Is traditional music theory relevant to recent popular music, and how should it evolve? Traditional music theory assumes fixed scales, predefined tuning, and the clear transmission of well-defined pitches. Modern popular music challenges these assumptions, incorporating alternative tuning frameworks and uncertainty in pitch perception. Inspired by Cartesian doubt and phenomenology, the presentation advocates suspending biases, prioritising observation, and generalising cautiously. Pitch perception is more complex than commonly assumed. Similarly to how a physical line’s dimension can exceed one, tones may resist classification as a single pitch. Modern popular music follows principles that differ from those that historically shaped scale construction. A new theoretical framework suggests that the tones themselves can define the tuning system rather than conforming to a predetermined structure. Another hypothesis proposes that repetition in music takes advantage of multi-stable pitch percepts. A more precise understanding of musical evolution prevents misinterpretations and false conclusions. It enhances music technology and AI applications while fostering creative exploration beyond classical constraints. Western classical music theory is a specific system, not a universal truth. Understanding modern popular music requires observing how it develops rather than imposing outdated frameworks.

Soil is a complex, dynamic material, with physical properties that depend on its biological content. We propose a cellular automaton model for self-organizing soil structure, where soil aggregates and serves as food for microbial species. These, in turn, produce nutrients that facilitate self-amplification, establishing a cyclical dynamic of consumption and regeneration. Our model explores the spatial interactions between these components and their role in sustaining a balanced ecosystem. The main results demonstrate that (1) spatial structure supports a stable living state, preventing population collapse or uncontrolled growth; (2) the spatial model allows for the coexistence of parasitic species, which exploit parts of the system without driving it to extinction; and (3) optimal growth conditions for microbes are associated to diverse length scales in the soil structure, suggesting that heterogeneity is key to ecosystem resilience. These findings highlight the importance of spatio-temporal dynamics of life in soil ecology.

Recent efforts have augmented Large Language Models (LLMs) with external resources (e.g., the Internet) or internal control flows (e.g., prompt chaining) for tasks requiring grounding or reasoning, creating a new class of “Language Agents”. While these agents have achieved substantial empirical success, we presently lack organizing principles to measure progress or guide further development. We draw on the rich history of cognitive science and symbolic artificial intelligence to propose ‘Cognitive Architectures for Language Agents’ – CoALA. CoALA describes a language agent with modular memory components, a structured action space to interact with internal memory and external environments, and a generalized decision-making process to choose actions. We use CoALA to survey and organize a large body of recent work, then identify actionable directions towards more capable agents.

Biological neurons use short stereotyped electrical impulses called “spikes” to compute and transmit information. The spike times, in addition to the spike rates, are known to play an important role in how neurons process information. SNNs are thus more biologically realistic than the real-valued artificial neural networks used in deep learning, and are arguably the most viable option if one wants to understand how the brain computes at the neuronal description level. But SNNs are also appealing for AI technology, because they can be implemented efficiently on low-power neuromorphic chips. SNNs have been studied for several decades, yet interest in them has surged recently due to a major breakthrough: surrogate gradient learning (SGL). SGL allows the training of SNNs with backpropagation, just like real-valued networks, outperforming other training methods by a large margin. In my group, we have demonstrated that SGL enables the learning of not only connection weights but also connection delays. These delays represent the time needed for one spike to travel from the emitting to the receiving neurons. Delays matter because they shift the spike arrival times, which should be synchronous to trigger an output spike. Thus plastic delays significantly enhance the expressivity of SNNs. If this fact is well established theoretically, efficient algorithms to learn these delays have been lacking. We proposed one such algorithm, based on a new convolution method known as dilated convolution with learnable spacings (DCLS), which outperforms previous proposals. Our results show that learning delays with DCLS in fully connected and convolutional SNNs strongly increases the accuracy on several temporal vision and audition tasks, leading to new state-of-the-arts. Our approach is relevant for the neuromorphic engineering community, as most existing neuromorphic chips support programmable delays. In the long term, our research could also provide insights into the role of myelin plasticity in the brain.