The focus of this talk is not about my research in AI or Robotics but my own journey on trying to do research and understand intelligence in a rapidly evolving research landscape. I will trace my path from conducting early-stage research during graduate school, to working on practical solutions within a startup environment, and finally to my current role where I participate in more structured research at a major tech company. Through these varied experiences, I will provide different perspectives on research and talk about how my core beliefs on intelligence have changed and sometimes even been compromised. There are no lessons to be learned from my stories, but hopefully they will be entertaining.

Understanding the neural mechanisms of reward-guided learning is a long-standing goal of computational neuroscience. Recent methodological innovations enable us to collect ever larger neural and behavioral datasets. This presents opportunities to achieve greater understanding of learning in the brain at scale, as well as methodological challenges. In the first part of the talk, I will discuss our recent insights into the mechanisms by which zebra finch songbirds learn to sing. Dopamine has been long thought to guide reward-based trial-and-error learning by encoding reward prediction errors. However, it is unknown whether the learning of natural behaviours, such as developmental vocal learning, occurs through dopamine-based reinforcement. Longitudinal recordings of dopamine and bird songs reveal that dopamine activity is indeed consistent with encoding a reward prediction error during naturalistic learning. In the second part of the talk, I will talk about recent work we are doing at DeepMind to develop tools for automatically discovering interpretable models of behavior directly from animal choice data. Our method, dubbed CogFunSearch, uses LLMs within an evolutionary search process in order to "discover" novel models in the form of Python programs that excel at accurately predicting animal behavior during reward-guided learning. The discovered programs reveal novel patterns of learning and choice behavior that update our understanding of how the brain solves reinforcement learning problems.

Digital platforms generate unprecedented traces of human behaviour, offering new methodological approaches to understanding collective action, polarisation, and social dynamics. Through analysis of millions of digital traces across multiple studies, we demonstrate how online behaviours predict offline action: Brexit-related tribal discourse responds to real-world events, machine learning models achieve 80% accuracy in predicting real-world protest attendance from digital signals, and social validation through "likes" emerges as a key driver of mobilization. Extending this approach to conspiracy narratives reveals how digital traces illuminate psychological mechanisms of belief and community formation. Longitudinal analysis of YouTube conspiracy content demonstrates how narratives systematically address existential, epistemic, and social needs, while examination of alt-tech platforms shows how emotions of anger, contempt, and disgust correlate with violence-legitimating discourse, with significant differences between narratives associated with offline violence versus peaceful communities. This work establishes digital traces as both methodological innovation and theoretical lens, demonstrating that computational social science can illuminate fundamental questions about polarisation, mobilisation, and collective behaviour across contexts from electoral politics to conspiracy communities.

Synthetic face datasets are increasingly used to overcome the limitations of real-world biometric data, including privacy concerns, demographic imbalance, and high collection costs. However, many existing methods lack fine-grained control over identity attributes and fail to produce paired, identity-consistent images under structured capture conditions. In this talk, I will present FLUXSynID, a framework for generating high-resolution synthetic face datasets with user-defined identity attribute distributions and paired document-style and trusted live capture images. The dataset generated using FLUXSynID shows improved alignment with real-world identity distributions and greater diversity compared to prior work. I will also discuss how FLUXSynID’s dataset and generation tools can support research in face recognition and morphing attack detection (MAD), enhancing model robustness in both academic and practical applications.

In a first part of the talk, I will present Predictive Coding Light (PCL), a novel unsupervised learning architecture for spiking neural networks. In contrast to conventional predictive coding approaches, which only transmit prediction errors to higher processing stages, PCL learns inhibitory lateral and top-down connectivity to suppress the most predictable spikes and passes a compressed representation of the input to higher processing stages. We show that PCL reproduces a range of biological findings and exhibits a favorable tradeoff between energy consumption and downstream classification performance on challenging benchmarks. A second part of the talk will feature our lab’s efforts to explain how infants and toddlers might learn abstract object representations without supervision. I will present deep learning models that exploit the temporal and multimodal structure of their sensory inputs to learn representations of individual objects, object categories, or abstract super-categories such as „kitchen object“ in a fully unsupervised fashion. These models offer a parsimonious account of how abstract semantic knowledge may be rooted in children's embodied first-person experiences.

