Commonsense Reasoning Research Articles

A noise audit of human-labeled benchmarks for machine commonsense reasoning

With the advent of large language models, evaluating and benchmarking these systems on important AI problems has taken on newfound importance. Such benchmarking typically involves comparing the predictions of a system against human labels (or a single ‘ground-truth’). However, much recent work in psychology has suggested that most tasks involving significant human judgment can have non-trivial degrees of noise. In his book, Kahneman suggests that noise may be a much more significant component of inaccuracy compared to bias, which has been studied more extensively in the AI community. This article proposes a detailed noise audit of human-labeled benchmarks in machine commonsense reasoning, an important current area of AI research. We conduct noise audits under two important experimental conditions: one in a smaller-scale but higher-quality labeling setting, and another in a larger-scale, more realistic online crowdsourced setting. Using Kahneman’s framework of noise, our results consistently show non-trivial amounts of level, pattern, and system noise, even in the higher-quality setting, with comparable results in the crowdsourced setting. We find that noise can significantly influence the performance estimates that we obtain of commonsense reasoning systems, even if the ‘system’ is a human; in some cases, by almost 10 percent. Labeling noise also affects performance estimates of systems like ChatGPT by more than 4 percent. Our results suggest that the default practice in the AI community of assuming and using a ‘single’ ground-truth, even on problems requiring seemingly straightforward human judgment, may warrant empirical and methodological re-visiting.

Overview of knowledge reasoning for knowledge graph

Knowledge graphs are large-scale semantic networks that considerably impact knowledge representation. Mining hidden knowledge from existing data, including triplet knowledge reasoning, is a primary objective of knowledge graphs. With the development of Neural Network (NN) and Deep Learning (DL), the interpretability of triplet knowledge reasoning gradually decreases; furthermore, what machines learn is not actual reasoning but digital reasoning shortcuts. To solve this problem, more background knowledge needs to be introduced into knowledge graphs: causal graphs can offer valuable causal logic knowledge for reasoning; temporal quadruples can provide essential temporal distribution details; and commonsense graphs can furnish pertinent commonsense understanding to support reasoning. In recent years, many scholars have incorporated additional background knowledge into knowledge graphs to construct more complex reasoning mechanisms. This paper reviews the basic concepts and definitions of knowledge reasoning and the reasoning methods used for knowledge graphs. Specifically, we dissect the reasoning methods into four categories: triplet reasoning, causal inference, temporal inference, and commonsense reasoning. Finally, we discuss the remaining challenges and research opportunities related to knowledge graph reasoning.

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

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

The Generalization and Robustness of Transformer-Based Language Models on Commonsense Reasoning

The advent of powerful transformer-based discriminative language models and, more recently, generative GPT-family models, has led to notable advancements in natural language processing (NLP), particularly in commonsense reasoning tasks. One such task is commonsense reasoning, where performance is usually evaluated through multiple-choice question-answering benchmarks. Till date, many such benchmarks have been proposed and `leaderboards' tracking state-of-the-art performance on those benchmarks suggest that transformer-based models are approaching human-like performance. However, due to documented problems such as hallucination and bias, the research focus is shifting from merely quantifying accuracy on the task to an in-depth, context-sensitive probing of LLMs' generalization and robustness. To gain deeper insight into diagnosing these models' performance in commonsense reasoning scenarios, this thesis addresses three main studies: the generalization ability of transformer-based language models on commonsense reasoning, the trend in confidence distribution of these language models confronted with ambiguous inference tasks, and a proposed risk-centric evaluation framework for both discriminative and generative language models.

