Generating Meaning: Active Inference and the Scope and Limits of Passive AI

Published: June 8, 2023

Prominent accounts of sentient behaviour depict brains as generative models organismic interaction with the world, raising points contact current work in Generative AI. However, because they contend control purposive, life-maintaining sensorimotor interactions, living organisms are inextricably anchored to body and world. Unlike passive learnt by AIs, must capture sensory consequences action. This allows embodied agents intervene upon their worlds ways that constantly put best test; providing a solid bedrock is—we argue—essential development genuine understanding. Here, we review resulting implications, consider future directions travel for

Language: Английский

---

Active Vision in Binocular Depth Estimation: A Top-Down Perspective

Biomimetics, Journal Year: 2023, Volume and Issue: 8(5), P. 445 - 445

Published: Sept. 21, 2023

Depth estimation is an ill-posed problem; objects of different shapes or dimensions, even if at distances, may project to the same image on retina. Our brain uses several cues for depth estimation, including monocular such as motion parallax and binocular diplopia. However, it remains unclear how computations required are implemented in biologically plausible ways. State-of-the-art approaches based deep neural networks implicitly describe a hierarchical feature detector. Instead, this paper we propose alternative approach that casts problem active inference. We show can be inferred by inverting generative model simultaneously predicts eyes' projections from 2D belief over object. Model inversion consists series homogeneous transformations Predictive Coding principles. Under assumption nonuniform fovea resolution, favors vision strategy fixates object with eyes, rendering more accurate. This not realized first fixating target then estimating depth; instead, combines two processes through action-perception cycles, similar mechanism saccades during recognition. The proposed requires only local (top-down bottom-up) message passing, which circuits.

Language: Английский

---

Correspondence Learning between Morphologically Different Robots Via Task Demonstrations

IEEE Robotics and Automation Letters, Journal Year: 2024, Volume and Issue: 9(5), P. 4463 - 4470

Published: March 27, 2024

We observe a large variety of robots in terms their bodies, sensors, and actuators.Given the commonalities skill sets, teaching each to different robot independently is inefficient not scalable when robotic landscape considered.If we can learn correspondences between sensorimotor spaces robots, expect that learned one be more directly easily transferred other robots.In this paper, propose method among two or may have morphologies.To specific, besides with similar morphologies degrees freedom, show fixed-based manipulator joint control differential drive mobile addressed within proposed framework.To set up correspondence considered, an initial base task demonstrated achieve same goal.Then, common latent representation along individual policies for achieving goal.After learning stage, observation new execution by becomes sufficient generate space pertaining task.We verified our system experiments where (1) need follow paths task, (2) trajectories (3) complexities required are robots.We also provide proof-of-the-concept realization real simulated robot.

Language: Английский

---

Invariant representations in abstract concept grounding – the physical world in grounded cognition

Psychonomic Bulletin & Review, Journal Year: 2024, Volume and Issue: unknown

Published: May 28, 2024

Abstract Grounded cognition states that mental representations of concepts consist experiential aspects. For example, the concept “cup” consists sensorimotor experiences from interactions with cups. Typical modalities in which are grounded are: The system (including interoception), emotion, action, language, and social Here, we argue this list should be expanded to include physical invariants (unchanging features motion; e.g., gravity, momentum, friction). Research on reasoning consistently demonstrates represented as fundamentally other grounding substrates, therefore qualify. We assess several theories representation (simulation, conceptual metaphor, spaces, predictive processing) their positions invariants. find classic theories, simulation metaphor theory, have not considered invariants, while spaces processing have. conclude included into core mechanisms theory well suited do this. Furthermore, very promising also integrated future.

Language: Английский

---

Modeling Motor Control in Continuous Time Active Inference: A Survey

IEEE Transactions on Cognitive and Developmental Systems, Journal Year: 2023, Volume and Issue: 16(2), P. 485 - 500

Published: Dec. 4, 2023

The way the brain selects and controls actions is still widely debated. Mainstream approaches based on Optimal Control focus stimulus-response mappings that optimize cost functions. Ideomotor theory cybernetics propose a different perspective: they suggest are selected controlled by activating action effects continuously matching internal predictions with sensations. Active Inference offers modern formulation of these ideas, in terms inferential mechanisms prediction-error-based control, which can be linked to neural living organisms. This article provides technical illustration models continuous time brief survey solve four kinds control problems; namely, goal-directed reaching movements, active sensing, resolution multisensory conflict during movement integration decision-making motor control. Crucially, Inference, all facets emerge from same optimization process - minimization Free Energy do not require designing separate Therefore, unitary perspective various aspects inform both study biological design artificial robotic systems.