We developed a novel paradigm measuring implicit identity recognition using Fast Periodic Visual Stimulation (FPVS) with EEG among 16 students and 12 police officers with normal face processing abilities. Participants' neural responses to a 1-Hz tagged oddball identity embedded within a 6-Hz image stream revealed implicit recognition with high-quality mugshots but not CCTV-like images, suggesting optimal resolution requirements. Our findings extend previous research by demonstrating that even unfamiliar identities can elicit robust neural recognition signatures through brief, repeated passive exposure. This approach offers potential for objective validation of face processing abilities in forensic applications, including assessment of facial examiners, Super-Recognisers, and eyewitnesses, potentially overcoming limitations of traditional behavioral assessment methods.

Recent advances in artificial intelligence have led to new forms of misinformation, including highly realistic “deepfake” synthetic media. We conduct three experiments to investigate how and why retail investors react to deepfake financial news. Results from the first two experiments provide evidence that investors use a “realism heuristic,” responding more intensely to audio and video deepfakes as their perceptual realism increases. In the third experiment, we introduce an intervention to prompt analytical thinking, varying whether participants make analytical judgments about credibility or intuitive investment judgments. When making intuitive investment judgments, investors are strongly influenced by both more and less realistic deepfakes. When making analytical credibility judgments, investors are able to discern the non-credibility of less realistic deepfakes but struggle with more realistic deepfakes. Thus, while analytical thinking can reduce the impact of less realistic deepfakes, highly realistic deepfakes are able to overcome this analytical scrutiny. Our results suggest that deepfake financial news poses novel threats to investors.

The human medial temporal lobe contains neurons that respond selectively to the semantic contents of a presented stimulus. These "concept cells" may respond to very different pictures of a given person and even to their written or spoken name. Their response latency is far longer than necessary for object recognition, they follow subjective, conscious perception, and they are found in brain regions that are crucial for declarative memory formation. It has thus been hypothesized that they may represent the semantic "building blocks" of episodic memories. In this talk I will present data from single unit recordings in the hippocampus, entorhinal cortex, parahippocampal cortex, and amygdala during paradigms involving object recognition and conscious perception as well as encoding of episodic memories in order to characterize the role of concept cells in these cognitive functions.

Fast periodic visual stimulation (FPVS) has emerged as a promising tool for assessing cognitive function in individuals with dementia. This technique leverages electroencephalography (EEG) to measure brain responses to rapidly presented visual stimuli, offering a non-invasive and objective method for evaluating a range of cognitive functions. Unlike traditional cognitive assessments, FPVS does not rely on behavioural responses, making it particularly suitable for individuals with cognitive impairment. In this talk I will highlight a series of studies that have demonstrated its ability to detect subtle deficits in recognition memory, visual processing and attention in dementia patients using EEG in the lab, at home and in clinic. The method is quick, cost-effective, and scalable, utilizing widely available EEG technology. FPVS holds significant potential as a functional biomarker for early diagnosis and monitoring of dementia, paving the way for timely interventions and improved patient outcomes.

Facial expressions are inherently dynamic, and our visual system is sensitive to subtle changes in their temporal sequence. However, researchers often use dynamic morphs of photographs—simplified, linear representations of motion—to study the neural correlates of dynamic face perception. To explore the brain's sensitivity to natural facial motion, we constructed a novel dynamic face database using generative neural networks, trained on a verified set of video-recorded emotional expressions. The resulting deepfakes, consciously indistinguishable from videos, enabled us to separate biological motion from photorealistic form. Results showed that conventional dynamic morphs elicit distinct responses in the brain compared to videos and photos, suggesting they violate expectations (n400) and have reduced social salience (late positive potential). This suggests that dynamic morphs misrepresent facial dynamism, resulting in misleading insights about the neural and behavioural correlates of face perception. Deepfakes and videos elicited largely similar neural responses, suggesting they could be used as a proxy for real faces in vision research, where video recordings cannot be experimentally manipulated. And yet, despite being consciously undetectable as fake, deepfakes elicited an expectation violation response in the brain. This points to a neural sensitivity to naturalistic facial motion, beyond conscious awareness. Despite some differences in neural responses, the realism and manipulability of deepfakes make them a valuable asset for research where videos are unfeasible. Using these stimuli, we proposed a novel marker for the conscious perception of naturalistic facial motion – Frontal delta activity – which was elevated for videos and deepfakes, but not for photos or dynamic morphs.

Photoreceptor loss is the primary cause behind vision impairment and blindness in diseases such as retinitis pigmentosa and age-related macular degeneration. However, the death of rods and cones allows retinoids to permeate the inner retina, causing retinal ganglion cells to become spontaneously hyperactive, severely reducing the signal-to-noise ratio, and creating interference in the communication between the surviving retina and the brain. Treatments aimed at blocking or reducing hyperactivity improve vision initiated from surviving photoreceptors and could enhance the signal fidelity generated by vision restoration methodologies.