CORECODE: A Common Sense Annotated Dialogue Dataset with Benchmark Tasks for Chinese Large Language Models

As an indispensable ingredient of intelligence, commonsense reasoning is crucial for large language models (LLMs) in real-world scenarios. In this paper, we propose CORECODE, a dataset that contains abundant commonsense knowledge manually annotated on dyadic dialogues, to evaluate the commonsense reasoning and commonsense conflict detection capabilities of Chinese LLMs. We categorize commonsense knowledge in everyday conversations into three dimensions: entity, event, and social interaction. For easy and consistent annotation, we standardize the form of commonsense knowledge annotation in open-domain dialogues as "domain: slot = value". A total of 9 domains and 37 slots are defined to capture diverse commonsense knowledge. With these pre-defined domains and slots, we collect 76,787 commonsense knowledge annotations from 19,700 dialogues through crowdsourcing. To evaluate and enhance the commonsense reasoning capability for LLMs on the curated dataset, we establish a series of dialogue-level reasoning and detection tasks, including commonsense knowledge filling, commonsense knowledge generation, commonsense conflict phrase detection, domain identification, slot identification, and event causal inference. A wide variety of existing open-source Chinese LLMs are evaluated with these tasks on our dataset. Experimental results demonstrate that these models are not competent to predict CORECODE's plentiful reasoning content, and even ChatGPT could only achieve 0.275 and 0.084 accuracy on the domain identification and slot identification tasks under the zero-shot setting. We release the data and codes of CORECODE at https://github.com/danshi777/CORECODE to promote commonsense reasoning evaluation and study of LLMs in the context of daily conversations.

Commonsense for Zero-Shot Natural Language Video Localization

Zero-shot Natural Language-Video Localization (NLVL) methods have exhibited promising results in training NLVL models exclusively with raw video data by dynamically generating video segments and pseudo-query annotations. However, existing pseudo-queries often lack grounding in the source video, resulting in unstructured and disjointed content. In this paper, we investigate the effectiveness of commonsense reasoning in zero-shot NLVL. Specifically, we present CORONET, a zero-shot NLVL framework that leverages commonsense to bridge the gap between videos and generated pseudo-queries via a commonsense enhancement module. CORONET employs Graph Convolution Networks (GCN) to encode commonsense information extracted from a knowledge graph, conditioned on the video, and cross-attention mechanisms to enhance the encoded video and pseudo-query representations prior to localization. Through empirical evaluations on two benchmark datasets, we demonstrate that CORONET surpasses both zero-shot and weakly supervised baselines, achieving improvements up to 32.13% across various recall thresholds and up to 6.33% in mIoU. These results underscore the significance of leveraging commonsense reasoning for zero-shot NLVL.

Graph Neural Prompting with Large Language Models

Large language models (LLMs) have shown remarkable generalization capability with exceptional performance in various language modeling tasks. However, they still exhibit inherent limitations in precisely capturing and returning grounded knowledge. While existing work has explored utilizing knowledge graphs (KGs) to enhance language modeling via joint training and customized model architectures, applying this to LLMs is problematic owing to their large number of parameters and high computational cost. Therefore, how to enhance pre-trained LLMs using grounded knowledge, e.g., retrieval-augmented generation, remains an open question. In this work, we propose Graph Neural Prompting (GNP), a novel plug-and-play method to assist pre-trained LLMs in learning beneficial knowledge from KGs. GNP encompasses various designs, including a standard graph neural network encoder, a cross-modality pooling module, a domain projector, and a self-supervised link prediction objective. Extensive experiments on multiple datasets demonstrate the superiority of GNP on both commonsense and biomedical reasoning tasks across different LLM sizes and settings. Code is available at https://github.com/meettyj/GNP.

TaskLAMA: Probing the Complex Task Understanding of Language Models

Structured Complex Task Decomposition (SCTD) is the problem of breaking down a complex real-world task (such as planning a wedding) into a directed acyclic graph over individual steps that contribute to achieving the task, with edges specifying temporal dependencies between steps. SCTD is an important component of assistive planning tools, and a challenge for commonsense reasoning systems. We probe how accurately SCTD can be done with the knowledge extracted from pre-trained Large Language Models (LLMs). We introduce a new high-quality human-annotated dataset for this problem and novel metrics to fairly assess performance of LLMs against several baselines. Our experiments reveal that LLMs are able to decompose complex tasks into individual steps effectively, with a relative improvement of 15% to 280% over the best baseline. We also propose a number of approaches to further improve their performance, with a relative improvement of 7% to 37%. However, we find that LLMs still struggle to predict pairwise temporal dependencies, which reveals a gap in their understanding of complex tasks.