Language: Английский

---

Dual Process Theory for Large Language Models: An overview of using Psychology to address hallucination and reliability issues

Adaptive Behavior, Journal Year: 2023, Volume and Issue: 32(4), P. 329 - 343

Published: Oct. 23, 2023

State-of-the-art Large Language Models have recently exhibited extraordinary linguistic abilities which surprisingly extended to reasoning. However, responses that are unreliable, false, or invented still a frequent issue. It has been argued scaling up strategies, as in increasing model size hardware power, might not be enough resolve the Recent research implemented Type 2 strategies (such Chain-of-Thought and Tree-of-Thought), mimic reasoning, from Dual Process Theory, interact with for improved results. The current paper reviews these light of Predicting Reflecting Framework understanding Theory suggests what Psychology, drawing executive functions, thinking disposition creativity, can further contribute possible implementations address hallucination reliability issues.

Language: Английский

---

Collective Predictive Coding Hypothesis: Symbol Emergence as Decentralized Bayesian Inference

Published: Aug. 15, 2023

Understanding the emergence of symbol systems, especially language, requires construction a computational model that reproduces both developmental learning process in everyday life and evolutionary dynamics throughout history. This study introduces collective predictive coding (CPC) hypothesis, which emphasizes models interdependence between forming internal representations through physical interactions with environment sharing utilizing meanings social semiotic within system. The total system is theorized from perspective {\it coding}. hypothesis draws inspiration studies grounded probabilistic generative language games, including Metropolis--Hastings naming game. Thus, playing such games among agents distributed manner can be interpreted as decentralized Bayesian inference shared by multi-agent Moreover, this explores potential link CPC free-energy principle, positing adheres to society-wide principle. Furthermore, paper provides new explanation for why large appear possess knowledge about world based on experience, even though they have neither sensory organs nor bodies.This reviews past approaches offers comprehensive survey related prior studies, presents discussion CPC-based generalizations. Future challenges cross-disciplinary research avenues are highlighted.

Language: Английский

---

Enhancing Robotic Perception through Synchronized Simulation and Physical Common-Sense Reasoning

Sensors, Journal Year: 2024, Volume and Issue: 24(7), P. 2249 - 2249

Published: March 31, 2024

We introduce both conceptual and empirical findings arising from the amalgamation of a robotics cognitive architecture with an embedded physics simulator, aligning principles outlined in intuitive literature. The employed robotic architecture, named CORTEX, leverages highly efficient distributed working memory known as deep state representation. This inherently encompasses fundamental ontology, persistency, geometric logical relationships among elements, tools for reading, updating, reasoning about its contents. Our primary objective is to investigate hypothesis that integration simulator into streamlines implementation various functionalities would otherwise necessitate extensive coding debugging efforts. Furthermore, we categorize these enhanced broad types based on nature problems they address. These include addressing challenges related occlusion, model-based perception, self-calibration, scene structural stability, human activity interpretation. To demonstrate outcomes our experiments, employ CoppeliaSim Kinova Gen3 arm Open-Manipulator-P real-world scenarios. Synchronization maintained between stream real events. Depending ongoing task, numerous queries are computed, results projected memory. Participating agents can then leverage this information enhance overall performance.

Language: Английский

---

Invariant Representations in Abstract Concept Grounding - The Physical World in Grounded Cognition

Published: April 27, 2024

Grounded cognition states that mental representations of concepts consist experiential aspects. For example, the concept ‘cup’ consists sensorimotor experiences from interactions with cups. Typical modalities in which are grounded are: The system (incl. interoception), emotion, action, language, and social Here, we argue this list should be expanded to include physical invariants (unchanging features motion; e.g., gravity, momentum, friction). Research on reasoning consistently demonstrates represented as fundamentally other grounding substrates, therefore qualify. We assess several theories representation (simulation, conceptual metaphor, spaces, predictive processing) their positions invariants. find classic theories, simulation metaphor theory, have not considered invariants, while spaces processing have. conclude included into core mechanisms theory well-suited do this. Meanwhile very promising also integrated future.

Language: Английский

---

Implementation of Engagement Detection for Human–Robot Interaction in Complex Environments

Sensors, Journal Year: 2024, Volume and Issue: 24(11), P. 3311 - 3311

Published: May 22, 2024

This study develops a comprehensive robotic system, termed the robot cognitive for complex environments, integrating three models: engagement model, intention and human–robot interaction (HRI) model. The system aims to enhance naturalness comfort of HRI by enabling robots detect human behaviors, intentions, emotions accurately. A novel dual-arm-hand mobile robot, Mobi, was designed demonstrate system’s efficacy. model utilizes eye gaze, head pose, action recognition determine suitable moment initiation, addressing potential contact anxiety. employs sentiment analysis emotion classification infer interactor’s intentions. integrated with Google Dialogflow, facilitates appropriate responses based on user feedback. performance validated in retail environment scenario, demonstrating its improve experience HRIs.

Language: Английский

---