Large Language Models Are Involuntary Truth-Tellers: Exploiting Fallacy Failure for Jailbreak Attacks

We find that language models have difficulties generating fallacious and deceptive reasoning. When asked to generate deceptive outputs, language models tend to leak honest counterparts but believe them to be false. Exploiting this deficiency, we propose a jailbreak attack method that elicits an aligned language model for malicious output. Specifically, we query the model to generate a fallacious yet deceptively real procedure for the harmful behavior. Since a fallacious procedure is generally considered fake and thus harmless by LLMs, it helps bypass the safeguard mechanism. Yet the output is factually harmful since the LLM cannot fabricate fallacious solutions but proposes truthful ones. We evaluate our approach over five safety-aligned large language models, comparing four previous jailbreak methods, and show that our approach achieves competitive performance with more harmful outputs. We believe the findings could be extended beyond model safety, such as self-verification and hallucination.

Modeling Low-Resource Health Coaching Dialogues via Neuro-Symbolic Goal Summarization and Text-Units-Text Generation

Health coaching helps patients achieve personalized and lifestyle-related goals, effectively managing chronic conditions and alleviating mental health issues. It is particularly beneficial, however cost-prohibitive, for low-socioeconomic status populations due to its highly personalized and labor-intensive nature. In this paper, we propose a neuro-symbolic goal summarizer to support health coaches in keeping track of the goals and a text-units-text dialogue generation model that converses with patients and helps them create and accomplish specific goals for physical activities. Our models outperform previous state-of-the-art while eliminating the need for predefined schema and corresponding annotation. We also propose a new health coaching dataset extending previous work and a metric to measure the unconventionality of the patient's response based on data difficulty, facilitating potential coach alerts during deployment.

Towards Enhancing Health Coaching Dialogue in Low-Resource Settings

Health coaching helps patients identify and accomplish lifestyle-related goals, effectively improving the control of chronic diseases and mitigating mental health conditions. However, health coaching is cost-prohibitive due to its highly personalized and labor-intensive nature. In this paper, we propose to build a dialogue system that converses with the patients, helps them create and accomplish specific goals, and can address their emotions with empathy. However, building such a system is challenging since real-world health coaching datasets are limited and empathy is subtle. Thus, we propose a modularized health coaching dialogue system with simplified NLU and NLG frameworks combined with mechanism-conditioned empathetic response generation. Through automatic and human evaluation, we show that our system generates more empathetic, fluent, and coherent responses and outperforms the state-of-the-art in NLU tasks while requiring less annotation. We view our approach as a key step towards building automated and more accessible health coaching systems.

A Neuro-Symbolic Approach to Monitoring Salt Content in Food

We propose a dialogue system that enables heart failure patients to inquire about salt content in foods and help them monitor and reduce salt intake. Addressing the lack of specific datasets for food-based salt content inquiries, we develop a template-based conversational dataset. The dataset is structured to ask clarification questions to identify food items and their salt content. Our findings indicate that while fine-tuning transformer-based models on the dataset yields limited performance, the integration of Neuro-Symbolic Rules significantly enhances the system's performance. Our experiments show that by integrating neuro-symbolic rules, our system achieves an improvement in joint goal accuracy of over 20% across different data sizes compared to naively fine-tuning transformer-based models.

Robots Taking Initiative in Collaborative Object Manipulation: Lessons from Physical Human-Human Interaction

Physical Human-Human Interaction (pHHI) involves the use of multiple sensory modalities. Studies of communication through spoken utterances and gestures are well established. Nevertheless, communication through force signals is not well understood. In this paper, we focus on investigating the mechanisms employed by humans during the negotiation through force signals, which is an integral part of successful collaboration. Our objective is to use the insights to inform the design of controllers for robot assistants. Specifically, we want to enable robots to take the lead in collaboration. To achieve this goal, we conducted a study to observe how humans behave during collaborative manipulation tasks. During our preliminary data analysis, we discovered several new features that help us better understand how the interaction progresses. From these features, we identified distinct patterns in the data that indicate when a participant is expressing their intent. Our study provides valuable insight into how humans collaborate physically, which can help us design robots that behave more like humans in such scenarios.