Small Language Model Can Self-Correct

Generative Language Models (LMs) such as ChatGPT have exhibited remarkable performance across various downstream tasks. Nevertheless, one of their most prominent drawbacks is generating inaccurate or false information with a confident tone. Previous studies have devised sophisticated pipelines and prompts to induce large LMs to exhibit the capability for self-correction. However, large LMs are explicitly prompted to verify and modify their answers separately rather than completing all steps spontaneously like humans. Moreover, these complex prompts are extremely challenging for small LMs to follow. In this paper, we introduce the Intrinsic Self-Correction (ISC) in generative language models, aiming to correct the initial output of LMs in a self-triggered manner, even for those small LMs with 6 billion parameters. Specifically, we devise a pipeline for constructing self-correction data and propose Partial Answer Masking (PAM), aiming to endow the model with the capability for intrinsic self-correction through fine-tuning. We conduct experiments using LMs with parameters sizes ranging from 6 billion to 13 billion in two tasks, including commonsense reasoning and factual knowledge reasoning. Our experiments demonstrate that the outputs generated using ISC outperform those generated without self-correction. We believe that the output quality of even small LMs can be further improved by empowering them with the ability to intrinsic self-correct.

RETRACTED: McOmet: Multimodal Fusion Transformer for Physical Audiovisual Commonsense Reasoning

This article, which was published in Proceedings of the Thirty-Seventh AAAI Conference on Artificial Intelligence (AAAI 2023), has been retracted by agreement between the authors and the journal. This statement refers to RETRACTED: McOmet: Multimodal Fusion Transformer for Physical Audiovisual Commonsense Reasoning.

Sentiment Analysis in the Age of Generative AI

In the rapidly advancing age of Generative AI, Large Language Models (LLMs) such as ChatGPT stand at the forefront of disrupting marketing practice and research. This paper presents a comprehensive exploration of LLMs’ proficiency in sentiment analysis, a core task in marketing research for understanding consumer emotions, opinions, and perceptions. We benchmark the performance of three state-of-the-art LLMs, i.e., GPT-3.5, GPT-4, and Llama 2, against established, high-performing transfer learning models. Despite their zero-shot nature, our research reveals that LLMs can not only compete with but in some cases also surpass traditional transfer learning methods in terms of sentiment classification accuracy. We investigate the influence of textual data characteristics and analytical procedures on classification accuracy, shedding light on how data origin, text complexity, and prompting techniques impact LLM performance. We find that linguistic features such as the presence of lengthy, content-laden words improve classification performance, while other features such as single-sentence reviews and less structured social media text documents reduce performance. Further, we explore the explainability of sentiment classifications generated by LLMs. The findings indicate that LLMs, especially Llama 2, offer remarkable classification explanations, highlighting their advanced human-like reasoning capabilities. Collectively, this paper enriches the current understanding of sentiment analysis, providing valuable insights and guidance for the selection of suitable methods by marketing researchers and practitioners in the age of Generative AI.

Mathematics, word problems, common sense, and artificial intelligence

The paper discusses the capacities and limitations of current artificial intelligence (AI) technology to solve word problems that combine elementary mathematics with commonsense reasoning. No existing AI systems can solve these reliably. We review three approaches that have been developed, using AI natural language technology: outputting the answer directly, outputting a computer program that solves the problem, and outputting a formalized representation that can be input to an automated theorem verifier. We review some benchmarks that have been developed to evaluate these systems and some experimental studies. We discuss the limitations of the existing technology at solving these kinds of problems. We argue that it is not clear whether these kinds of limitations will be important in developing AI technology for pure mathematical research, but that they will be important in applications of mathematics, and may well be important in developing programs capable of reading and understanding mathematical content written by humans.

