Dark, Beyond Deep: A Paradigm Shift to Cognitive AI with Humanlike Common Sense

Abstract

Recent progress in deep learning is essentially based on a "big data for small tasks" paradigm, under which massive amounts of data are used to train a classifier for a single narrow task. In this paper, we call for a shift that flips this paradigm upside down. Specifically, we propose a "small data for big tasks" paradigm, wherein a single artificial intelligence (AI) system is challenged to develop "common sense", enabling it to solve a wide range of tasks with little training data. We illustrate the potential power of this new paradigm by reviewing models of common sense that synthesize recent breakthroughs in both machine and human vision. We identify functionality, physics, intent, causality, and utility (FPICU) as the five core domains of cognitive AI with humanlike common sense. When taken as a unified concept, FPICU is concerned with the questions of "why" and "how", beyond the dominant "what" and "where" framework for understanding vision. They are invisible in terms of pixels but nevertheless drive the creation, maintenance, and development of visual scenes. We therefore coin them the "dark matter" of vision. Just as our universe cannot be understood by merely studying observable matter, we argue that vision cannot be understood without studying FPICU. We demonstrate the power of this perspective to develop cognitive AI systems with humanlike common sense by showing how to observe and apply FPICU with little training data to solve a wide range of challenging tasks, including tool use, planning, utility inference, and social learning. In summary, we argue that the next generation of AI must embrace "dark" humanlike common sense for solving novel tasks.

arXiv commentFor high quality figures, please refer to http://wellyzhang.github.io/attach/dark.pdf

References in corpus (37)

• Efficient Estimation of Word Representations in Vector Space
• Distributed Representations of Words and Phrases and their Compositionality
• Playing Atari with Deep Reinforcement Learning
• Trust Region Policy Optimization
• ShapeNet: An Information-Rich 3D Model Repository
• End-to-End Training of Deep Visuomotor Policies
• VQA: Visual Question Answering
• Learning to Communicate with Deep Multi-Agent Reinforcement Learning
• Data-Driven Grasp Synthesis - A Survey
• Counterfactual Multi-Agent Policy Gradients
• Stabilising Experience Replay for Deep Multi-Agent Reinforcement Learning
• AI2-THOR: An Interactive 3D Environment for Visual AI
• Learning Multiagent Communication with Backpropagation
• Building Generalizable Agents with a Realistic and Rich 3D Environment
• A Study on Overfitting in Deep Reinforcement Learning
• Emergence of Grounded Compositional Language in Multi-Agent Populations
• Causal Discovery with Continuous Additive Noise Models
• Formalizing Neurath's Ship: Approximate Algorithms for Online Causal Learning
• MINOS: Multimodal Indoor Simulator for Navigation in Complex Environments
• Schema Networks: Zero-shot Transfer with a Generative Causal Model of Intuitive Physics
• Configurable 3D Scene Synthesis and 2D Image Rendering with Per-Pixel Ground Truth using Stochastic Grammars
• HoME: a Household Multimodal Environment
• Learning Object Arrangements in 3D Scenes using Human Context
• Emergence of Language with Multi-agent Games: Learning to Communicate with Sequences of Symbols
• Visual Affordance and Function Understanding: A Survey
• Learning Perceptual Inference by Contrasting
• VRKitchen: an Interactive 3D Virtual Environment for Task-oriented Learning
• RAVEN: A Dataset for Relational and Analogical Visual rEasoNing
• Emergence of Linguistic Communication from Referential Games with Symbolic and Pixel Input
• Emergent Communication in a Multi-Modal, Multi-Step Referential Game
• Understanding Human Gaze Communication by Spatio-Temporal Graph Reasoning
• Learning Social Affordance for Human-Robot Interaction
• PerspectiveNet: 3D Object Detection from a Single RGB Image via Perspective Points
• Holistic++ Scene Understanding: Single-view 3D Holistic Scene Parsing and Human Pose Estimation with Human-Object Interaction and Physical Commonsense
• Learning Grid Cells as Vector Representation of Self-Position Coupled with Matrix Representation of Self-Motion
• Cooperative Holistic Scene Understanding: Unifying 3D Object, Layout, and Camera Pose Estimation
• Theory-based Causal Transfer: Integrating Instance-level Induction and Abstract-level Structure Learning

Cited by in corpus (17)

• Deep Sequential Feature Learning in Clinical Image Classification of Infectious Keratitis
• A Survey of Embodied AI: From Simulators to Research Tasks
• A Review of Emerging Research Directions in Abstract Visual Reasoning
• Tac-Man: Tactile-Informed Prior-Free Manipulation of Articulated Objects
• Instance-wise or Class-wise? A Tale of Neighbor Shapley for Concept-based Explanation
• Deep Learning Methods for Abstract Visual Reasoning: A Survey on Raven's Progressive Matrices
• HALMA: Humanlike Abstraction Learning Meets Affordance in Rapid Problem Solving
• Congestion-aware Multi-agent Trajectory Prediction for Collision Avoidance
• Learning Low-Level Causal Relations using a Simulated Robotic Arm
• Memory Efficient Class-Incremental Learning for Image Classification
• Abstract Spatial-Temporal Reasoning via Probabilistic Abduction and Execution
• Human-Robot Interaction in a Shared Augmented Reality Workspace
• Learning by Fixing: Solving Math Word Problems with Weak Supervision
• Using Human-Guided Causal Knowledge for More Generalized Robot Task Planning
• Learning Triadic Belief Dynamics in Nonverbal Communication from Videos
• ACRE: Abstract Causal REasoning Beyond Covariation
• IntentQA: Intent Question Answering in Videos by Cognitive Context Reasoning