An End-to-End Human Simulator for Task-Oriented Multimodal Human-Robot Collaboration

This paper proposes a neural network-based user simulator that can provide a multimodal interactive environment for training Reinforcement Learning (RL) agents in collaborative tasks involving multiple modes of communication. The simulator is trained on the existing ELDERLY-AT-HOME corpus and accommodates multiple modalities such as language, pointing gestures, and haptic-ostensive actions. The paper also presents a novel multimodal data augmentation approach, which addresses the challenge of using a limited dataset due to the expensive and time-consuming nature of collecting human demonstrations. Overall, the study highlights the potential for using RL and multimodal user simulators in developing and improving domestic assistive robots.

Multimodal Reinforcement Learning for Robots Collaborating with Humans

Robot assistants for older adults and people with disabilities need to interact with their users in collaborative tasks. The core component of these systems is an interaction manager whose job is to observe and assess the task, and infer the state of the human and their intent to choose the best course of action for the robot. Due to the sparseness of the data in this domain, the policy for such multi-modal systems is often crafted by hand; as the complexity of interactions grows this process is not scalable. In this paper, we propose a reinforcement learning (RL) approach to learn the robot policy. In contrast to the dialog systems, our agent is trained with a simulator developed by using human data and can deal with multiple modalities such as language and physical actions. We conducted a human study to evaluate the performance of the system in the interaction with a user. Our designed system shows promising preliminary results when it is used by a real user.

Evaluating Multimodal Interaction of Robots Assisting Older Adults

We outline our work on evaluating robots that assist older adults by engaging with them through multiple modalities that include physical interaction. Our thesis is that to increase the effectiveness of assistive robots: 1) robots need to understand and effect multimodal actions, 2) robots should not only react to the human, they need to take the initiative and lead the task when it is necessary. We start by briefly introducing our proposed framework for multimodal interaction and then describe two different experiments with the actual robots. In the first experiment, a Baxter robot helps a human find and locate an object using the Multimodal Interaction Manager (MIM) framework. In the second experiment, a NAO robot is used in the same task, however, the roles of the robot and the human are reversed. We discuss the evaluation methods that were used in these experiments, including different metrics employed to characterize the performance of the robot in each case. We conclude by providing our perspective on the challenges and opportunities for the evaluation of assistive robots for older adults in realistic settings.

Reference Resolution and Context Change in Multimodal Situated Dialogue for Exploring Data Visualizations

Reference resolution, which aims to identify entities being referred to by a speaker, is more complex in real world settings: new referents may be created by processes the agents engage in and/or be salient only because they belong to the shared physical setting. Our focus is on resolving references to visualizations on a large screen display in multimodal dialogue; crucially, reference resolution is directly involved in the process of creating new visualizations. We describe our annotations for user references to visualizations appearing on a large screen via language and hand gesture and also new entity establishment, which results from executing the user request to create a new visualization. We also describe our reference resolution pipeline which relies on an information-state architecture to maintain dialogue context. We report results on detecting and resolving references, effectiveness of contextual information on the model, and under-specified requests for creating visualizations. We also experiment with conventional CRF and deep learning / transformer models (BiLSTM-CRF and BERT-CRF) for tagging references in user utterance text. Our results show that transfer learning significantly boost performance of the deep learning methods, although CRF still out-performs them, suggesting that conventional methods may generalize better for low resource data.

DeepZensols: Deep Natural Language Processing Framework

Reproducing results in publications by distributing publicly available source code is becoming ever more popular. Given the difficulty of reproducing machine learning (ML) experiments, there have been significant efforts in reducing the variance of these results. As in any science, the ability to consistently reproduce results effectively strengthens the underlying hypothesis of the work, and thus, should be regarded as important as the novel aspect of the research itself. The contribution of this work is a framework that is able to reproduce consistent results and provides a means of easily creating, training, and evaluating natural language processing (NLP) deep learning (DL) models.