Enabling High-Level Machine Reasoning with Cognitive Neuro-Symbolic Systems

High-level reasoning can be defined as the capability to generalize over knowledge acquired via experience, and to exhibit robust behavior in novel situations. Such form of reasoning is a basic skill in humans, who seamlessly use it in a broad spectrum of tasks, from language communication to decision making in complex situations. When it manifests itself in understanding and manipulating the everyday world of objects and their interactions, we talk about common sense or commonsense reasoning. State-of-the-art AI systems don’t possess such capability: for instance, Large Language Models have recently become popular by demonstrating remarkable fluency in conversing with humans, but they still make trivial mistakes when probed for commonsense competence; on a different level, performance degradation outside training data prevents self-driving vehicles to safely adapt to unseen scenarios, a serious and unsolved problem that limits the adoption of such technology. In this paper we propose to enable high-level reasoning in AI systems by integrating cognitive architectures with external neuro-symbolic components. We illustrate a hybrid framework centered on ACT-R, and we discuss the role of generative models in recent and future applications.

Building Intelligent Systems by Combining Machine Learning and Automated Commonsense Reasoning

We present an approach to building systems that emulate human-like intelligence. Our approach uses machine learning technology (including generative AI systems) to extract knowledge from pictures, text, etc., and represents it as (pre-defined) predicates. Next, we use the s(CASP) automated commonsense reasoning system to check the consistency of this extracted knowledge and reason over it in a manner very similar to how a human would do it. We have used our approach for building systems for visual question answering, task-specific chatbots that can ``understand" human dialogs and interactively talk to them, and autonomous driving systems that rely on commonsense reasoning. Essentially, our approach emulates how humans process knowledge where they use sensing and pattern recognition to gain knowledge (Kahneman's System 1 thinking, akin to using a machine learning model), and then use reasoning to draw conclusions, generate response, or take actions (Kahneman's System 2 thinking, akin to automated reasoning).

Information bottleneck based knowledge selection for commonsense reasoning

KG-augmented models usually endow existing models with external knowledge graphs, which achieve promising performance in various knowledge-intensive tasks, such as commonsense reasoning. Existing methods mainly first exploited heuristic ways for retrieving the relevant knowledge subgraphs according to the input, and then utilized some effective encoders, such as GNNs, to encode the symbolic knowledge into the neural reasoning networks. However, whether the whole retrieved knowledge subgraphs are really relevant or useful for the reasoning process was seldom considered. Actually, according to our observations and analysis, most retrieved knowledge is noisy and useless to the reasoning models, which would hurt the final performance. To remedy this, this paper proposes information bottleneck based knowledge selection (IBKS), which is able to select useful knowledge from the retrieved knowledge subgraph. Expectedly, the selected knowledge could better improve the commonsense reasoning ability of the model. Moreover, IBKS is model-agnostic and could be plugged into any existing KG-augmented model. Extensive experimental results show that IBKS could effectively improve commonsense reasoning performance.

ICORPP: Interleaved commonsense reasoning and probabilistic planning on robots

Robot sequential decision-making in the real world is a challenge because it requires the robots to simultaneously reason about the current world state and dynamics, while planning actions to accomplish complex tasks. On the one hand, declarative languages and reasoning algorithms support representing and reasoning with commonsense knowledge. But these algorithms are not good at planning actions toward maximizing cumulative reward over a long, unspecified horizon. On the other hand, probabilistic planning frameworks, such as Markov decision processes (MDPs) and partially observable MDPs (POMDPs), support planning to achieve long-term goals under uncertainty. But they are ill-equipped to represent or reason about knowledge that is not directly related to actions. In this article, we present an algorithm, called iCORPP, to simultaneously estimate the current world state, reason about world dynamics, and construct task-oriented controllers. In this process, robot decision-making problems are decomposed into two interdependent (smaller) subproblems that focus on reasoning to “understand the world” and planning to “achieve the goal” respectively. The developed algorithm has been implemented and evaluated both in simulation and on real robots using everyday service tasks, such as indoor navigation, and dialog management. Results show significant improvements in scalability, efficiency, and adaptiveness, compared to competitive baselines including handcrafted action policies.

