, N = 54) and the Brown-necked raven (C. ruficollis, N = 30), in a series of experiments targeting spatial memory, inhibitory control, and innovation. Following testing, individuals were fitted with GPS loggers that recorded their movements for 1–7 years after release. We estimated two movement-derived cognitive proxies: (1) the entropy of movement patterns, calculated for both stopping locations and the network of connections between them, to assess the tendency to forage opportunistically versus reusing previously learned routes; and (2) anomalous diffusion exponents, to evaluate the degree of goal-directed travel. We found that Fan-tailed ravens outperformed Brown-necked ravens across all laboratory tasks. In the wild, Fan-tailed ravens exhibited consistently higher routines (i.e., lower movement entropy) and higher anomalous diffusion values than Brown-necked ravens. Thus, not only did Fan-tailed ravens show enhanced spatial memory, inhibitory control, and innovation capacity, but they also exhibited highly organised movement patterns that relied on learnt routes and goal-directed movement. Our results suggest that cognitive traits measured under controlled conditions may be reflected in large-scale movement strategies in natural settings. Applying this framework to many species offers the potential to disentangle the role of ecological and social pressures shaping cognition, complementing previous cross-species comparative efforts.

ABSTRACT:  From approaching social partners to foraging, the fitness of newborn animals depends on their ability to make adaptive decisions in the absence of prior experience.  Without prior experience, decisions can be guided by biases shaped through evolution. However, innate biases expose individuals to the risk of errors. For instance, a newborn with an innate preference for maternal colours may mistakenly approach irrelevant, misleading objects that share the features of target stimuli. Which strategies newborn animals use to minimise the costs of biases remains unknown. To address this issue, we modelled a self-supervised strategy where inexperienced animals leverage their innate internal biases and the information present in the environment to maximise adaptive choices. Our model provides a set of testable predictions. First, innate biases tend to focus on cues that are rare in the background but frequent in the target stimuli (e.g., red colour), thus reducing false positives. Second, the evolution of multiple biases enables animals to benefit from the presence of co-occurring cues (e.g., the co-occurrence of red colour, movement against gravity and a face-like pattern present in the mother hen) for more robust identification of relevant stimuli. This combination supports the emergence of weak biases, whose weakness reduces the risk of wrong choices for single stimuli that partially resemble target objects. The presence of multiple weak biases is particularly advantageous in complex environments where multiple stimuli are present. Overall, a strategy that requires the simultaneous co-occurrence of independent and rare stimuli can explain the occurrence of multiple weak biases observed in newborn animals. This simple self-supervised strategy can support effective choices in both biological and artificial minds, with applications from animal cognition to developmental psychology and artificial intelligence.

ABSTRACT:  To interact effectively with the world, animals coordinate how they move and how they sense. In natural settings, perceivers adapt their motor-sensory strategies as they approach and explore objects, dynamically shaping both the generation and interpretation of sensory cues. Previous attempts to explain this process by reducing it solely to neuronal representations have failed to capture the mechanisms underlying dynamic perception. In this study, we use precise behavioral tracking to investigate the initial phase of object interaction, asking: What motor-sensory strategies support object detection during natural approach? Using high-resolution video tracking of rats freely exploring objects, we analyzed how head and whisker dynamics evolved across sequential contacts. We found that whisker-object interactions converged toward a distinct line, termed α^*, within a motor-sensory plane, where small changes in voluntary whisker movements produced large changes in whisker curvature, a sensory correlate of contact force. This convergence was actively controlled, predicted head movements and marked the completion of object approach. Convergence to α^* was consistent across different objects, suggesting it serves as an invariant motor-sensory contingency for object detection. Finally, proximity to α^* predicted the emergence of touch-induced pumps, a rapid motor-sensory reflex that further facilitated convergence within a single whisking cycle. Together, these results reveal that object detection is a closed-loop dynamic process, in which animals actively steer motor-sensory dynamics toward a robust detection-optimized state. More broadly, our findings exemplify how animals converge to embodied percepts through closed-loop coupling of movement and sensing; this principle holds clear implications for designing adaptive and robust artificial systems.

ABSTRACT:  Reinforcement learning (RL) is thought to underlie the acquisition of vocal skills like birdsong and speech, where sounding like one’s “tutor” is rewarding. However, what RL strategy generates the rich sound inventories for song or speech? To answer this question, we developed an actor-critic model of birdsong learning to explain juvenile zebra finches’ efficient learning of multiple syllables. We replace a single actor with multiple independent actors that jointly maximize a common intrinsic reward and accurately reproduce the birds’ empirical learning trajectories. We propose that our competitive-cooperative multi-actor RL (MARL) algorithm is key for the efficient learning of the action inventory of complex skills and lay the foundation for developing efficient learning algorithms in biorobotics and AI.

