Editorial: AI and neuroscience: integrating knowledge, reasoning, and theory of mind
Open Access
- 19 May 2026
- journal article
- editorial
- Published by Frontiers Media SA in Frontiers in Computational Neuroscience
- Vol. 20, 1859797
Abstract
Artificial intelligence and neuroscience have been converging on a shared problem about how to explain and ultimately build systems (biological or synthetic) that can acquire knowledge from experience, reason under uncertainty, and coordinate perspectives from self-model representations to other minds (Hassabis et al., 2017;Langley et al., 2022;Limanowski & Blankenburg, 2013;Nawaz et al., 2025). The "AI and Neuroscience: Integrating Knowledge, Reasoning, and Theory of Mind" theme captures this convergence by explicitly highlighting a broad body of research linking accounts of neural information processing with computational architectures that can learn, generalize, and remain interpretable. At a broad level, the contributions in this collection can be read as collectively operating across three complementary levels. First, they investigate biological substrates of computation and the representational constraints that come with real neural tissue. Second, they advance architectures and modeling frameworks that treat cognition as an evolving repertoire of learned competencies rather than a set of isolated tasks. Third, they address human-AI coupling, i.e., how AI systems can extend cognition without displacing the very internal knowledge structures that make reasoning and perspective-taking possible in the first place.Edwards (Edwards, 2025) extends the functional contextual N-Frame model (Edwards, 2023(Edwards, , 2024) ) as an integrative theory linking predictive coding (Friston, 2018;Spratling, 2017), QBism-style observer dependence (Fuchs, 2017(Fuchs, , 2023;;Fuchs et al., 2014), and evolutionary dynamics to explain how belief updating and decision-making in humans (and potentially AI) can be modeled within a quantum-cognitive functional contextual formalism. The paper positions consciousness as an active, context-sensitive participant in "actualization" (bridging quantum potentiality to classical outcomes), uses cognitive fallacies (e.g., conjunction effects) as structured signatures of quantum context-dependent inference rather than mere cognitive classical irrationality, and proposes testable boundaries/parameters relevant to AI consciousness beyond standard AI benchmark performance. In the Research Topic's terms, it offers a unifying scaffold for knowledge (context-shaped representations), reasoning (state updates under constraints), and the foundations needed for theory of mind (self-referential observer modeling), while explicitly connecting these claims to experimentally oriented predictions and formal operator-style descriptions, in order to potentially develop safe and ethical AI.Tütüncü & Gonzalez-Franco (Tütüncü & Gonzalez-Franco, 2025) argues that "algorithmic suffering" is best treated as a comparative neuroscientific lens rather than as a claim that machines literally feel pain. They identify parallels between human and AI cognition in reward prediction error, Bayesian belief updating, and risk anticipation, while emphasizing a decisive asymmetry: in humans, prediction errors can become globally integrated into a conscious self-model and experienced as threats to meaning and integrity, whereas in current AI they remain numerical updates without phenomenology. Framed within this Research Topic, the paper helps clarify where AI's performance (its observable behavior and outputs) may mimic cognition while still falling short of the selfhood, meaning, and experiential integration that would be required for genuinely human mind-like understanding.Küchler and colleagues (Küchler et al., 2025) provide a clear "wetware-to-logic" bridge by demonstrating that engineered in vitro neuronal networks with controlled topology can implement basic Boolean operations (including NAND/OR) using stimulation and readout on high-density microelectrode arrays. Importantly, the work goes beyond a proof-ofprinciple gate: it interrogates encoding and decoding choices (rate-based versus spike-timingbased schemes such as TTFS), highlighting how representational format constrains how information can be reliably encoded, transmitted, and thus computed. Such minimal networks may serve as building blocks for hybrid intelligence and biocomputing. In the context of this Research Topic, the paper grounds "knowledge and reasoning" in the physical realities of neural signalling and frames neural computation as an input-output mapping that can be characterized and engineered.Chowdhury and colleagues (Chowdhury et al., 2025) review the rapidly developing landscape of inner speech recognition (ISR), i.e., the decoding of covert/inner speech from neural signals, and position machine learning as the primary driver of progress across the ISR pipeline (signal acquisition, preprocessing, feature extraction, and classification). The review synthesizes cognitive models of inner speech alongside a comparative analysis of machine learning (ML) and deep learning (DL) approaches (e.g., support vector machines (SVMs) and random forests versus convolutional neural network (CNN-based models)), while emphasizing persistent barriers such as signal-to-noise limitations, inter-subject variability, interpretability, and challenges for real-time deployment. Conceptually, inner speech sits at the intersection of language, self-regulation, and planning; as such, ISR research provides a natural bridge toward higher-order cognitive functions (i.e., neural signals → inner speech decoding → internal cognition). In the context of this Research Topic, the paper can be read as exemplifying the "integrating knowledge and reasoning" theme as an applied neuroscience-to-AI translation problem with significant clinical and assistive-technology implications.Prudkov (Prudkov, 2025) advances a hypothesis-and-theory account of AGI grounded in the "goals-means correspondence," arguing that the central challenge for general intelligence is not merely producing competent outputs, but establishing...Keywords
This publication has 10 references indexed in Scilit:
- A review of neuro-symbolic AI integrating reasoning and learning for advanced cognitive systemsIntelligent Systems with Applications, 2025
- A functional contextual, observer-centric, quantum mechanical, and neuro-symbolic approach to solving the alignment problem of artificial general intelligence: safe AI through intersecting computational psychological neuroscience and LLM architecture for emergent theory of mindFrontiers in Computational Neuroscience, 2024
- Functional contextual implementation of an evolutionary, entropy-based, and embodied free energy framework: Utilizing Lagrangian mechanics and evolutionary game theory’s truth vs. fitness test of the veridicality of phenomenological experienceFrontiers in Psychology, 2023
- Theory of Mind and Preference Learning at the Interface of Cognitive Science, Neuroscience, and AI: A ReviewFrontiers in Artificial Intelligence, 2022
- Relational Density Theory: Nonlinearity of Equivalence Relating Examined through Higher-Order Volumetric-Mass-DensityPerspectives on Behavior Science, 2020
- Does predictive coding have a future?Nature Neuroscience, 2018
- Neuroscience-Inspired Artificial IntelligenceNeuron, 2017
- A review of predictive coding algorithmsBrain and Cognition, 2017
- An introduction to QBism with an application to the locality of quantum mechanicsAmerican Journal of Physics, 2014
- Minimal self-models and the free energy principleFrontiers in Human Neuroscience, 2013