Commonsense reasoning and automatic generation of IoT contextual knowledge: An Answer Set Programming approach

The increasing interest in searching for intelligent, proactive, and autonomous environments leads to the necessity of accessing contextual knowledge: knowledge about changes that occur in the environment and even about the capabilities of the devices that compose a specific deployment. This knowledge would allow carrying out commonsense reasoning (exhibit human-like comprehension) according to the current situation, enabling decision making and task planning. One of the most used forms of knowledge specification is ontologies, but ontological knowledge modeling is usually a manual process, making it a costly task. In addition, ontologies can be found that have been designed for specifying knowledge about IoT-related concepts, but many are overly complex or do not have an adequate orientation that allows modeling the system’s capabilities. There are many different types of reasoners, but Answer Set Solvers are of particular interest for commonsense reasoning. This paper proposes an ontology to represent contextual knowledge about IoT device capabilities, and an Answer Set Programming-based reasoner for the automatic generation of this knowledge. The main challenge is to demonstrate that contextual knowledge can be generated through commonsense reasoning processes implemented with Answer Set Programming, and that the resulting knowledge can be used for decision making (also through commonsense reasoning). This work shows how the generated knowledge is correct through different use cases, presents an application example that demonstrates the benefits of using the generated knowledge, and analyzes the reasoner performance to demonstrate that the execution time is adequate.

Cleavage politics in ordinary reasoning: How common sense divides

ABSTRACT This study explores how the ideological divide between the radical right and liberal left is anchored in the common sense reasoning of ordinary citizens. Across Western Europe, cleavage research has documented a divide in attitudes towards immigration and cultural liberalization which some view as a new cleavage of ‘universalism’ versus ‘particularism’. Yet it is unclear how this squares with the fact that most citizens are non-ideologues with only moderate political interest and knowledge. Building on Clifford Geertz, Pierre Bourdieu, and recent research into the politics of group identities, we theorize how even in the absence of fully worked-out political ideologies, citizens mobilize forms of common sense reasoning based on group distinctions and moral intuitions. We apply this approach in a rare qualitative comparison of political reasoning on both sides of the new divide, analyzing narrative interviews (N=64) with groups positioned on its opposing sides: workers or small owners voting for the radical right; and sociocultural professionals voting for the liberal left. We show how ‘common sense particularism’ and ‘common sense universalism’ draw on distinct intuitions regarding the worthy self, the scope of solidarity, and the normative status of community. These intuitions resonate with class-specific social experiences and group referents.

A Novel Model of Narrative Memory for Conscious Agents

The aim of Artificial General Intelligence (AGI) is to build machines with cognitive abilities on par with humans. These abilities include sensing, reasoning, decision-making, and social interaction. Although great strides have been achieved in creating intelligent agents that can observe and form opinions about their surroundings, problems still arise when it comes to their ability to reason and make decisions. The lack of a narrative structure inside these intelligent beings is a major drawback. This study suggests creating a narrative memory model with a dedicated module for use by self-aware agents. The model incorporates narrative memory to improve the agent's perception, understanding, and decision-making capabilities. The conscious agent will be able to create, store, and recall narrative-like representations of prior experiences with the help of the narrative memory module. To help the agent make sense of its surroundings, this narrative structure will give a method for categorizing and linking data. The agent will be able to recognize patterns, determine causal linkages, and extrapolate future outcomes by drawing on its narrative memory. There are several benefits to incorporating narrative memory into the brain of a self-aware agent. First, it'll give the agent a richer context from which to draw insights and make judgments. Second, the agent's perception and interpretation of complicated circumstances, as well as its general reasoning abilities, will be bolstered by the narrative memory module. Thirdly, the agent will be able to develop over time by absorbing new information and incorporating it into its preexisting narrative memory. This study uses a narrative memory-based approach to close the gap between conscious agents and human-like reasoning. The goal of this study is to improve the cognitive capacities of conscious agents by giving them the capacity to create and employ narrative memories, thus facilitating more nuanced perception, comprehension, and decision-making. The findings of this study could have far-reaching implications for the fields of artificial intelligence, robotics, and human-computer interaction by accelerating the creation of AGI systems with human-level cognitive capacities.

Learning to Agree on Vision Attention for Visual Commonsense Reasoning

Learning to Agree on Vision Attention for Visual Commonsense Reasoning