ABSTRACT:  This paper posits reafferent sensing as a minimal viable process that allows biological agents to successfully exploit causally relevant relations in a noisy, complex environment. We argue that most current approaches to causal learning, both in biology / comparative cognition as well as in classical reinforcement learning (RL) frameworks, lack explanation of the ability of the agents to exploit causal structure of their environment, and instead begin with the agent already capable of instrumental causal interaction to further explain causal learning and reasoning abilities. We argue that reafference enables successful navigation and exploitation of causal structures, acting as a precursor to more cognitively advanced causal learning abilities. We further propose that reafferent mechanisms can be explicitly modelled in RL, to advance a more biologically plausible model of embodied, instrumental causal interaction with the world.

ABSTRACT:  Why are brains the way they are? Are their circuit architectures and synaptic plasticity rules in some sense ‘optimal’? If so, in what sense, or in what contexts? We address these questions using olfactory associative memory in the fruit fly Drosophila. Flies can learn to associate a particular odour with a reward (e.g., food) or punishment (e.g., shock) and thereafter approach or avoid the trained odour. These associative memories are stored in Kenyon cells in the mushroom body, by weakening synapses from odour-responsive Kenyon cells onto mushroom body output neurons (MBONs) that lead to incorrect actions (e.g., odour+punishment weakens KC->Approach synapses). Why weaken incorrect actions rather than strengthening correct actions? Notably, synaptic depression is also used for learning in the vertebrate cerebellum, which has a remarkably similar architecture to the insect mushroom body, suggesting that using depression may be functionally advantageous.

We show both analytically and using simulations that depression outperforms potentiation for discriminating odours with overlapping KC representations, under a particular condition: if behaviour depends on the relative, not the absolute, difference between Avoid vs. Approach MBON activities. To test whether behaviour depends on the relative difference, we measured aversive learning for a range of odour concentrations and punishment intensities, in an individual-fly T-maze (n=3303 flies). We automatically tracked the flies’ decisions to enter or leave the side with the punished odour, and from the statistical distributions of these stochastic decisions, we inferred the mean and variance of the flies’ underlying preference for/against the odour. Bayesian modelling indicated that the data best fit a model where behaviour depends on the relative, not the absolute, difference between Avoid and Approach. These results suggest that flies learn by synaptic depression because, in the mushroom body, it is computationally superior to synaptic potentiation. These results illustrate how quantitative analysis of natural behaviour illuminates neural mechanisms underlying learned decision-making.

ABSTRACT:  Shannon's information theory states that information rate increases in proportion to the logarithm of signal power (Watts). However, information transmission is costly, so should neurons maximise information rate (bits/s), or information rate per Watt (energy efficiency, bits/J)? In practice, neurons incur two types of energy costs: fixed (e.g. infrastructure) and variable (e.g. firing rate). These fixed costs guarantee that energy efficiency increases only up to a threshold value of signal power, and then decreases. This pattern of results applies to a variety of physiological quantities: synaptic conductance, spike rate, axon diameter, the distribution (pdf) of axon diameters, and the layout of ganglion cell receptive fields on the retina. Here, we show that these disparate findings are predicted by a single equation (which is normally associated with foraging behaviour): the marginal value theorem (MVT).

ABSTRACT:  While we have a growing understanding of the environmental features bees use for navigation, far less is known about the fine-scale motor patterns that underpin their route-following during long-distance flights. In particular, the altitude bees maintain at the landscape scale, and the consistency of their trajectories across repeated trips, remain poorly understood. Resolving these behaviours at a fine scale is critical for testing predictions about the behavioural strategies bees use to navigate.

Here, we tracked honeybee flights to and from a feeder located up to 350 m from the hive using a drone-based tracking system capable of capturing bee trajectories with millisecond temporal resolution and centimetre-level spatial precision. This approach enables the reconstruction of detailed, continuous flight paths over landscape scales. Our preliminary results show that individual bees tended to follow similar routes on repeated return flights, yet routes differed markedly between individuals from the same hive. Outbound and inbound flights also differed both within and across bees. Notably, landscape features such as field boundaries often coincided with changes in flight characteristics, including speed and altitude.

Our ongoing work aims to determine whether bees exhibit structured oscillatory behaviors analogous to those observed in walking insects, or theoretically proposed for measuring flight altitude, and whether such patterns encode navigational uncertainty or serve to optimize visual sampling in complex environments.